System

Abstract

To provide a system.SOLUTION: A system includes: means for collecting and accumulating flaming data of the past; means for collecting and accumulating data of effective words; means for analyzing using a natural language processing technique on the basis of the flaming data and the effective words; means for predicting economical effects of public information on the basis of an analysis result; and means for supporting at least either correction of creation of public information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] There is a risk that information posted by companies on social media or websites will cause outrage and damage to the company's image. There is also the issue that the impact of the information released is weak, resulting in a low economic effect. [Means for solving the problem]

[0005] We provide a system that uses data on past controversies and effective words to predict the economic impact of publishing content on social media or websites using AI. Furthermore, we can determine the amount of loss caused by using past controversy data when editing content, preventing damage to a company's image in advance. [Brief explanation of the drawings]

[0006] [Figure 1]1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 2 is a sequence diagram showing a flow of processing in the data processing system according to the first embodiment of the first form example. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1 of Embodiment 1. [Figure 13] FIG. 10 is a sequence diagram showing a processing flow of a data processing system in a second embodiment of the second form example. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 of Embodiment Example 2. [Figure 15] FIG. 10 is a sequence diagram showing the flow of processing in a data processing system according to a third embodiment of the third embodiment. [Figure 16] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 3 of Embodiment 3. [Figure 17] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in the first embodiment of the first form example when an emotion engine is combined. [Figure 18] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1 of Form Example 1 when an emotion engine is combined. [Figure 19] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in the second embodiment of the second form example when an emotion engine is combined. [Figure 20] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 of Form Example 2 when an emotion engine is combined. [Figure 21] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in the third embodiment of the third form example when an emotion engine is combined. [Figure 22] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 3 of Form Example 3 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0007] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0008] First, the terms used in the following description will be explained.

[0009] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, the processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), an APU (Accelerated Processing Unit), or a TPU (TENSOR PROCESSING UNIT (registered trademark)).

[0010] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0011] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0012] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0013] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0014] [First embodiment]

[0015] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0016] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0017] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0018] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0019] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0020] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0021] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0022] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0023] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0024] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0025] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0026] Next, the specific processing by the specific processing unit 290 of the data processing device 12 will be described.

[0027] "Example 1"

[0028] One embodiment of the present invention is an information disclosure support system for companies that utilizes AI. This system utilizes data on past flame wars and data on effective words, and predicts the economic impact of publishing information on social media or websites based on this data. Specifically, past flame war data is stored in a database and analyzed by AI. Data on effective words is also stored in the same way, and the published information is edited and created based on this data.

[0029] "Example 2"

[0030] Furthermore, another embodiment of the present invention has a function to grasp the amount of loss when using past flame data during correction. Specifically, the amount of loss in the event of a flame is estimated from past flame data and used as reference information for correction. This makes it possible to specifically quantify the risk of a flame and disclose information taking that risk into account.

[0031] "Example 3"

[0032] Furthermore, a further embodiment of the present invention has a function to prevent damage to a company's image in advance. Specifically, before information created or corrected by AI is made public, it evaluates whether the information poses a risk of causing a backlash and prompts corrections as necessary. This makes it possible to prevent damage to a company's image in advance.

[0033] The processing flow of each embodiment will be described below.

[0034] "Example 1"

[0035] Step 1: Accumulate data on past controversies and effective words in a database.

[0036] Step 2: AI analyzes this data and edits or creates public information.

[0037] Step 3: Based on the economic impact predicted by AI, the optimal timing and content of information disclosure are determined.

[0038] "Example 2"

[0039] Step 1: Estimate the amount of loss in the event of a fire based on past fire data.

[0040] Step 2: Use the estimated loss amount as a reference for corrections.

[0041] Step 3: Quantify the risk of a backlash and disclose information based on that risk.

[0042] "Example 3"

[0043] Step 1: Before information created or edited by AI is made public, assess whether it poses a risk of causing outrage.

[0044] Step 2: Based on the evaluation results, encourage correction of information as necessary.

[0045] Step 3: Publish the corrected information to prevent damage to the company's image.

[0046] Example 1

[0047] Next, a description will be given of Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0048] When companies release information on social media or their websites, they face the challenge of being unable to predict the risk of a public outcry and the economic impact. In particular, there is a need for methods to utilize past examples of public outcry and effective keywords to reduce the risk of public information and maximize the economic impact. Another important issue is preventing damage to a company's image in advance.

[0049] The identification process by the identification processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means. In this invention, the server includes a means for collecting and storing past flame data, a means for collecting and storing data on effective words, a means for analyzing this data using natural language processing technology, a means for predicting the economic effect of public information based on the analysis results, and a means for supporting the editing and creation of public information. This allows companies to reduce the risk of public information becoming a flame and maximize the economic effect. It also makes it possible to prevent damage to the company's image in advance.

[0050] "Past flame war data" refers to information about past flame wars that have occurred on social media or websites, and specifically includes the content of posts, comments, number of retweets, and user reactions.

[0051] "Effective word data" refers to information about keywords and phrases that elicit positive responses on social media and websites, and specifically includes keywords and phrases that have generated positive responses from users.

[0052] "Natural language processing technology" refers to technology that enables computers to understand, analyze, and generate human language, and specifically includes sentiment analysis and keyword extraction.

[0053] "Economic impact forecasting" refers to assessing how much traffic the public information will attract on social media or websites, or how much of a risk there is of it causing a backlash.

[0054] "Support for editing and creating public information" refers to improving the quality of information created by users by reducing the risk of online outrage and suggesting effective words.

[0055] "Preventing damage to a company's image" refers to measures taken to prevent a company's reputation from being damaged due to information that the company makes public.

[0056] MODE FOR CARRYING OUT THE INVENTION

[0057] This invention is an information disclosure support system that reduces the risk of flame wars and maximizes the economic impact when companies disclose information on social media or their websites. This system collects and stores data on past flame wars and effective words, and performs analysis based on this data using natural language processing technology. Furthermore, it predicts the economic impact of the disclosed information based on the analysis results and supports the editing and creation of the disclosed information.

[0058] Data collection and storage

[0059] The server collects past flame war data from social media sites and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags, and obtains data such as the number of comments and retweets related to the flame war. It also uses marketing research tools to collect data on effective keywords. For example, it uses Google® Analytics or SEMrush to analyze how much traffic specific keywords attracted. The collected data is stored in a database such as MySQL® or PostgreSQL®.

[0060] Analyzing the data

[0061] The server analyzes the accumulated flame data using Tensorflow (registered trademark) and PyTorch (registered trademark). Specifically, it uses natural language processing (NLP) techniques to analyze the sentiment of posts and identify the factors that cause flame wars. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive responses. For example, it uses models such as Word2Vec and BERT to evaluate the relevance of keywords.

[0062] Economic impact forecast

[0063] Based on the analysis results, the server predicts the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. This prediction is made by combining past data with AI models.

[0064] Editing and creating public information

[0065] The user creates public information and sends it from their device to the server. For example, they may enter a company's new product introduction or campaign announcement. The server assesses the risk of a controversy based on the information sent. Specifically, it analyzes the entered text using NLP technology and compares it with past controversy data to assess the risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a controversy, so please change it to this expression." The user then amends the information based on the server's suggestions and creates the final public information. The revised information can then be sent back to the server for final confirmation.

[0066] Specific examples

[0067] Example 1: Flaming risk assessment

[0068] A user creates a description of a new product and sends it from their device to the server. The server evaluates the risk of this description causing a backlash based on past data on backlash. For example, if the expression "This product is superior to other companies' products" is judged to increase the risk of backlash, the server will suggest changing the expression to something like "This product is well-received by many users."

[0069] Example 2: Effective word suggestions

[0070] The user creates a campaign announcement and sends it from their device to the server. The server then uses a database of effective words to suggest keywords that will elicit a positive response to the announcement. For example, if keywords such as "limited," "free," and "special offer" are deemed effective, the server will suggest adding these keywords to the announcement.

[0071] Prompt Sentence Examples

[0072] "We have created an introduction for a new product. Please assess the risk of this introduction causing outrage if it is published on social media, and make any necessary revisions."

[0073] "We've created a campaign announcement. Can you suggest some effective words to elicit a positive response to this announcement?"

[0074] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0075] Step 1: Collect and store data

[0076] The server collects past flame war data from social media sites and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags and obtains data such as comments and retweets related to the flame war. The input is the specific keywords or hashtags, and the output is the collected flame war data. The server stores the collected data in a database such as MySQL (registered trademark) or PostgreSQL (registered trademark).

[0077] Step 2: Effective Word Data Collection

[0078] The server uses marketing research tools to collect data on effective keywords. For example, it uses Google Analytics or SEMrush to analyze how much traffic a specific keyword attracted. The input is the specific keyword, and the output is data on effective keywords. The server stores the collected data in a database.

[0079] Step 3: Analyze the data

[0080] The server uses TensorFlow (registered trademark) and PyTorch (registered trademark) to analyze the accumulated flame data. Specifically, it uses natural language processing (NLP) technology to analyze the sentiment of the posts and identify the factors that cause the flames. The input is the accumulated flame data, and the output is the results of identifying the factors that cause the flames. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive reactions. The input is data on effective words, and the output is keywords and phrases that elicit positive reactions.

[0081] Step 4: Forecasting economic impact

[0082] The server uses the analysis results to predict the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. The input is the analysis results, and the output is the predicted economic impact.

[0083] Step 5: Create and submit your publication

[0084] A user creates public information and sends it to the server from their terminal. For example, they input a company's new product introduction or campaign announcement. The input is the public information created by the user, and the output is the public information sent to the server.

[0085] Step 6: Evaluate public information and make recommendations

[0086] The server evaluates the risk of a flame war based on the information sent. Specifically, it analyzes the input text using NLP technology and compares it with past flame war data to evaluate the risk. The input is public information created by the user, and the output is the evaluation result of the flame war risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a flame war, so please change it to this expression." The input is the evaluation result of the flame war risk, and the output is the suggested correction.

[0087] Step 7: Revise and finalize the public information

[0088] The user modifies the information based on the server's suggestions and creates the final published information. The modified information can be sent back to the server for final confirmation. The input is the server's suggested modifications, and the output is the modified published information.

[0089] (Application example 1)

[0090] Next, a description will be given of Application Example 1 of Embodiment Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0091] When companies publish advertisements on social media or their websites, they are required to predict the economic impact by utilizing data on past controversies and effective words. However, there is no system that effectively utilizes this data to support the editing and creation of advertisement copy. Furthermore, there is a lack of methods to prevent damage to a company's image in advance by predicting controversy risk scores and effective word scores. This creates a problem in that companies are forced to operate while taking on risks when publishing advertisements.

[0092] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0093] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, a means including AI for predicting the economic effect of publishing on social media or a website based on this data, a means for supporting the correction and creation of advertising copy, and a means for predicting a flame risk score and an effective word score, thereby enabling companies to reduce the risk of publishing advertisements and create effective advertising copy.

[0094] "Past flame war data" refers to information about past flame war cases that have occurred on social media or websites.

[0095] "Effective word data" refers to information about words and phrases that are highly effective in advertising and information disclosure.

[0096] "AI that predicts economic impact" refers to artificial intelligence that predicts the economic impact of information published on social media or websites based on past data.

[0097] "Means to support the editing and creation of advertising copy" refers to a function that provides support for companies to revise advertising copy created by them based on past data or to create new copy.

[0098] The "flame risk score" is an index that quantifies the likelihood that a particular advertisement copy or piece of information will cause a flame war on social media or a website.

[0099] "Effective word score" refers to an indicator that quantifies the likelihood that a particular ad copy or piece of information will be highly effective.

[0100] "Measures to prevent damage to a company's image in advance" refers to a function that prevents a company's image from being damaged by information that the company makes public.

[0101] The system for implementing this invention includes AI that utilizes past data on flaming and effective words to predict the economic impact of advertising copy to be published on social media and websites. It also has the function of supporting the correction and creation of advertising copy, and predicting the flaming risk score and effective word score.

[0102] System configuration

[0103] Hardware:

[0104] server

[0105] Smartphone

[0106] software:

[0107] Python (registered trademark)

[0108] pandas

[0109] scikit-learn (registered trademark)

[0110] joblib

[0111] Data Handling

[0112] The server stores past flame data and effective word data in a database. These data are saved as CSV files and loaded using pandas. The loaded data is converted into a numeric vector using TfidfVectorizer.

[0113] Training and saving the model

[0114] The server trains a Logistic Regression model based on the data converted to a numeric vector using TfidfVectorizer. The trained model is saved using joblib. This lays the foundation for predicting flame risk scores and effective word scores.

[0115] Ad copy rating

[0116] When a user enters an ad copy using their smartphone, the server uses the stored model to predict the ad copy's flame war risk score and effective word score, allowing users to understand the risk and effectiveness of the ad copy in advance.

[0117] Specific examples

[0118] For example, if a user enters the following ad text:

[0119] To celebrate the launch of our new products, we're holding a special sale!

[0120] The server predicts the flame risk score and effective word score for this ad copy. The predicted results are a flame risk score of 0.2 and an effective word score of 0.8.

[0121] Prompt Sentence Examples

[0122] Predict the flame risk score and effective word score for the following ad copy.

[0123] Ad copy: "To celebrate the launch of our new product, we're offering a special sale!"

[0124] In this way, companies can reduce the risk of advertising and create effective advertising copy.

[0125] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0126] Step 1:

[0127] The server loads past flame data and effective word data from the database. These data are stored as CSV files and are loaded using pandas. The input is a CSV file and the output is a data frame. Specifically, the data is loaded using the pandas.read_csv function.

[0128] Step 2:

[0129] The server converts the read data into a numeric vector using TfidfVectorizer. The input is a data frame, and the output is a numeric vector. Specifically, the server converts text data into a numeric vector using the fit_transform method of TfidfVectorizer.

[0130] Step 3:

[0131] The server trains a LogisticRegression model based on the converted data. The input is a numeric vector, and the output is the trained model. Specifically, the model is trained using the fit method of LogisticRegression.

[0132] Step 4:

[0133] The server saves the trained model and TfidfVectorizer using joblib. The input is the trained model and TfidfVectorizer, and the output is the saved model file. Specifically, the joblib.dump function is used to save the model and vectorizer.

[0134] Step 5:

[0135] The user inputs the ad copy using a smartphone. The input is the text of the ad copy, and the output is a request to the server. Specifically, the ad copy is input through a smartphone application and sent to the server.

[0136] Step 6:

[0137] The server converts the received ad copy into a numeric vector using TfidfVectorizer. The input is the ad copy text, and the output is a numeric vector. Specifically, the server converts the ad copy into a numeric vector using the transform method of the saved TfidfVectorizer.

[0138] Step 7:

[0139] The server predicts the flame risk score and effective word score based on the ad copy converted to a numeric vector. The input is a numeric vector, and the output is the flame risk score and effective word score. Specifically, the server predicts the scores using the predict_proba method of the saved LogisticRegression model.

[0140] Step 8:

[0141] The server returns the predicted flame risk score and effective word score to the user. The input is the predicted score, and the output is a response to the user. Specifically, the score is displayed through a smartphone application.

[0142] Example 2

[0143] Next, a description will be given of Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0144] In today's information society, when companies or individuals release information on social media or their websites, they are required to use past data on online outrage to predict risks and take appropriate measures. However, existing systems lack the means to effectively utilize past data on online outrage and to specifically estimate the amount of loss, which makes it difficult to adequately manage risk. Furthermore, they lack concrete measures to prevent damage to a company's image in advance.

[0145] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0146] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, an artificial intelligence system for predicting the economic impact of information disclosure based on this data, a means for analyzing past flame data and estimating loss amounts, a means for storing the estimated loss amounts in a database, and a means for providing risk information including the estimated loss amounts when a user discloses information. This makes it possible to estimate specific loss amounts based on past flame data, quantify risk, and take appropriate measures when disclosing information. It also provides specific means for preventing damage to a company's image in advance.

[0147] "Past flame war data" refers to data on past flame war cases that occurred on online platforms such as social media and websites.

[0148] "Effective word data" refers to data on keywords and phrases that are effective when used when publishing information on social media or websites.

[0149] The "economic impact of information disclosure" refers to the predicted economic impact or benefits when information is disclosed on social media or a website.

[0150] "Artificial intelligence" is a system that uses machine learning and data analysis techniques to learn patterns from past data and make future predictions and decisions.

[0151] "Amount of loss" refers to the estimated amount of economic loss based on past cases of online outrage.

[0152] A "database" is a system for efficiently storing, managing, and retrieving data.

[0153] "Risk information" refers to information about risks that should be taken into consideration when disclosing information, and specifically includes estimated losses.

[0154] "Damage to corporate image" refers to a decline in a company's reputation and brand value.

[0155] This invention is a system for utilizing past flame war data to quantify the risk of information disclosure and take appropriate measures. Specific embodiments of this system are described below.

[0156] Server Roles

[0157] The server retrieves past online flame data from a database, analyzes it, and estimates the amount of loss. Specifically, the server uses the following hardware and software:

[0158] Hardware: A server machine with a powerful processor and sufficient memory

[0159] Software: Relational database management systems (RDBMS) such as MySQL (registered trademark) and PostgreSQL (registered trademark), data analysis libraries such as Python (registered trademark) pandas and scikit-learn (registered trademark)

[0160] The server first executes an SQL query to retrieve past flame data from the database. Next, it converts the retrieved data into a data frame using the Python (registered trademark) pandas library and performs preprocessing. Using the preprocessed data, it estimates the loss amount using the Linear Regression model of scikit-learn (registered trademark). The estimated loss amount is then saved back to the database.

[0161] Device Role

[0162] When a user discloses information, the terminal acquires risk information from the server and displays it to the user. Specifically, the terminal uses the following hardware and software:

[0163] Hardware: Personal computers, tablets, smartphones, etc.

[0164] Software: Web browser, HTML, CSS, JavaScript (registered trademark)

[0165] The device sends an HTTP GET request to the server and receives risk information in JSON format, which is then displayed on a web page using HTML and JavaScript.

[0166] User Roles

[0167] Users can check the risk information through their devices and adjust the content of the information they disclose. Specifically, users check the estimated loss amount displayed on the web page, and if the risk is high, they can take measures such as changing the content of the information they disclose.

[0168] Specific examples

[0169] Example 1: Acquiring and analyzing flame data

[0170] The server retrieves past social media flame data from a MySQL® database, using, for example, the following SQL query:

[0171] sql

[0172] SELECT FROM flame_data WHERE date >= '2020-01-01';

[0173] The acquired data is analyzed using the Python (registered trademark) pandas library to estimate the loss amount in the event of a fire. For example, the following code is used:

[0174] Python (registered trademark)

[0175] import pandas as pd

[0176] from sklearn.linear_model import LinearRegression

[0177] Loading data

[0178] data = pd.read_sql('SELECT FROM flame_data WHERE date >= "2020-01-01"', con=database_connection)

[0179] Estimated loss amount

[0180] model = LinearRegression()

[0181] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[0182] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[0183] Example 2: Displaying risk information

[0184] When a user makes a disclosure, the device retrieves the risk information from the server and displays it in an HTML page, for example using code like this:

[0185] html

[0186] <!DOCTYPE html>

[0187]

[0188]

[0189] <title> Risk Information< / title>

[0190]

[0191]

[0192] <h1> Risk Information< / h1>

[0193] Estimated loss amount: \

[0194] <script>

[0195] fetch('https: / / example.com / api / risk_info')

[0196] .then(response => response.json())

[0197] .then(data => {

[0198] document.getElementById('loss_amount').textContent = data.loss_amount;

[0199] });

[0200] < / script>

[0201]

[0202]

[0203] Prompt Sentence Examples

[0204] Examples of prompts to input to a generative AI model might include:

[0205] "Generate Python (registered trademark) code that uses past social media flame war data to estimate the amount of losses incurred during a flame war."

[0206] Using this prompt sentence, the generative AI model can generate the appropriate Python code.

[0207] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0208] Step 1:

[0209] The server retrieves past flame data from the database. It receives database connection information and an SQL query as input, and generates a data frame containing the flame data as output. Specifically, the server executes the SQL query to extract the past flame data. For example, the following SQL query is used:

[0210] sql

[0211] SELECT FROM flame_data WHERE date >= '2020-01-01';

[0212] This query retrieves flame data for the specified time period.

[0213] Step 2:

[0214] The server analyzes the acquired data. It receives a data frame as input and generates a preprocessed data frame as output. Specifically, the server completes missing values ​​in the data and removes unnecessary columns. For example, it preprocesses the data using the Python (registered trademark) pandas library.

[0215] Step 3:

[0216] The server uses the preprocessed data to estimate the loss amount. It receives the preprocessed data frame as input and generates the estimated loss amount as output. Specifically, the server uses a LinearRegression model from scikit-learn (registered trademark) to estimate the loss amount. For example, the model is trained as follows to predict the loss amount:

[0217] Python (registered trademark)

[0218] model = LinearRegression()

[0219] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[0220] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[0221] This process allows the amount of loss to be estimated.

[0222] Step 4:

[0223] The server stores the estimated loss amount in a database. It receives the estimated loss amount and database connection information as input, and generates the loss amount stored in the database as output. Specifically, the server executes an SQL query to insert the estimated result into the database. For example, it uses the following SQL query:

[0224] sql

[0225] INSERT INTO risk_info (date, predicted_loss) VALUES ('2023-10-01', 500000);

[0226] This query results in the estimated loss amount being stored in a database.

[0227] Step 5:

[0228] When a user discloses information, the device obtains risk information from the server. It receives an HTTP GET request as input and generates JSON data containing the risk information as output. Specifically, the device sends an HTTP GET request to the server and receives the risk information. For example, the following JavaScript code is used:

[0229] javascript

[0230] fetch('https: / / example.com / api / risk_info')

[0231] .then(response => response.json())

[0232] .then(data => {

[0233] document.getElementById('loss_amount').textContent = data.predicted_loss;

[0234] });

[0235] Through this process, risk information is obtained.

[0236] Step 6:

[0237] The terminal displays the acquired risk information to the user. It receives JSON data containing risk information as input and generates an HTML page displaying the risk information as output. Specifically, the terminal uses HTML and JavaScript to display the risk information on a web page. For example, it can be implemented as follows:

[0238] html

[0239] <!DOCTYPE html>

[0240]

[0241]

[0242] <title> Risk Information< / title>

[0243]

[0244]

[0245] <h1> Risk Information< / h1>

[0246] Estimated loss amount: \

[0247] <script>

[0248] fetch('https: / / example.com / api / risk_info')

[0249] .then(response => response.json())

[0250] .then(data => {

[0251] document.getElementById('loss_amount').textContent = data.predicted_loss;

[0252] });

[0253] < / script>

[0254]

[0255]

[0256] This process displays risk information to the user.

[0257] Step 7:

[0258] The user checks the risk information through the terminal. As input, the user receives the displayed risk information, and as output, adjusts the content of the information disclosure. Specifically, the user checks the estimated loss amount displayed on the web page, and if the risk is high, takes action such as changing the disclosure content.

[0259] (Application example 2)

[0260] Next, a description will be given of Application Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0261] There is a need for a system that can quantify the risk of a controversy and predict the amount of loss in advertising campaigns using past controversy data. However, current systems have the problem of being unable to specifically quantify the risk of a controversy, disclose information based on that risk, or provide specific advice for reducing the risk.

[0262] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0263] In this invention, the server includes means for utilizing past flame data, means for utilizing data on effective words, means including a generative AI model for predicting the economic impact of publishing on social media or a website based on this data, means for quantifying the risk of a flame based on past flame data when the content of an advertising campaign is input, means for estimating the amount of loss based on the flame risk, and means for providing specific advice for reducing the risk. This makes it possible to specifically quantify the risk of a flame in an advertising campaign and provide information disclosure based on that risk and specific advice for reducing the risk.

[0264] "Past flame war data" refers to data such as negative reactions, comments, and criticisms that occurred on social media or websites in the past.

[0265] "Effective word data" refers to data on keywords and phrases that are considered effective in eliciting positive responses on social media and websites.

[0266] A "generative AI model" refers to an algorithm or system that uses machine learning or artificial intelligence techniques to generate a specific output from input data.

[0267] "Means for quantifying the risk of a backlash" refers to a method or system that uses data on past backlashes to numerically express the possibility that a particular advertising campaign will cause a backlash.

[0268] "Means for estimating loss amount" refers to a method or system for calculating expected economic loss based on the risk of a fire.

[0269] "Means for providing specific advice" refers to methods or systems that suggest specific actions or measures to reduce the risk of online outrage.

[0270] The system for carrying out the present invention operates in cooperation with three entities: a server, a terminal, and a user. A specific embodiment of the system will be described below.

[0271] Server Roles

[0272] The server stores data on past flame wars and effective keywords, and uses a generative AI model based on this data to predict economic impact. The server uses the following software and hardware:

[0273] Software: Python (registered trademark), Pandas, Scikit-learn (registered trademark)

[0274] Hardware: A server with a powerful processor and large memory capacity

[0275] The server loads past flame data and converts it into a data frame. Next, it sets features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model. This allows it to quantify the flame risk and predict loss amounts by inputting the details of an advertising campaign.

[0276] Device Role

[0277] The terminal provides an interface for users to input the content of their advertising campaigns. The terminal uses the following software and hardware:

[0278] Software: Web browser, mobile application

[0279] Hardware: Smartphones, tablets, computers

[0280] When a user enters details of an ad campaign through their device, the data is sent to a server, which then calculates the risk of a potential online outrage and predicts the amount of damage. The prediction is then sent back to the device and displayed to the user.

[0281] User Roles

[0282] Users can input details of their advertising campaigns using the terminal, check the predicted results of the risk of outrage and the amount of loss, and receive specific advice from the server on how to reduce the risk and take appropriate measures.

[0283] Specific examples

[0284] For example, if a user inputs a 30-day ad campaign length, 100,000 people as the audience size, and 50 negative comments, the server will use this data to quantify the risk of outrage and calculate the expected loss. As a result, the estimated loss amount is displayed and specific advice on how to reduce the risk is provided.

[0285] Prompt Sentence Examples

[0286] "If your ad campaign is 30 days long, your audience size is 100,000, and you have 50 negative comments, what is your estimated loss?"

[0287] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[0288] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0289] Step 1:

[0290] A user uses a terminal to enter details of an advertising campaign.

[0291] Inputs: length of ad campaign, audience size, number of negative comments

[0292] Output: The input data is sent to the server

[0293] What it does: Users enter details about their advertising campaign through a smartphone or computer interface and click the "Submit" button.

[0294] Step 2:

[0295] Based on the data received by the server, past flame data is read.

[0296] Input: Details of the ad campaign submitted by the user

[0297] Output: Historical fire data converted into a data frame

[0298] Specific operation: The server uses the Python (registered trademark) Pandas library to read past fire data from a CSV file and convert it into a data frame.

[0299] Step 3:

[0300] The server sets the features and target and trains a linear regression model.

[0301] Input: Historical fire data converted into a data frame

[0302] Output: A trained linear regression model

[0303] How it works: The server uses the Scikit-learn (registered trademark) library to set features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model.

[0304] Step 4:

[0305] Based on the detailed data of the advertising campaign received by the server from the user, the risk of a backlash is quantified and the amount of loss is predicted.

[0306] Input: Detailed advertising campaign data received from users, a trained linear regression model

[0307] Output: Fire risk figure and predicted loss amount

[0308] How it works: The server inputs the received detailed data of the advertising campaign as features, and uses a trained linear regression model to quantify the risk of a backlash and predict the amount of loss.

[0309] Step 5:

[0310] The server returns the predicted results of the risk of a firestorm and the amount of loss to the terminal.

[0311] Input: Flame risk value and predicted loss amount

[0312] Output: The prediction results are displayed on the terminal.

[0313] Specific operation: The server sends the prediction results in JSON format to the device, which receives them and displays them to the user.

[0314] Step 6:

[0315] The server generates specific advice to reduce the risk and sends it to the terminal.

[0316] Input: Flame risk value and predicted loss amount

[0317] Output: Specific advice for reducing risk

[0318] Specific operation: The server uses the generative AI model to generate specific advice for reducing the risk of flame wars and sends it to the device, which receives it and displays it to the user.

[0319] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[0320] Example 3

[0321] Next, a description will be given of a third embodiment of the third embodiment. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0322] In today's information society, there is an increasing risk of a company becoming involved in a scandal when publishing information on social media or on its website. This risk can damage a company's image and cause financial losses, so companies are required to assess the risk of a scandal in advance and revise information as necessary. However, the current system does not adequately assess the risk of a scandal or revise information, which is a major issue for companies.

[0323] The specific processing by the specific processing unit 290 of the data processing device 12 in the third embodiment is realized by the following means.

[0324] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means for using a generative AI model that generates information based on these data, means for assessing the flame war risk before the generated information is made public, means for encouraging the correction of information if there is a flame war risk, and means for predicting the economic effect of publishing the information on social media or a website based on these means. This enables companies to prevent damage to their corporate image by assessing the flame war risk before publishing information and correcting the information as necessary.

[0325] "Past flame war data" refers to data on past flame wars that occurred as a result of information that was made public on social media, websites, etc.

[0326] "Effective word data" refers to data on words and phrases that are considered effective in eliciting positive responses when publishing information on social media or websites.

[0327] A "generative AI model" is an artificial intelligence model that generates information based on prompts entered by the user.

[0328] The "means for assessing the risk of a controversy" is a means for analyzing and assessing the possibility that the generated information will cause a controversy when it is made public.

[0329] "Means to encourage correction of information" refers to means to encourage users to correct information that is assessed as having a risk of causing a flame war.

[0330] "Means for predicting economic effects" refers to a means for predicting the economic impact that information published on social media or websites will have on a company.

[0331] MODE FOR CARRYING OUT THE INVENTION

[0332] This invention is a system that prevents damage to a company's image by assessing the risk of a controversy beforehand when a company publishes information on social media or a website, and correcting the information as necessary. This system consists of three main elements: a server, a terminal, and a user.

[0333] 1. Program Generation

[0334] The server generates a program to prevent damage to a company's image in advance. This program uses a generative AI model to create or edit information and evaluate the risk of that information causing a backlash before it is made public. Specifically, OpenAI's GPT-4 (registered trademark) is used as the generative AI model.

[0335] 2. Program Processing

[0336] The server uses a generative AI model (e.g., OpenAI's GPT-4®) to generate information based on the prompt entered by the user. The generated information is then sent to an inflammability risk assessment module. This module uses natural language processing (NLP) techniques to analyze the content of the information and assess whether it poses an inflammability risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information poses an inflammability risk.

[0337] The terminal sends the prompt sentence entered by the user to the server, and receives the generated information and the risk assessment result returned from the server. The user checks the risk assessment result on the terminal and corrects the information as necessary.

[0338] 3. Examples of concrete examples and prompts

[0339] As a concrete example, consider a user creating a press release for a new product. The user types the following prompt into the terminal:

[0340] Example prompt sentence:

[0341] "Write a press release for your new product. Call it 'EcoSmart' and highlight its environmentally friendly features."

[0342] The server inputs this prompt into a generative AI model and generates the following information:

[0343] Examples of generated information:

[0344] "EcoSmart is a new product that utilizes the latest environmental technology. It is energy efficient and made from recyclable materials."

[0345] The server then sends the generated information to the flame risk assessment module, which performs a risk assessment. If the assessment results in a "risk," the server sends a message to the user urging them to make corrections. The user can then check the suggested corrections on their device and make any necessary corrections to prevent damage to the company's image.

[0346] This system enables companies to assess the risk of information being exposed to controversy before it is released and correct the information as necessary, thereby preventing damage to the company's image in advance. The flow of the identification process in the third embodiment will be described with reference to FIG.

[0347] Step 1:

[0348] The user enters a prompt statement.

[0349] The user inputs a prompt into the input field of the terminal, for example, "Please write a press release for a new product. The product name should be 'EcoSmart' and the product's environmentally friendly features should be emphasized." The input prompt is temporarily stored in the terminal's memory.

[0350] Step 2:

[0351] The terminal sends the prompt to the server.

[0352] The terminal converts the prompt text entered by the user into an appropriate format and sends a request to the server's API endpoint. Specifically, it sends the prompt text to the server using an HTTP POST request. The input is the prompt text, and the output is the request sent to the server.

[0353] Step 3:

[0354] The server generates information using a generative AI model.

[0355] The server inputs the received prompt into a generative AI model (e.g., OpenAI's GPT-4 (registered trademark)) to generate information. The generative AI model analyzes the prompt and generates information based on the appropriate context. The input is the prompt, and the output is the generated information. For example, the information generated is, "EcoSmart is a new product that utilizes the latest environmental technology. It is highly energy efficient and uses recyclable materials."

[0356] Step 4:

[0357] The server sends the generated information to a flame risk assessment module.

[0358] The server sends the generated information to the flame risk assessment module for risk assessment. This module uses natural language processing (NLP) technology to analyze the content of the information and assess whether it poses a flame risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information contains a flame risk. The input is the generated information, and the output is the risk assessment result.

[0359] Step 5:

[0360] The server generates the risk assessment results and sends them to the terminal.

[0361] The server receives the evaluation results from the flame risk assessment module and sends them to the terminal. The evaluation results include judgments such as "no risk" or "risk present." The input is the risk assessment result, and the output is the transmission to the terminal.

[0362] Step 6:

[0363] The terminal displays the risk assessment results to the user.

[0364] The terminal displays the risk assessment results received from the server to the user. For example, a message such as "This information has the risk of causing a firestorm. We recommend that you modify it" is displayed. The input is the risk assessment result, and the output is what is displayed to the user.

[0365] Step 7:

[0366] The user modifies the information as needed.

[0367] The user checks the risk assessment results displayed on the device and modifies the information as necessary. The modified information is sent back to the server, where it undergoes the same process before finally being published as safe information. The input is the modified information, and the output is the re-evaluated information.

[0368] (Application example 3)

[0369] Next, a description will be given of Application Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0370] When companies release information on social media or their websites, they are required to assess the risk of a backlash in advance to prevent damage to their corporate image. However, conventional methods do not fully utilize data on past backlashes or effective keywords, and are insufficient in predicting economic effects. In addition, they lack the functionality to suggest specific corrections, making it difficult for companies to respond quickly and appropriately. This often results in companies taking unnecessary risks.

[0371] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 3 is realized by the following means.

[0372] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effects of disclosing information based on these data, means for assessing the flame war risk of information using a generative AI model, and means for presenting specific correction proposals if there is a flame war risk. This enables companies to assess the flame war risk before disclosing information and obtain specific correction proposals as necessary.

[0373] "Past flame war data" refers to information about past flame war cases that occurred on social media, websites, etc., and specifically includes data on the causes and effects of the flame wars, as well as related keywords.

[0374] "Data on effective words" refers to words and phrases used on social media, websites, etc. that are said to be particularly effective in eliciting positive responses.

[0375] "Artificial intelligence that predicts the economic impact of information disclosure" is an artificial intelligence technology used to predict the impact that disclosed information will have on a company's economic activity, and makes predictions based on past data and trends.

[0376] A "generative AI model" is an artificial intelligence model that generates text using natural language processing techniques and has the ability to generate sentences based on specific prompts.

[0377] "Means for assessing the risk of online outrage" refers to a means for assessing the possibility that publicly disclosed information will cause online outrage, and determines the risk using past online outrage data and generative AI models.

[0378] "Means for presenting specific revision proposals" refers to means for presenting specific revision proposals to reduce the risk of information being assessed as having a risk of causing outrage.

[0379] The system for carrying out the present invention operates in cooperation with three entities: a server, a terminal, and a user. A specific embodiment of the system will be described below.

[0380] System Configuration

[0381] The server includes means for utilizing data on past flame wars, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effects of disclosing information, means for assessing the risk of information becoming a flame war using a generative AI model, and means for presenting specific suggestions for correction if there is a risk of becoming a flame war.

[0382] A terminal is a device through which a user inputs information and communicates with a server. Specifically, this applies to smartphones and personal computers.

[0383] The users are corporate public relations personnel, marketing personnel, etc., and are responsible for inputting information and receiving feedback from the server.

[0384] Program processing

[0385] The server first receives information entered by the user, such as advertising copy and campaign details to be published on social media or on the website.

[0386] The server then evaluates the information's risk of causing an uproar based on past data on uproars and effective keywords. Specifically, it uses a generative AI model to determine whether the input information poses a risk of causing an uproar.

[0387] If the evaluation determines that there is a risk of outrage, the server generates specific revision suggestions, which use a generative AI model to suggest specific wording or phrases to reduce the risk.

[0388] Finally, the server sends the evaluation results and suggested modifications to the user's device, where the user can receive them and modify the information as needed.

[0389] Hardware and software used

[0390] Hardware: Servers, smartphones, PCs

[0391] Software: OpenAI API, Python (registered trademark)

[0392] Specific examples

[0393] For example, suppose a user enters the following ad copy:

[0394] This product is better than the competition. Buy it now!

[0395] The server receives this copy and generates the following prompt:

[0396] Evaluate the ad copy below and determine if it poses a risk of causing controversy.

[0397] Advertisement: This product is better than the competition. Buy it now!

[0398] If there is a risk of outrage, please provide specific suggestions for correction.

[0399] The generative AI model evaluates the prompt and suggests the following corrections:

[0400] This ad copy contains language that disparages other companies' products, and there is a risk of it causing outrage. We recommend that you revise it as follows:

[0401] Suggested fix: "This product has many satisfied customers. Buy it now and experience the difference!"

[0402] In this way, users can receive feedback from the server and correct their information, reducing the risk of backlash and preventing damage to the company's image.

[0403] The flow of the specific processing in Application Example 3 will be described with reference to FIG.

[0404] Step 1:

[0405] The user uses the terminal to input information such as advertising copy and campaign details.

[0406] Input: Ad copy and campaign details

[0407] Output: The entered information is sent to the server

[0408] How it works: The user enters the ad copy into the input form on their smartphone or computer and presses the submit button. The device then sends this information to the server.

[0409] Step 2:

[0410] The server compares the information it receives with past flame data and effective word data.

[0411] Input: User-submitted information, past flame data, effective word data

[0412] Output: Matching results (data necessary for assessing the risk of a firestorm)

[0413] Specific operation: The server retrieves past flame data and effective word data from the database and compares it with the information received from the user.

[0414] Step 3:

[0415] The server uses a generative AI model to assess the risk of information becoming viral.

[0416] Input: Matching results, generative AI model

[0417] Output: Flame risk assessment results

[0418] Specific operation: The server inputs the matching results into the generative AI model, generates a prompt sentence, and performs an evaluation. The generative AI model determines the risk of flame wars and returns the result to the server.

[0419] Step 4:

[0420] If the server is at risk of being burned, specific correction suggestions are generated.

[0421] Input: Flame risk assessment results

[0422] Output: Revision proposal

[0423] Specific behavior: The server uses the generative AI model again to generate specific revision suggestions to reduce the risk of flame wars. The generative AI model generates revision suggestions based on the prompt sentence and returns the results to the server.

[0424] Step 5:

[0425] The server sends the evaluation results and suggested modifications to the user's device.

[0426] Input: Evaluation results, proposed corrections

[0427] Output: Evaluation results and suggested fixes sent to the user's device

[0428] Specific operation: The server compiles the evaluation results and revision suggestions and sends them to the user's device. The user can check this information on their device and revise the ad copy as necessary.

[0429] Step 6:

[0430] The user corrects the information based on the suggested corrections and sends it back to the server.

[0431] Input: Information corrected based on proposed amendment

[0432] Output: The modified information is sent to the server

[0433] Specific operation: The user modifies the ad copy based on the suggested revisions received from the server and sends it back to the server. The server receives the revised information and re-evaluates it.

[0434] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0435] "Example 1"

[0436] In one example of an invention incorporating an emotion engine, an emotion engine that recognizes user emotions is combined with a system that includes AI that utilizes data on past flame wars and effective words, and predicts the economic impact of publishing information on social media or websites based on this data. Specifically, the system analyzes emotions from text data posted by users, and corrects or creates public information based on the results. For example, if a user posts text expressing joy, that information is used to publish information that reinforces a positive image of a company's products or services.

[0437] "Example 2"

[0438] As a second example of the invention incorporating an emotion engine, in the system described in claim 2, an emotion engine that recognizes the user's emotions is combined with the means for determining the amount of loss when past flame data is used during editing. Specifically, the user's emotions are analyzed, and the risk of a flame war is assessed based on the results. For example, if a user posts text that expresses anger, that information is used to avoid publishing information that increases the risk of a flame war.

[0439] "Example 3"

[0440] As a third embodiment of the invention incorporating an emotion engine, in the system described in claim 3, an emotion engine that recognizes user emotions is combined with a means for preventing damage to the company's image in advance. Specifically, the system analyzes the user's emotions and, based on the results, publishes information to prevent damage to the company's image. For example, if a user posts text expressing sadness, that information is used to publish information about the company's response and service improvements.

[0441] The processing flow of each embodiment will be described below.

[0442] "Example 1"

[0443] Step 1: A user posts text data on a social networking site or website.

[0444] Step 2: The emotion engine analyzes the user's emotions from the text data.

[0445] Step 3: The AI ​​predicts the economic impact of the posted text data based on past flame war data and data on effective words.

[0446] Step 4: Based on the analysis results of the emotion engine and the prediction results of the AI, edit or create public information.

[0447] "Example 2"

[0448] Step 1: A user posts text data on a social networking site or website.

[0449] Step 2: The emotion engine analyzes the user's emotions from the text data.

[0450] Step 3: The AI ​​uses past flame war data to determine the amount of loss caused by the posted text data.

[0451] Step 4: Evaluate the risk of a social media outrage based on the results of the emotion engine analysis and the AI's understanding of the loss amount.

[0452] "Example 3"

[0453] Step 1: A user posts text data on a social networking site or website.

[0454] Step 2: The emotion engine analyzes the user's emotions from the text data.

[0455] Step 3: Based on past data on controversy and effective words, the AI ​​predicts the impact of the posted text data on the company's image.

[0456] Step 4: Based on the analysis results of the emotion engine and the prediction results of the AI, disclose information to prevent damage to the company's image.

[0457] Example 1

[0458] Next, a description will be given of Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0459] When companies release information on social media or websites, they are required to predict the economic impact by utilizing data on past scandals and effective keywords. However, there is a lack of systems to effectively utilize this data, making it difficult to prevent damage to corporate image and economic loss. Furthermore, there is a lack of methods to analyze user sentiment and edit or create information based on that sentiment, preventing effective corporate information disclosure strategies.

[0460] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0461] In this invention, the server includes means for collecting and storing data on past flame wars, means for collecting and storing data on effective words, means for using artificial intelligence to predict the economic impact of publishing on social media or websites based on this data, means for analyzing emotions from text data posted by users, and means for editing or creating published information based on the results of the emotion analysis.This enables companies to predict the economic impact by utilizing data on past flame wars and effective words, and to publish information based on user emotions.

[0462] "Past flame war data" refers to detailed information on past flame war cases that have occurred on social media or the internet.

[0463] "Effective word data" refers to information on words and phrases that are considered effective in marketing and advertising.

[0464] "SNS" is an abbreviation for social networking service, which refers to an online platform where users can share information and interact with each other.

[0465] A "website" is a collection of information made public on the Internet, and refers to an online page where companies and individuals disseminate information.

[0466] "Economic impact" refers to the impact that a particular action or event has on the economy, specifically including increases or decreases in sales and profits.

[0467] "Artificial intelligence" refers to technology that uses machine learning and data analysis to mimic human intelligence.

[0468] "User" refers to any individual or entity that uses the System or Services.

[0469] "Text data" refers to digital data that contains text information.

[0470] "Sentiment analysis" refers to the technology of identifying emotions from text data and evaluating the type and intensity of those emotions.

[0471] "Public information" refers to information that is made publicly available through social media or websites.

[0472] "Correction" refers to the act of revising existing text and changing it to more appropriate expression.

[0473] "Creating" refers to the act of generating new text or information.

[0474] MODE FOR CARRYING OUT THE INVENTION

[0475] This invention is an AI-based information disclosure support system for companies that uses data on past flame wars and effective keywords to predict the economic impact of information published on social media and websites. It also includes a function to analyze user sentiment and edit or create published information based on the results.

[0476] Data collection and storage

[0477] The server collects past flame war cases that have occurred on social media and the internet. Specifically, the data is obtained using a web scraping tool (e.g., Beautiful Soup). The collected data is stored in a database (e.g., MySQL (registered trademark), PostgreSQL (registered trademark)). Similarly, data on effective keywords is collected and stored in the same database.

[0478] Analyzing the data

[0479] The server analyzes the accumulated flame data using an AI model (e.g., TensorFlow (registered trademark), PyTorch (registered trademark)). Specifically, it uses natural language processing (NLP) technology to extract the causes and patterns of flame wars. Data on effective words is also analyzed using an AI model to identify which words are effective in what situations. These analysis results are stored in a database.

[0480] Economic impact forecast

[0481] The server predicts the economic impact of information published on social media and websites based on past flame war data and data on effective keywords. Specifically, it uses a machine learning model to simulate the impact the published information will have. The prediction results are stored in a database.

[0482] Combining Emotion Engines

[0483] The device sends the text data posted by the user to an emotion engine (e.g., IBM Watson (registered trademark) or Microsoft (registered trademark) Azure (registered trademark) emotion analysis API). The server receives the analysis results returned from the emotion engine and stores them in a database. Based on the results of the emotion analysis, the server edits or creates public information. Specifically, based on text that shows positive emotion, it generates information that reinforces a positive image of a company's products or services. The generated information is published on social media and websites.

[0484] Specific examples

[0485] For example, if a user posts the text "I really like the new product!", the device sends this text to the emotion engine. The server receives the emotion analysis result of "joy" from the emotion engine, and based on this analysis result generates positive information such as "Many users like the new product," which is then published on the SNS.

[0486] Prompt Sentence Examples

[0487] Examples of prompts to input to a generative AI model might include:

[0488] "Based on past data on flame wars and data on effective words, please predict the economic impact of information published on social media and websites. Also, please analyze user sentiment and correct the published information based on the results."

[0489] The above is an embodiment of the invention.

[0490] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0491] Step 1:

[0492] The server collects past incidents of online outrage that have occurred on social media and the internet. Specifically, it acquires data using a web scraping tool (e.g., Beautiful Soup). The input is text data from social media and websites, and the output is the collected data on incidents of online outrage. This data is stored in a database (e.g., MySQL (registered trademark), PostgreSQL (registered trademark)).

[0493] Step 2:

[0494] The server collects data on words that are considered effective in marketing and advertising. This is done by utilizing existing marketing databases and expert knowledge. The input is word data from the marketing database, and the output is the collected data on effective words. This data is also stored in the same database.

[0495] Step 3:

[0496] The server analyzes the accumulated flame data using an AI model (e.g., TensorFlow (registered trademark), PyTorch (registered trademark)). Specifically, it uses natural language processing (NLP) technology to extract the causes and patterns of flames. The input is the flame data stored in the database, and the output is the analysis results of the causes and patterns of flames. These analysis results are stored in the database.

[0497] Step 4:

[0498] The server also uses an AI model to analyze the effective word data and determine which words are effective in what situations. The input is the effective word data stored in the database, and the output is the analysis results of the effective words. These analysis results are also stored in the database.

[0499] Step 5:

[0500] The server predicts the economic impact of information published on social media and websites based on past flame data and data on effective keywords. Specifically, it uses a machine learning model to simulate the extent of the reaction the published information will have. The input is the flame data and the analysis results of effective keywords, and the output is the predicted economic impact. The predicted results are stored in a database.

[0501] Step 6:

[0502] The device sends the text data posted by the user to an emotion engine (e.g., IBM Watson or Microsoft Azure's emotion analysis API). The input is the text data posted by the user, and the output is the emotion analysis result by the emotion engine.

[0503] Step 7:

[0504] The server receives the analysis results returned from the emotion engine and stores them in a database. The input is the emotion analysis results, and the output is the emotion analysis results stored in the database.

[0505] Step 8:

[0506] The server corrects and creates public information based on the results of the sentiment analysis. Specifically, it generates information that reinforces a positive image of a company's products and services based on text that indicates positive sentiment. The input is the sentiment analysis results, and the output is the generated public information. The generated information is published on social media and websites.

[0507] Step 9:

[0508] For example, if a user posts the text "I really like the new product!", the device sends this text to the emotion engine. The server receives the emotion analysis result of "joy" from the emotion engine, and based on this analysis result generates positive information such as "Many users like the new product," which is then published on the SNS.

[0509] (Application example 1)

[0510] Next, a description will be given of Application Example 1 of Embodiment Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0511] When a company publishes information on social media or its website, it is important to predict the economic impact by utilizing data on past controversies and effective keywords. However, conventional systems lack the functionality to analyze user sentiment and generate optimal advertising copy based on that, making it difficult to improve the company's image and maximize the economic impact.

[0512] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means. In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including AI for predicting the economic effect of publishing on SNS or a website based on this data, means for analyzing user emotions, means for generating advertising copy based on the results of the emotion analysis, and means for predicting the economic effect of the generated advertising copy. This enables companies to generate optimal advertising copy based on user emotions and maximize the economic effect.

[0513] "Past flame war data" refers to data on negative reactions, critical comments, posts, etc. that have occurred in the past on social media or websites.

[0514] "Effective word data" refers to data on keywords and phrases that have generated positive responses and high economic impact on social media and websites.

[0515] "AI that predicts economic impact" is artificial intelligence that uses past data to predict the extent of economic impact that information published on social media or websites will have.

[0516] "Means for analyzing user emotions" refers to technology that analyzes emotions from text data posted by users and identifies emotional states such as positive, negative, and neutral.

[0517] "Means for generating advertising copy based on the results of sentiment analysis" refers to a technology that automatically generates optimal advertising copy using the results of user sentiment analysis.

[0518] The "means for predicting the economic effect of the generated advertising copy" is a technique for predicting the degree of economic impact that the generated advertising copy will have.

[0519] The following system configuration will be described as an embodiment of the present invention.

[0520] The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including AI for predicting the economic impact when such data is published on social media or a website, a means for analyzing user emotions, a means for generating advertising copy based on the results of the emotion analysis, and a means for predicting the economic impact of the generated advertising copy.

[0521] Hardware and software used

[0522] Hardware: Servers, smartphones

[0523] Software: Python (registered trademark), pandas, scikit-learn (registered trademark), TextBlob, OpenAI API

[0524] Data processing and calculation

[0525] 1. Data loading: The server loads past flame data and effective word data from a CSV file.

[0526] 2. Vectorization: The server vectorizes the text data using TfidfVectorizer.

[0527] 3. Model training: The server trains the economic impact prediction model using LinearRegression.

[0528] 4. Sentiment Analysis: The server uses TextBlob to analyze the sentiment of the user's posted text.

[0529] 5. Ad copy generation: The server uses the OpenAI API to generate ad copy based on sentiment.

[0530] 6. Economic impact prediction: The server uses the trained model to predict the economic impact of the generated ad copy.

[0531] Specific examples

[0532] When a user uses a smartphone to post the text "I want to create an advertisement for a new smartphone," the server performs the following process.

[0533] 1. The server performs sentiment analysis on the user's posted text using TextBlob to detect positive sentiment.

[0534] 2. The server uses the OpenAI API to generate the following prompt text and create an ad copy.

[0535] Example prompt: "Create a positive advertisement text for the following product: I want to create an advertisement for a new smartphone."

[0536] 3. The server predicts the economic effect based on the generated advertising copy.

[0537] In this way, companies can generate optimal advertising copy based on user emotions and maximize economic benefits.

[0538] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0539] Step 1:

[0540] The server reads past flame data and effective word data from a CSV file. As input, it receives the CSV file of past flame data and effective word data, and as output, it stores this data in memory in data frame format. Specifically, it uses the pandas library to read the CSV file and convert it to a data frame.

[0541] Step 2:

[0542] The server vectorizes the text data using TfidfVectorizer. It receives the text string of past flame data as input and obtains a vector representation of the text data as output. Specifically, it initializes TfidfVectorizer and vectorizes the text data using the fit_transform method.

[0543] Step 3:

[0544] The server trains an economic impact prediction model using LinearRegression. It receives vectorized text data and corresponding economic impact data as input and obtains a trained economic impact prediction model as output. Specifically, it initializes a LinearRegression model and trains the model using the fit method.

[0545] Step 4:

[0546] The server uses TextBlob to analyze the sentiment of user-posted text. It receives the text data posted by the user as input and obtains the sentiment analysis result (positive, negative, or neutral) as output. Specifically, it creates a TextBlob object and analyzes the sentiment using the sentiment property.

[0547] Step 5:

[0548] The server uses the OpenAI API to generate ad copy based on emotions. It receives the emotion analysis results and the user's posted text as input, and obtains the generated ad copy as output. Specifically, it generates a prompt based on the emotion analysis results and sends a request to the OpenAI API to generate the ad copy.

[0549] Step 6:

[0550] The server uses the trained model to predict the economic impact of the generated ad copy. It receives the generated ad copy as input and obtains the predicted economic impact value as output. Specifically, it vectorizes the generated ad copy and predicts the economic impact using the trained model's predict method.

[0551] Example 2

[0552] Next, a description will be given of Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0553] In today's information society, when companies or individuals publish information on social media or their websites, it is important to properly assess the risk of a controversy and prevent economic losses before they occur. However, conventional systems were unable to fully utilize past controversy data or user sentiment, resulting in insufficient assessment of the risk of a controversy. Furthermore, there was a lack of means to prevent damage to a company's image in advance. This made risk management difficult when disclosing information, and there was a high possibility of economic losses and damage to the company's image occurring.

[0554] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0555] In this invention, the server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic impact of information disclosure based on this data, a means for analyzing user input text in real time and recognizing emotions, and a means for assessing the risk of flame wars based on the results of the emotion analysis. This makes it possible to specifically quantify the risk of flame wars before information disclosure, thereby preventing economic losses and damage to the company's image.

[0556] "Past flame war data" refers to information about past flame war cases that occurred on social media, websites, etc., and specifically includes data such as the content of posts, comments, and related losses.

[0557] "Effective word data" refers to data on keywords and phrases that are believed to have a positive impact when information is released.

[0558] "Artificial intelligence that predicts the economic impact of information disclosure" refers to machine learning models and algorithms that use past data to predict the economic impact of information disclosure.

[0559] "Means for analyzing user-entered text in real time and recognizing emotions" refers to technology for instantly analyzing text entered by a user and identifying the emotions contained in that text.

[0560] "Means for assessing the risk of outrage based on the results of sentiment analysis" refers to technology that uses the results of sentiment analysis to assess the likelihood that input text will cause outrage.

[0561] "Means of determining the amount of loss" refers to technology for identifying and quantifying the economic losses incurred during a social media outburst based on data from past social media outbursts.

[0562] "Means to prevent damage to a company's image in advance" refers to technology that assesses the risk of a backlash before information is made public, preventing a company's reputation and image from being damaged.

[0563] MODE FOR CARRYING OUT THE INVENTION

[0564] This invention is a system that utilizes past flame data, analyzes user input text in real time to recognize emotions, and evaluates the risk of a flame war. Specific embodiments of this system are described below.

[0565] Server Processing

[0566] The server first collects data on past online scandals. It uses SQL queries from the database to retrieve data on past online scandals, including posts, comments, and associated losses. It then trains a model to predict losses based on the collected data. It preprocesses the data using Python's Pandas library and builds a regression model using Scikit-learn. The trained model is saved as a file using the Pickle library for later use.

[0567] Terminal handling

[0568] The device analyzes the text entered by the user in real time and calls the emotion analysis API to recognize emotions. Specifically, it receives the user's text input and analyzes the emotion of the text using the Google Cloud Natural Language API. The analysis results are displayed to the user on the device. For example, if the emotion "anger" is recognized, the user is notified.

[0569] User operations

[0570] Users input text into the system, which evaluates the risk of that text sparking a controversy. For example, they input the content of a social media post or blog article. The system evaluates the risk of the input text sparking a controversy based on past controversy data and the results of sentiment analysis. The evaluation results are displayed to the user, who can then modify the content of their post based on the results.

[0571] Specific examples

[0572] Example prompts to input to the generative AI model

[0573] "Design a system that uses past data on flaming incidents to assess the risk of specific text becoming flaming. Include functionality to analyze user-entered text in real time, recognize emotions using an emotion engine, and assess the risk of flaming based on the results."

[0574] This system allows users to quantify and confirm the risk of a controversy before posting, making it possible to disclose information while taking the risks into consideration.

[0575] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0576] System program processing flow

[0577] Server Processing

[0578] Step 1: Collect data on past flame wars

[0579] The server collects data on past incidents of online outrage from a database, specifically using SQL queries to retrieve data such as posts, comments, and associated losses.

[0580] Input: Database connection information, SQL query

[0581] Output: Past fire dataset

[0582] Specific operation: The server executes the SQL query "SELECT FROM EnjoData WHERE Date BETWEEN (registered trademark) '2018-01-01' AND '2023-01-01';" to retrieve the required data from the database.

[0583] Step 2: Preprocessing the data

[0584] The server preprocesses the collected data, specifically by using the Pandas library in Python (registered trademark) to fill in missing values ​​and remove outliers.

[0585] Input: Historical fire dataset

[0586] Output: Preprocessed dataset

[0587] Specific behavior: The server uses Pandas to perform operations such as "df.fillna(0)" and "df.dropna()" to clean up the data.

[0588] Step 3: Training the loss prediction model

[0589] The server trains a model to predict loss amounts based on the preprocessed data. Specifically, it builds a regression model using Scikit-learn (registered trademark).

[0590] Input: Preprocessed dataset

[0591] Output: A trained loss prediction model

[0592] Specific operation: The server executes code such as "from sklearn.linear_model import LinearRegression" and "model = LinearRegression().fit(X_train, y_train)" to train the model.

[0593] Step 4: Save the model

[0594] The server saves the trained model as a file for later use. It uses the Pickle library to save the model.

[0595] Input: A trained loss prediction model

[0596] Output: Saved model file

[0597] Specific operation: The server executes the code "import pickle" and "with open('loss_prediction_model.pkl', 'wb') as file: pickle.dump(model, file)" to save the model.

[0598] Terminal handling

[0599] Step 5: Receiving User Input

[0600] The device receives text entered by the user, for example, from a text box in a web form or mobile app.

[0601] Input: User text input

[0602] Output: Received text data

[0603] Specific operation: The device says, "The user enters 'I can't use this product at all' into the web form."

[0604] Step 6: Call the sentiment analysis API

[0605] The device sends the input text to the sentiment analysis API to recognize emotions, and uses the Google Cloud Natural Language API to analyze the sentiment of the text.

[0606] Input: Received text data

[0607] Output: Emotion analysis results

[0608] Specific operation: The device executes the code "response = client.analyze_sentiment(document=document)" and "sentiment = response.document_sentiment" to analyze the sentiment.

[0609] Step 7: Viewing the analysis results

[0610] The device displays the emotion analysis results to the user. For example, if the emotion "anger" is recognized, the device notifies the user.

[0611] Input: Sentiment analysis results

[0612] Output: A message to inform the user

[0613] Specific behavior: The device "displays to the user 'This text contains angry sentiment.'"

[0614] User operations

[0615] Step 8: Entering Text

[0616] A user inputs text into the system, for example, the content of a social media post or blog post.

[0617] Input:Text content

[0618] Output: Send text data to the system

[0619] Specific operation: The user types "This service is completely unusable" and submits it to the system.

[0620] Step 9: Identify risk of outrage

[0621] The user checks the results of the flame risk assessment provided by the system, which assesses the risk based on past flame data and the results of sentiment analysis.

[0622] Input: Evaluation results from the system

[0623] Output: Display evaluation results to the user

[0624] Specific operation: The system displays the evaluation result, "This post has a high risk of causing a firestorm," and the user confirms it.

[0625] In this way, users can confirm the risk of a controversy in concrete terms and quantify it before posting, making it possible to disclose information while taking the risks into account.

[0626] (Application example 2)

[0627] Next, a description will be given of Application Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0628] In today's information society, when companies or individuals publish information on social media or their websites, they are required to use past data on flame wars to assess the risk of flame wars and prevent economic losses. However, conventional systems lack the means to analyze user emotions and assess the risk of flame wars, making it difficult to provide advance warnings of high-risk posts.

[0629] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0630] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic impact of posting on a social networking site or website based on this data, means for analyzing user emotions, means for assessing the risk of a flame war, and means for issuing advance warnings of high-risk posts. This makes it possible to assess the risk of a flame war taking user emotions into consideration and to issue advance warnings of high-risk posts.

[0631] "Past flame war data" refers to data on flame wars that have occurred in the past due to information that was made public on social media, websites, etc.

[0632] "Effective word data" refers to data on words and phrases that are likely to provoke a positive response when publishing information on social media or a website.

[0633] "Artificial intelligence" is the technology that enables computer systems to learn, reason, and self-correct by imitating human intelligence.

[0634] "Means for analyzing user emotions" refers to technology that identifies emotions from text and comments posted by users and evaluates the type and intensity of those emotions.

[0635] The "means for assessing the risk of a backlash" is a technology that quantifies and evaluates the possibility that the release of information will cause a backlash based on past backlash data and the results of user sentiment analysis.

[0636] "Means for providing advance warnings about high-risk posts" refers to technology that issues advance warnings to users about posts that are evaluated as having a high risk of causing a flame war.

[0637] The system for carrying out this invention consists of three main elements: a server, a terminal, and a user. The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic effects of posting on social media or a website based on this data, a means for analyzing user emotions, a means for assessing the risk of flame wars, and a means for issuing advance warnings about high-risk posts.

[0638] The server stores past flame data and effective word data in a database, and uses artificial intelligence to predict economic impact based on this data. Specifically, the server uses a text analysis engine (e.g., TextBlob) to analyze user posts and identify their sentiment. Furthermore, it uses a machine learning model (e.g., LinearRegression) to assess the risk of flame wars based on past flame war data and the results of sentiment analysis.

[0639] When a user posts to a social networking site or website, the device receives the results of a flame war risk assessment provided by the server and displays a warning for posts with a high risk, allowing the user to reconsider the content of their post and reduce the risk of a flame war.

[0640] For example, if a user tries to post "This product is completely unusable!", the device will send this text to the server. The server will then use a text analysis engine to analyze the sentiment and identify anger. It will then use a machine learning model to assess the risk of outrage and, if a high risk is detected, will display a warning on the device.

[0641] Example prompt sentence:

[0642] Enter your user post: This product is completely unusable!

[0643] This system allows users to check the risk of a controversy before posting and avoid high-risk posts, thereby preventing economic losses for companies and individuals.

[0644] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0645] Step 1:

[0646] The user uses the device to input text to post on a social networking site or website. The input text is sent from the device to the server. The input data is the content of the user's post.

[0647] Step 2:

[0648] The server passes the received text to a text analysis engine (e.g., TextBlob) for sentiment analysis. The text analysis engine identifies the emotion of the text and evaluates the type (e.g., anger, joy, sadness) and intensity of the emotion. The input data is the user's post, and the output data is the type and intensity of the emotion.

[0649] Step 3:

[0650] The server evaluates the risk of a controversy using a machine learning model (e.g., Linear Regression) based on the results of the emotion analysis. The machine learning model receives past controversy data and the emotion analysis results as input, and quantifies the risk of a controversy. The input data is the type and intensity of emotion, as well as past controversy data, and the output data is a numerical value for the risk of a controversy.

[0651] Step 4:

[0652] The server sends the results of the risk assessment to the terminal. The terminal displays a warning to the user if the risk is high based on the received risk value. The input data is the risk value, and the output data is a warning message.

[0653] Step 5:

[0654] The user checks the warning message displayed on the terminal and reconsiders the content of the post. If necessary, the user corrects the content of the post and sends it back to the server. The input data is the warning message, and the output data is the corrected content of the post.

[0655] This series of processes allows users to check the risk of a controversy before posting and avoid high-risk posts, thereby preventing economic losses for companies and individuals.

[0656] Example 3

[0657] Next, a description will be given of a third embodiment of the third embodiment. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0658] Companies are being asked to prevent the risk of information they disclose becoming a hot topic in advance and to prevent damage to their corporate image. However, with conventional systems, it is difficult to consistently generate and edit information, evaluate the risk of a hot topic, and analyze user sentiment, and effective countermeasures are often not possible. There is also a lack of means to predict the economic impact by utilizing data on past hot topics and effective words. This means that companies are taking on significant risks when disclosing information.

[0659] The identification process by the identification processing unit 290 of the data processing device 12 in Example 3 is realized by the following means. In this invention, the server includes means for utilizing past flame data, means for utilizing data on effective words, means for using a generative AI model for generating information based on these data, means for correcting the generated information, means for evaluating the flame risk of the generated or corrected information, means for prompting correction if the flame risk is high, means for using an emotion engine for analyzing user emotions, means for publishing information based on the analysis results, and means for predicting the economic effect of publishing the information on an SNS or website based on these data. This makes it possible to prevent a damage to the company's image by evaluating in advance the flame risk of information published by a company and prompting correction as necessary.

[0660] "Past flame war data" refers to data on past flame wars that occurred as a result of information that was made public on social media, websites, etc.

[0661] "Effective word data" refers to data on words and phrases that are considered effective in eliciting positive responses when information is released.

[0662] A "generative AI model" is an artificial intelligence model that generates natural language text based on an input prompt.

[0663] "Correction" is the process of checking the grammar and expression of the generated information for appropriateness and correcting it if necessary.

[0664] "Flame risk assessment" is the process of analyzing the possibility of a flame war occurring when generated or edited information is made public, and determining the risk.

[0665] "Methods to encourage corrections" are methods of notifying users that information that is deemed to pose a high risk of causing a backlash needs to be corrected.

[0666] An "emotion engine" is a technology that analyzes text posted by users and identifies the emotions (joy, sadness, anger, etc.) contained in that text.

[0667] "Information disclosure" is the process of publishing information based on the analysis results on social media, websites, etc. in order to prevent a company's image from being damaged.

[0668] "Economic impact forecasting" is the process of predicting the economic impact that publicly disclosed information will have on a company.

[0669] MODE FOR CARRYING OUT THE INVENTION

[0670] This invention is a system that prevents damage to a company's image by assessing in advance the risk of information that a company discloses becoming a target of criticism and prompting corrections as necessary. A specific embodiment of this system will be described below.

[0671] Hardware and software used

[0672] Hardware: Servers, terminals

[0673] Software: Generative AI model, emotion engine, database management system (DBMS)

[0674] System configuration

[0675] 1. The server stores past flame data in a database and has the means to utilize this data.

[0676] 2. The server stores data on effective words in a database and has the means to utilize this data.

[0677] 3. The server has a means to generate information using a generative AI model.

[0678] 4. The server has a means of correcting the generated information.

[0679] 5. The server has a means to assess the risk of flame wars for the information generated or edited.

[0680] 6. The server has the means to prompt corrections if there is a high risk of outrage.

[0681] 7. The server has a means to use an emotion engine to analyze the user's emotions.

[0682] 8. The server has a means to publish information based on the analysis results.

[0683] 9. The server has the means to predict the economic impact of publishing this data on social media or on a website.

[0684] System Operation

[0685] 1. The server uses the generative AI model to generate information that the company is planning to release. For example, the server inputs a prompt, "Please generate a description that explains the features of the new product," into the generative AI model to generate a description of the new product.

[0686] 2. The server uses AI to correct the generated information. Specifically, it checks grammar and expressions for appropriateness, and makes corrections as necessary.

[0687] 3. The server analyzes the corrected information and assesses the risk of a social media outrage. Natural language processing (NLP) technology is used to detect risk factors within the text.

[0688] 4. If the server is judged to be at high risk of being exposed to malicious content, a notification is sent to the device urging the user to make corrections. The user receives the notification and corrects the information as necessary.

[0689] 5. The server uses an emotion engine to analyze the user's emotions. For example, if a user posts "The price of the new product is too high," the server analyzes the text and detects the emotion of sadness.

[0690] 6. Based on the analysis results, the server publishes information to prevent damage to the company's image, such as explanations about prices and information about service improvements.

[0691] Specific examples

[0692] For example, consider the case where a company uses this system when announcing a new product.

[0693] 1. The server inputs the prompt "Please generate an introductory text that describes the features of the new product" into the generative AI model.

[0694] 2. The generative AI model generates a description such as, "This new product features high performance and an easy-to-use design."

[0695] 3. The server uses AI to check the grammar of the description it generates and corrects it to, "It features high performance and an easy-to-use design."

[0696] 4. The server analyzes the revised description using NLP technology and determines that the expression "high performance" may be implicitly criticizing other companies' products.

[0697] 5. The server sends a notification to the device prompting the correction, suggesting that the expression "high performance" be changed to "multi-functional."

[0698] 6. The user receives the notification and the message is revised to read, "It features a versatile and user-friendly design."

[0699] 7. The server uses an emotion engine to analyze the text posted by the user, "The price of the new product is too high," and detects the emotion of sadness.

[0700] 8. Based on the analysis results, the server publishes an "Explanation of the New Product's Price," explaining that "This new product uses high-quality materials, which is why it is priced higher."

[0701] In this way, the system for preventing damage to a company's image in advance functions concretely. The flow of the identification process in the third embodiment will be described with reference to FIG.

[0702] Step 1:

[0703] The server inputs a prompt into the generative AI model to generate information. Specifically, the prompt is "Please generate an introduction that explains the features of our new product." Based on this prompt, the generative AI model generates an introduction such as "This new product is characterized by its high performance and easy-to-use design." The input is the prompt, and the output is the generated introduction.

[0704] Step 2:

[0705] The server uses AI to correct the generated information. Specifically, it checks the grammar of the generated introduction and confirms the appropriateness of the expression. For example, the sentence "It features high performance and an easy-to-use design" is corrected to "It features high performance and an easy-to-use design." The input is the generated introduction, and the output is the corrected introduction.

[0706] Step 3:

[0707] The server analyzes the corrected information and evaluates the risk of a controversy. Specifically, it uses natural language processing (NLP) technology to detect risk factors within the text. For example, it determines that the expression "high performance" may be an implicit criticism of another company's product. The input is the corrected introduction, and the output is the evaluation result of the risk of a controversy.

[0708] Step 4:

[0709] If the server determines that there is a high risk of a flame war, it sends a notification to the terminal urging the user to make corrections. Specifically, it notifies the user of the parts that need to be corrected and the reasons for the corrections. For example, it suggests changing the expression "high performance" to "multi-functional." The input is the evaluation result of the flame war risk, and the output is a notification urging the user to make the corrections.

[0710] Step 5:

[0711] The user receives the notification and corrects the information as necessary. Specifically, the description is corrected based on the notification. For example, the sentence "It features high performance and an easy-to-use design" is corrected to "It features multifunctionality and an easy-to-use design." The input is the notification prompting the correction, and the output is the corrected description.

[0712] Step 6:

[0713] The server uses an emotion engine to analyze the user's emotions. Specifically, it analyzes the text posted by the user and identifies the type of emotion (joy, sadness, anger, etc.). For example, if a user posts "The price of the new product is too high," the text is analyzed to detect the emotion of sadness. The input is the text posted by the user, and the output is the result of the emotion analysis.

[0714] Step 7:

[0715] Based on the analysis results, the server publishes information to prevent damage to the company's image. Specifically, it uses the analysis results to provide explanations about service improvements and prices. For example, it publishes an "explanation about the price of a new product" and explains that "this new product uses high-quality materials, and therefore the price is set higher." The input is the result of the sentiment analysis, and the output is the information to be published.

[0716] (Application example 3)

[0717] Next, a description will be given of Application Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0718] In today's information society, there is an increasing risk that information released by a company will cause a backlash, which could damage the company's image and lead to financial losses. Furthermore, releasing information without regard for user feelings can worsen the relationship between the company and its users. To solve these problems, it is necessary to evaluate the risk of a backlash before releasing information and optimize the information while taking user feelings into consideration.

[0719] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 3 is realized by the following means. In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effect of information disclosure based on this data, means for analyzing user emotions, and means for optimizing information based on the analysis results. This makes it possible to evaluate the risk of flame wars before information disclosure and to optimize information taking user emotions into consideration.

[0720] "Past online outrage data" refers to data on cases of online outrage caused by information that was made public in the past.

[0721] "Effective word data" is data on words and phrases that are expected to elicit a positive response when used when disclosing information.

[0722] "Artificial intelligence that predicts the economic impact of disclosure" is artificial intelligence that calculates the expected economic impact if information is made public.

[0723] "Means for analyzing user emotions" refers to technology for analyzing the emotions expressed by a user from text, voice, etc.

[0724] "Means for optimizing information" refers to technology for correcting and adjusting the information to be made public in an optimal form based on analysis results and predictive data.

[0725] The system for carrying out this invention comprises a server and a user terminal. The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic effects of disclosure, a means for analyzing user sentiment, and a means for optimizing information based on the analysis results.

[0726] Hardware and software used

[0727] Hardware:

[0728] Server (cloud-based)

[0729] User device (smartphone, PC)

[0730] software:

[0731] Sentiment analysis engines (e.g., IBM Watson, Google Cloud Natural Language API)

[0732] Flaming risk assessment AI model (e.g., OpenAI GPT-4 (registered trademark))

[0733] Ad generation AI model (e.g., OpenAI GPT-4 (registered trademark))

[0734] Data processing and calculation

[0735] 1. Ad generation:

[0736] The user device sends basic information about the advertising campaign (target audience, product information, etc.) to the server.

[0737] The server uses the ad generation AI model to generate ad copy.

[0738] 2. Flaming Risk Assessment:

[0739] The server inputs the generated advertising copy into an AI model for assessing the risk of a controversy and performs a risk assessment.

[0740] If the risk is high, the server will suggest a correction to the user terminal.

[0741] 3. Emotion analysis:

[0742] The server collects users' emotional data (e.g., social media posts, feedback, etc.).

[0743] A sentiment analysis engine is used to analyze the user's sentiment.

[0744] Based on the analysis results, the server optimizes the advertising content.

[0745] Specific examples

[0746] For example, if you are running an advertising campaign for a new smartphone, you can send the following prompt message from the user's device to the server:

[0747] Prompt Sentence Examples

[0748] Target Audience: Young people (18-25 years old)

[0749] Product Information: New smartphone "XPhone" with high-performance camera, long battery life, and latest OS

[0750] Tone of the ad copy: Casual and friendly

[0751] Based on this prompt, the server uses a generative AI model to generate ad copy, and an AI model for assessing the risk of flame wars evaluates the risk. If the risk is high, it suggests modifications, and finally, a sentiment analysis engine optimizes the ad content based on the user's emotions.

[0752] The above is an embodiment of the present invention.

[0753] The flow of the specific processing in Application Example 3 will be described with reference to FIG.

[0754] Step 1:

[0755] The user inputs basic information about the advertising campaign (target audience, product information, tone of the advertising copy, etc.) into the user terminal, which is then sent to the server.

[0756] Step 2:

[0757] The server generates ad copy using an ad generation AI model based on the received basic information. The generated ad copy is temporarily stored on the server.

[0758] Step 3:

[0759] The server inputs the generated ad copy into an AI model for assessing the risk of a potential flame war and performs a risk assessment. The results of the risk assessment are output, and if the risk is high, the server generates a revision proposal.

[0760] Step 4:

[0761] The server sends the risk assessment results and proposed revisions to the user terminal, and the user checks the proposed revisions and revises the advertisement copy as necessary.

[0762] Step 5:

[0763] The server collects users' emotional data (e.g., social media posts, feedback, etc.) and inputs the collected emotional data into a sentiment analysis engine.

[0764] Step 6:

[0765] The server uses an emotion analysis engine to analyze the user's emotions, and the analysis results are output and stored in the server.

[0766] Step 7:

[0767] The server optimizes the ad content based on the analysis results, and the optimized ad copy is stored on the server.

[0768] Step 8:

[0769] The server sends the optimized ad copy to the user terminal, and the user checks the optimized ad copy and publishes it as the final ad copy.

[0770] The above is the processing flow of this program.

[0771] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0772] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (registered trademark) (Internet search engine).<URL: https: / / openai.com / blog / chatgpt> ) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0773] Another example of generative AI is Gemini (registered trademark) (Internet search engine). <url: https: gemini.google.com ?hl="ja">) are listed.

[0774] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0775] [Second embodiment]

[0776] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0777] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0778] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0779] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0780] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0781] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0782] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0783] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0784] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0785] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0786] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0787] Next, the specific processing by the specific processing unit 290 of the data processing device 12 will be described.

[0788] "Example 1"

[0789] One embodiment of the present invention is an information disclosure support system for companies that utilizes AI. This system utilizes data on past flame wars and data on effective words, and predicts the economic impact of publishing information on social media or websites based on this data. Specifically, past flame war data is stored in a database and analyzed by AI. Data on effective words is also stored in the same way, and the published information is edited and created based on this data.

[0790] "Example 2"

[0791] Furthermore, another embodiment of the present invention has a function to grasp the amount of loss when using past flame data during correction. Specifically, the amount of loss in the event of a flame is estimated from past flame data and used as reference information for correction. This makes it possible to specifically quantify the risk of a flame and disclose information taking that risk into account.

[0792] "Example 3"

[0793] Furthermore, a further embodiment of the present invention has a function to prevent damage to a company's image in advance. Specifically, before information created or corrected by AI is made public, it evaluates whether the information poses a risk of causing a backlash and prompts corrections as necessary. This makes it possible to prevent damage to a company's image in advance.

[0794] The processing flow of each embodiment will be described below.

[0795] "Example 1"

[0796] Step 1: Accumulate data on past controversies and effective words in a database.

[0797] Step 2: AI analyzes this data and edits or creates public information.

[0798] Step 3: Based on the economic impact predicted by AI, the optimal timing and content of information disclosure are determined.

[0799] "Example 2"

[0800] Step 1: Estimate the amount of loss in the event of a fire based on past fire data.

[0801] Step 2: Use the estimated loss amount as a reference for corrections.

[0802] Step 3: Quantify the risk of a backlash and disclose information based on that risk.

[0803] "Example 3"

[0804] Step 1: Before information created or edited by AI is made public, assess whether it poses a risk of causing outrage.

[0805] Step 2: Based on the evaluation results, encourage correction of information as necessary.

[0806] Step 3: Publish the corrected information to prevent damage to the company's image.

[0807] Example 1

[0808] Next, a description will be given of Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0809] When companies release information on social media or their websites, they face the challenge of being unable to predict the risk of a public outcry and the economic impact. In particular, there is a need for methods to utilize past examples of public outcry and effective keywords to reduce the risk of public information and maximize the economic impact. Another important issue is preventing damage to a company's image in advance.

[0810] The identification process by the identification processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means. In this invention, the server includes a means for collecting and storing past flame data, a means for collecting and storing data on effective words, a means for analyzing this data using natural language processing technology, a means for predicting the economic effect of public information based on the analysis results, and a means for supporting the editing and creation of public information. This allows companies to reduce the risk of public information becoming a flame and maximize the economic effect. It also makes it possible to prevent damage to the company's image in advance.

[0811] "Past flame war data" refers to information about past flame war cases that occurred on social media or websites, and specifically includes the content of posts, comments, number of retweets, and user reactions.

[0812] "Effective word data" refers to information about keywords and phrases that elicit positive responses on social media and websites, and specifically includes keywords and phrases that have generated positive responses from users.

[0813] "Natural language processing technology" refers to technology that enables computers to understand, analyze, and generate human language, and specifically includes sentiment analysis and keyword extraction.

[0814] "Economic impact forecasting" refers to assessing how much traffic the public information will attract on social media or websites, or how much of a risk there is of it causing a backlash.

[0815] "Support for editing and creating public information" refers to improving the quality of information created by users by reducing the risk of online outrage and suggesting effective words.

[0816] "Preventing damage to a company's image" refers to measures taken to prevent a company's reputation from being damaged due to information that the company makes public.

[0817] MODE FOR CARRYING OUT THE INVENTION

[0818] This invention is an information disclosure support system that reduces the risk of flame wars and maximizes the economic impact when companies disclose information on social media or their websites. This system collects and stores data on past flame wars and effective words, and performs analysis based on this data using natural language processing technology. Furthermore, it predicts the economic impact of the disclosed information based on the analysis results and supports the editing and creation of the disclosed information.

[0819] Data collection and storage

[0820] The server collects past flame war data from social media and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags, and obtains data such as the number of comments and retweets related to the flame war. It also uses marketing research tools to collect data on effective keywords. For example, it uses Google Analytics or SEMrush to analyze how much traffic specific keywords attracted. The collected data is stored in a database such as MySQL (registered trademark) or PostgreSQL (registered trademark).

[0821] Analyzing the data

[0822] The server uses TensorFlow (registered trademark) and PyTorch (registered trademark) to analyze the accumulated flame war data. Specifically, it uses natural language processing (NLP) technology to analyze the sentiment of posts and identify the factors that cause flame wars. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive responses. For example, it uses models such as Word2Vec and BERT to evaluate the relevance of keywords.

[0823] Economic impact forecast

[0824] Based on the analysis results, the server predicts the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. This prediction is made by combining past data with AI models.

[0825] Editing and creating public information

[0826] The user creates public information and sends it from their device to the server. For example, they may enter a company's new product introduction or campaign announcement. The server assesses the risk of a controversy based on the information sent. Specifically, it analyzes the entered text using NLP technology and compares it with past controversy data to assess the risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a controversy, so please change it to this expression." The user then amends the information based on the server's suggestions and creates the final public information. The revised information can then be sent back to the server for final confirmation.

[0827] Specific examples

[0828] Example 1: Flaming risk assessment

[0829] A user creates a description of a new product and sends it from their device to the server. The server evaluates the risk of this description causing a backlash based on past data on backlash. For example, if the expression "This product is superior to other companies' products" is judged to increase the risk of backlash, the server will suggest changing the expression to something like "This product is well-received by many users."

[0830] Example 2: Effective word suggestions

[0831] The user creates a campaign announcement and sends it from their device to the server. The server then uses a database of effective words to suggest keywords that will elicit a positive response to the announcement. For example, if keywords such as "limited," "free," and "special offer" are deemed effective, the server will suggest adding these keywords to the announcement.

[0832] Prompt Sentence Examples

[0833] "We have created an introduction for a new product. Please assess the risk of this introduction causing outrage if it is published on social media, and make any necessary revisions."

[0834] "We've created a campaign announcement. Can you suggest some effective words to elicit a positive response to this announcement?"

[0835] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0836] Step 1: Collect and store data

[0837] The server collects past flame war data from social media sites and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags and obtains data such as comments and retweets related to the flame war. The input is the specific keywords or hashtags, and the output is the collected flame war data. The server stores the collected data in a database such as MySQL (registered trademark) or PostgreSQL (registered trademark).

[0838] Step 2: Effective Word Data Collection

[0839] The server uses marketing research tools to collect data on effective keywords. For example, it uses Google Analytics or SEMrush to analyze how much traffic a specific keyword attracted. The input is the specific keyword, and the output is data on effective keywords. The server stores the collected data in a database.

[0840] Step 3: Analyze the data

[0841] The server uses TensorFlow (registered trademark) and PyTorch (registered trademark) to analyze the accumulated flame data. Specifically, it uses natural language processing (NLP) technology to analyze the sentiment of the posts and identify the factors that cause the flames. The input is the accumulated flame data, and the output is the results of identifying the factors that cause the flames. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive reactions. The input is data on effective words, and the output is keywords and phrases that elicit positive reactions.

[0842] Step 4: Forecasting economic impact

[0843] The server uses the analysis results to predict the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. The input is the analysis results, and the output is the predicted economic impact.

[0844] Step 5: Create and submit your publication

[0845] A user creates public information and sends it to the server from their terminal. For example, they input a company's new product introduction or campaign announcement. The input is the public information created by the user, and the output is the public information sent to the server.

[0846] Step 6: Evaluate public information and make recommendations

[0847] The server evaluates the risk of a flame war based on the information sent. Specifically, it analyzes the input text using NLP technology and compares it with past flame war data to evaluate the risk. The input is public information created by the user, and the output is the evaluation result of the flame war risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a flame war, so please change it to this expression." The input is the evaluation result of the flame war risk, and the output is the suggested correction.

[0848] Step 7: Revise and finalize the public information

[0849] The user modifies the information based on the server's suggestions and creates the final published information. The modified information can be sent back to the server for final confirmation. The input is the server's suggested modifications, and the output is the modified published information.

[0850] (Application example 1)

[0851] Next, a description will be given of Application Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0852] When companies publish advertisements on social media or their websites, they are required to predict the economic impact by utilizing data on past controversies and effective words. However, there is no system that effectively utilizes this data to support the editing and creation of advertisement copy. Furthermore, there is a lack of methods to prevent damage to a company's image in advance by predicting controversy risk scores and effective word scores. This creates a problem in that companies are forced to operate while taking on risks when publishing advertisements.

[0853] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0854] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, a means including AI for predicting the economic effect of publishing on social media or a website based on this data, a means for supporting the correction and creation of advertising copy, and a means for predicting a flame risk score and an effective word score, thereby enabling companies to reduce the risk of publishing advertisements and create effective advertising copy.

[0855] "Past flame war data" refers to information about past flame war cases that have occurred on social media or websites.

[0856] "Effective word data" refers to information about words and phrases that are highly effective in advertising and information disclosure.

[0857] "AI that predicts economic impact" refers to artificial intelligence that predicts the economic impact of information published on social media or websites based on past data.

[0858] "Means to support the editing and creation of advertising copy" refers to a function that provides support for companies to revise advertising copy created by them based on past data or to create new copy.

[0859] The "flame risk score" is an index that quantifies the likelihood that a particular advertisement copy or piece of information will cause a flame war on social media or a website.

[0860] "Effective word score" refers to an indicator that quantifies the likelihood that a particular ad copy or piece of information will be highly effective.

[0861] "Measures to prevent damage to a company's image in advance" refers to a function that prevents a company's image from being damaged by information that the company makes public.

[0862] The system for implementing this invention includes AI that utilizes past data on flaming and effective words to predict the economic impact of advertising copy to be published on social media and websites. It also has the function of supporting the correction and creation of advertising copy, and predicting the flaming risk score and effective word score.

[0863] System configuration

[0864] Hardware:

[0865] server

[0866] Smartphone

[0867] software:

[0868] Python (registered trademark)

[0869] pandas

[0870] scikit-learn (registered trademark)

[0871] joblib

[0872] Data Handling

[0873] The server stores past flame data and effective word data in a database. These data are saved as CSV files and loaded using pandas. The loaded data is converted into a numeric vector using TfidfVectorizer.

[0874] Training and saving the model

[0875] The server trains a Logistic Regression model based on the data converted to a numeric vector using TfidfVectorizer. The trained model is saved using joblib. This lays the foundation for predicting flame risk scores and effective word scores.

[0876] Ad copy rating

[0877] When a user enters an ad copy using their smartphone, the server uses the stored model to predict the ad copy's flame war risk score and effective word score, allowing users to understand the risk and effectiveness of the ad copy in advance.

[0878] Specific examples

[0879] For example, if a user enters the following ad text:

[0880] To celebrate the launch of our new products, we're holding a special sale!

[0881] The server predicts the flame risk score and effective word score for this ad copy. The predicted results are a flame risk score of 0.2 and an effective word score of 0.8.

[0882] Prompt Sentence Examples

[0883] Predict the flame risk score and effective word score for the following ad copy.

[0884] Ad copy: "To celebrate the launch of our new product, we're offering a special sale!"

[0885] In this way, companies can reduce the risk of advertising and create effective advertising copy.

[0886] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0887] Step 1:

[0888] The server loads past flame data and effective word data from the database. These data are stored as CSV files and are loaded using pandas. The input is a CSV file and the output is a data frame. Specifically, the data is loaded using the pandas.read_csv function.

[0889] Step 2:

[0890] The server converts the read data into a numeric vector using TfidfVectorizer. The input is a data frame, and the output is a numeric vector. Specifically, the server converts text data into a numeric vector using the fit_transform method of TfidfVectorizer.

[0891] Step 3:

[0892] The server trains a LogisticRegression model based on the converted data. The input is a numeric vector, and the output is the trained model. Specifically, the model is trained using the fit method of LogisticRegression.

[0893] Step 4:

[0894] The server saves the trained model and TfidfVectorizer using joblib. The input is the trained model and TfidfVectorizer, and the output is the saved model file. Specifically, the joblib.dump function is used to save the model and vectorizer.

[0895] Step 5:

[0896] The user inputs the ad copy using a smartphone. The input is the text of the ad copy, and the output is a request to the server. Specifically, the ad copy is input through a smartphone application and sent to the server.

[0897] Step 6:

[0898] The server converts the received ad copy into a numeric vector using TfidfVectorizer. The input is the ad copy text, and the output is a numeric vector. Specifically, the server converts the ad copy into a numeric vector using the transform method of the saved TfidfVectorizer.

[0899] Step 7:

[0900] The server predicts the flame risk score and effective word score based on the ad copy converted to a numeric vector. The input is a numeric vector, and the output is the flame risk score and effective word score. Specifically, the server predicts the scores using the predict_proba method of the saved LogisticRegression model.

[0901] Step 8:

[0902] The server returns the predicted flame risk score and effective word score to the user. The input is the predicted score, and the output is a response to the user. Specifically, the score is displayed through a smartphone application.

[0903] Example 2

[0904] Next, a description will be given of Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0905] In today's information society, when companies or individuals release information on social media or their websites, they are required to use past data on online outrage to predict risks and take appropriate measures. However, existing systems lack the means to effectively utilize past data on online outrage and to specifically estimate the amount of loss, which makes it difficult to adequately manage risk. Furthermore, they lack concrete measures to prevent damage to a company's image in advance.

[0906] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0907] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, an artificial intelligence system for predicting the economic impact of information disclosure based on this data, a means for analyzing past flame data and estimating loss amounts, a means for storing the estimated loss amounts in a database, and a means for providing risk information including the estimated loss amounts when a user discloses information. This makes it possible to estimate specific loss amounts based on past flame data, quantify risk, and take appropriate measures when disclosing information. It also provides specific means for preventing damage to a company's image in advance.

[0908] "Past flame war data" refers to data on past flame war cases that occurred on online platforms such as social media and websites.

[0909] "Effective word data" refers to data on keywords and phrases that are effective when used when publishing information on social media or websites.

[0910] The "economic impact of information disclosure" refers to the predicted economic impact or benefits when information is disclosed on social media or a website.

[0911] "Artificial intelligence" is a system that uses machine learning and data analysis techniques to learn patterns from past data and make future predictions and decisions.

[0912] "Amount of loss" refers to the estimated amount of economic loss based on past cases of online outrage.

[0913] A "database" is a system for efficiently storing, managing, and retrieving data.

[0914] "Risk information" refers to information about risks that should be taken into consideration when disclosing information, and specifically includes estimated losses.

[0915] "Damage to corporate image" refers to a decline in a company's reputation and brand value.

[0916] This invention is a system for utilizing past flame war data to quantify the risk of information disclosure and take appropriate measures. Specific embodiments of this system are described below.

[0917] Server Roles

[0918] The server retrieves past online flame data from a database, analyzes it, and estimates the amount of loss. Specifically, the server uses the following hardware and software:

[0919] Hardware: A server machine with a powerful processor and sufficient memory

[0920] Software: Relational database management systems (RDBMS) such as MySQL (registered trademark) and PostgreSQL (registered trademark), data analysis libraries such as Python (registered trademark) pandas and scikit-learn (registered trademark)

[0921] The server first executes an SQL query to retrieve past flame data from the database. Next, it converts the retrieved data into a data frame using the Python (registered trademark) pandas library and performs preprocessing. Using the preprocessed data, it estimates the loss amount using the Linear Regression model of scikit-learn (registered trademark). The estimated loss amount is then saved back to the database.

[0922] Device Role

[0923] When a user discloses information, the terminal acquires risk information from the server and displays it to the user. Specifically, the terminal uses the following hardware and software:

[0924] Hardware: Personal computers, tablets, smartphones, etc.

[0925] Software: Web browser, HTML, CSS, JavaScript

[0926] The device sends an HTTP GET request to the server and receives risk information in JSON format, which is then displayed on a web page using HTML and JavaScript.

[0927] User Roles

[0928] Users can check the risk information through their devices and adjust the content of the information they disclose. Specifically, users check the estimated loss amount displayed on the web page, and if the risk is high, they can take measures such as changing the content of the information they disclose.

[0929] Specific examples

[0930] Example 1: Acquiring and analyzing flame data

[0931] The server retrieves past social media flame data from a MySQL® database, using, for example, the following SQL query:

[0932] sql

[0933] SELECT FROM flame_data WHERE date >= '2020-01-01';

[0934] The acquired data is analyzed using the Python (registered trademark) pandas library to estimate the loss amount in the event of a fire. For example, the following code is used:

[0935] Python (registered trademark)

[0936] import pandas as pd

[0937] from sklearn.linear_model import LinearRegression

[0938] Loading data

[0939] data = pd.read_sql('SELECT FROM flame_data WHERE date >= "2020-01-01"', con=database_connection)

[0940] Estimated loss amount

[0941] model = LinearRegression()

[0942] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[0943] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[0944] Example 2: Displaying risk information

[0945] When a user makes a disclosure, the device retrieves the risk information from the server and displays it in an HTML page, for example using code like this:

[0946] html

[0947] <!DOCTYPE html>

[0948]

[0949]

[0950] <title> Risk Information< / title>

[0951]

[0952]

[0953] <h1> Risk Information< / h1>

[0954] Estimated loss amount: \

[0955] <script>

[0956] fetch('https: / / example.com / api / risk_info')

[0957] .then(response => response.json())

[0958] .then(data => {

[0959] document.getElementById('loss_amount').textContent = data.loss_amount;

[0960] });

[0961] < / script>

[0962]

[0963]

[0964] Prompt Sentence Examples

[0965] Examples of prompts to input to a generative AI model might include:

[0966] "Generate Python (registered trademark) code that uses past social media flame war data to estimate the amount of losses incurred during a flame war."

[0967] Using this prompt sentence, the generative AI model can generate the appropriate Python code.

[0968] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0969] Step 1:

[0970] The server retrieves past flame data from the database. It receives database connection information and an SQL query as input, and generates a data frame containing the flame data as output. Specifically, the server executes the SQL query to extract the past flame data. For example, the following SQL query is used:

[0971] sql

[0972] SELECT FROM flame_data WHERE date >= '2020-01-01';

[0973] This query retrieves flame data for the specified time period.

[0974] Step 2:

[0975] The server analyzes the acquired data. It receives a data frame as input and generates a preprocessed data frame as output. Specifically, the server completes missing values ​​in the data and removes unnecessary columns. For example, it preprocesses the data using the Python (registered trademark) pandas library.

[0976] Step 3:

[0977] The server uses the preprocessed data to estimate the loss amount. It receives the preprocessed data frame as input and generates the estimated loss amount as output. Specifically, the server uses a LinearRegression model from scikit-learn (registered trademark) to estimate the loss amount. For example, the model is trained as follows to predict the loss amount:

[0978] Python (registered trademark)

[0979] model = LinearRegression()

[0980] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[0981] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[0982] This process allows the amount of loss to be estimated.

[0983] Step 4:

[0984] The server stores the estimated loss amount in a database. It receives the estimated loss amount and database connection information as input, and generates the loss amount stored in the database as output. Specifically, the server executes an SQL query to insert the estimated result into the database. For example, it uses the following SQL query:

[0985] sql

[0986] INSERT INTO risk_info (date, predicted_loss) VALUES ('2023-10-01', 500000);

[0987] This query results in the estimated loss amount being stored in a database.

[0988] Step 5:

[0989] When a user discloses information, the device obtains risk information from the server. It receives an HTTP GET request as input and generates JSON data containing the risk information as output. Specifically, the device sends an HTTP GET request to the server and receives the risk information. For example, the following JavaScript code is used:

[0990] javascript

[0991] fetch('https: / / example.com / api / risk_info')

[0992] .then(response => response.json())

[0993] .then(data => {

[0994] document.getElementById('loss_amount').textContent = data.predicted_loss;

[0995] });

[0996] Through this process, risk information is obtained.

[0997] Step 6:

[0998] The terminal displays the acquired risk information to the user. It receives JSON data containing risk information as input and generates an HTML page displaying the risk information as output. Specifically, the terminal uses HTML and JavaScript to display the risk information on a web page. For example, it can be implemented as follows:

[0999] html

[1000] <!DOCTYPE html>

[1001]

[1002]

[1003] <title> Risk Information< / title>

[1004]

[1005]

[1006] <h1> Risk Information< / h1>

[1007] Estimated loss amount: \

[1008] <script>

[1009] fetch('https: / / example.com / api / risk_info')

[1010] .then(response => response.json())

[1011] .then(data => {

[1012] document.getElementById('loss_amount').textContent = data.predicted_loss;

[1013] });

[1014] < / script>

[1015]

[1016]

[1017] This process displays risk information to the user.

[1018] Step 7:

[1019] The user checks the risk information through the terminal. As input, the user receives the displayed risk information, and as output, adjusts the content of the information disclosure. Specifically, the user checks the estimated loss amount displayed on the web page, and if the risk is high, takes action such as changing the disclosure content.

[1020] (Application example 2)

[1021] Next, a description will be given of Application Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1022] There is a need for a system that can quantify the risk of a controversy and predict the amount of loss in advertising campaigns using past controversy data. However, current systems have the problem of being unable to specifically quantify the risk of a controversy, disclose information based on that risk, or provide specific advice for reducing the risk.

[1023] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1024] In this invention, the server includes means for utilizing past flame data, means for utilizing data on effective words, means including a generative AI model for predicting the economic impact of publishing on social media or a website based on this data, means for quantifying the risk of a flame based on past flame data when the content of an advertising campaign is input, means for estimating the amount of loss based on the flame risk, and means for providing specific advice for reducing the risk. This makes it possible to specifically quantify the risk of a flame in an advertising campaign and provide information disclosure based on that risk and specific advice for reducing the risk.

[1025] "Past flame war data" refers to data such as negative reactions, comments, and criticisms that occurred on social media or websites in the past.

[1026] "Effective word data" refers to data on keywords and phrases that are considered effective in eliciting positive responses on social media and websites.

[1027] A "generative AI model" refers to an algorithm or system that uses machine learning or artificial intelligence techniques to generate a specific output from input data.

[1028] "Means for quantifying the risk of a backlash" refers to a method or system that uses data on past backlashes to numerically express the possibility that a particular advertising campaign will cause a backlash.

[1029] "Means for estimating loss amount" refers to a method or system for calculating expected economic loss based on the risk of a fire.

[1030] "Means for providing specific advice" refers to methods or systems that suggest specific actions or measures to reduce the risk of online outrage.

[1031] The system for carrying out the present invention operates in cooperation with three entities: a server, a terminal, and a user. A specific embodiment of the system will be described below.

[1032] Server Roles

[1033] The server stores data on past flame wars and effective keywords, and uses a generative AI model based on this data to predict economic impact. The server uses the following software and hardware:

[1034] Software: Python (registered trademark), Pandas, Scikit-learn (registered trademark)

[1035] Hardware: A server with a powerful processor and large memory capacity

[1036] The server loads past flame data and converts it into a data frame. Next, it sets features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model. This allows it to quantify the flame risk and predict loss amounts by inputting the details of an advertising campaign.

[1037] Device Role

[1038] The terminal provides an interface for users to input the content of their advertising campaigns. The terminal uses the following software and hardware:

[1039] Software: Web browser, mobile application

[1040] Hardware: Smartphones, tablets, computers

[1041] When a user enters details of an ad campaign through their device, the data is sent to a server, which then calculates the risk of a potential online outrage and predicts the amount of damage. The prediction is then sent back to the device and displayed to the user.

[1042] User Roles

[1043] Users can input details of their advertising campaigns using the terminal, check the predicted results of the risk of outrage and the amount of loss, and receive specific advice from the server on how to reduce the risk and take appropriate measures.

[1044] Specific examples

[1045] For example, if a user inputs a 30-day ad campaign length, 100,000 people as the audience size, and 50 negative comments, the server will use this data to quantify the risk of outrage and calculate the expected loss. As a result, the estimated loss amount is displayed and specific advice on how to reduce the risk is provided.

[1046] Prompt Sentence Examples

[1047] "If your ad campaign is 30 days long, your audience size is 100,000, and you have 50 negative comments, what is your estimated loss?"

[1048] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[1049] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1050] Step 1:

[1051] A user uses a terminal to enter details of an advertising campaign.

[1052] Inputs: length of ad campaign, audience size, number of negative comments

[1053] Output: The input data is sent to the server

[1054] What it does: Users enter details about their advertising campaign through a smartphone or computer interface and click the "Submit" button.

[1055] Step 2:

[1056] Based on the data received by the server, past flame data is read.

[1057] Input: Details of the ad campaign submitted by the user

[1058] Output: Historical fire data converted into a data frame

[1059] Specific operation: The server uses the Python (registered trademark) Pandas library to read past fire data from a CSV file and convert it into a data frame.

[1060] Step 3:

[1061] The server sets the features and target and trains a linear regression model.

[1062] Input: Historical fire data converted into a data frame

[1063] Output: A trained linear regression model

[1064] How it works: The server uses the Scikit-learn (registered trademark) library to set features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model.

[1065] Step 4:

[1066] Based on the detailed data of the advertising campaign received by the server from the user, the risk of a backlash is quantified and the amount of loss is predicted.

[1067] Input: Detailed advertising campaign data received from users, a trained linear regression model

[1068] Output: Fire risk figure and predicted loss amount

[1069] How it works: The server inputs the received detailed data of the advertising campaign as features, and uses a trained linear regression model to quantify the risk of a backlash and predict the amount of loss.

[1070] Step 5:

[1071] The server returns the predicted results of the risk of a firestorm and the amount of loss to the terminal.

[1072] Input: Flame risk value and predicted loss amount

[1073] Output: The prediction results are displayed on the terminal.

[1074] Specific operation: The server sends the prediction results in JSON format to the device, which receives them and displays them to the user.

[1075] Step 6:

[1076] The server generates specific advice to reduce the risk and sends it to the terminal.

[1077] Input: Flame risk value and predicted loss amount

[1078] Output: Specific advice for reducing risk

[1079] Specific operation: The server uses the generative AI model to generate specific advice for reducing the risk of flame wars and sends it to the device, which receives it and displays it to the user.

[1080] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[1081] Example 3

[1082] Next, a description will be given of Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1083] In today's information society, there is an increasing risk of a company becoming involved in a backlash when publishing information on social media or on its website. This risk can damage a company's image and cause financial losses, so companies are required to assess the risk of a backlash in advance and revise information as necessary. However, the current system does not adequately assess the risk of a backlash or revise information, which is a major issue for companies.

[1084] The specific processing by the specific processing unit 290 of the data processing device 12 in the third embodiment is realized by the following means.

[1085] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means for using a generative AI model that generates information based on these data, means for assessing the flame war risk before the generated information is made public, means for encouraging the correction of information if there is a flame war risk, and means for predicting the economic effect of publishing the information on social media or a website based on these means. This enables companies to prevent damage to their corporate image by assessing the flame war risk before publishing information and correcting the information as necessary.

[1086] "Past flame war data" refers to data on past flame wars that occurred as a result of information that was made public on social media, websites, etc.

[1087] "Effective word data" refers to data on words and phrases that are considered effective in eliciting positive responses when publishing information on social media or websites.

[1088] A "generative AI model" is an artificial intelligence model that generates information based on prompts entered by the user.

[1089] The "means for assessing the risk of a controversy" is a means for analyzing and assessing the possibility that the generated information will cause a controversy when it is made public.

[1090] "Means to encourage correction of information" refers to means to encourage users to correct information that is assessed as having a risk of causing a flame war.

[1091] "Means for predicting economic effects" refers to a means for predicting the economic impact that information published on social media or websites will have on a company.

[1092] MODE FOR CARRYING OUT THE INVENTION

[1093] This invention is a system that prevents damage to a company's image by assessing the risk of a controversy beforehand when a company publishes information on social media or a website, and correcting the information as necessary. This system consists of three main elements: a server, a terminal, and a user.

[1094] 1. Program Generation

[1095] The server generates a program to prevent damage to a company's image in advance. This program uses a generative AI model to create or edit information and evaluate the risk of that information causing a backlash before it is made public. Specifically, OpenAI's GPT-4 (registered trademark) is used as the generative AI model.

[1096] 2. Program Processing

[1097] The server uses a generative AI model (e.g., OpenAI's GPT-4 (registered trademark)) to generate information based on the prompt sentence entered by the user. The generated information is then sent to an inflammability risk assessment module. This module uses natural language processing (NLP) techniques to analyze the content of the information and assess whether it poses an inflammability risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information poses an inflammability risk.

[1098] The terminal sends the prompt sentence entered by the user to the server, and receives the generated information and the risk assessment result returned from the server. The user checks the risk assessment result on the terminal and corrects the information as necessary.

[1099] 3. Examples of concrete examples and prompts

[1100] As a concrete example, consider a user creating a press release for a new product. The user types the following prompt into the terminal:

[1101] Example prompt sentence:

[1102] "Write a press release for your new product. Call it 'EcoSmart' and highlight its environmentally friendly features."

[1103] The server inputs this prompt into a generative AI model and generates the following information:

[1104] Examples of generated information:

[1105] "EcoSmart is a new product that utilizes the latest environmental technology. It is energy efficient and made from recyclable materials."

[1106] The server then sends the generated information to the flame risk assessment module, which performs a risk assessment. If the assessment results in a "risk," the server sends a message to the user urging them to make corrections. The user can then check the suggested corrections on their device and make any necessary corrections to prevent damage to the company's image.

[1107] This system enables companies to assess the risk of information being exposed to controversy before it is released and correct the information as necessary, thereby preventing damage to the company's image in advance. The flow of the identification process in the third embodiment will be described with reference to FIG.

[1108] Step 1:

[1109] The user enters a prompt statement.

[1110] The user inputs a prompt into the input field of the terminal, for example, "Please write a press release for a new product. The product name should be 'EcoSmart' and the product's environmentally friendly features should be emphasized." The input prompt is temporarily stored in the terminal's memory.

[1111] Step 2:

[1112] The terminal sends the prompt to the server.

[1113] The terminal converts the prompt text entered by the user into an appropriate format and sends a request to the server's API endpoint. Specifically, it sends the prompt text to the server using an HTTP POST request. The input is the prompt text, and the output is the request sent to the server.

[1114] Step 3:

[1115] The server generates information using a generative AI model.

[1116] The server inputs the received prompt into a generative AI model (e.g., OpenAI's GPT-4 (registered trademark)) to generate information. The generative AI model analyzes the prompt and generates information based on the appropriate context. The input is the prompt, and the output is the generated information. For example, the information generated is, "EcoSmart is a new product that utilizes the latest environmental technology. It is highly energy efficient and uses recyclable materials."

[1117] Step 4:

[1118] The server sends the generated information to a flame risk assessment module.

[1119] The server sends the generated information to the flame risk assessment module for risk assessment. This module uses natural language processing (NLP) technology to analyze the content of the information and assess whether it poses a flame risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information contains a flame risk. The input is the generated information, and the output is the risk assessment result.

[1120] Step 5:

[1121] The server generates the risk assessment results and sends them to the terminal.

[1122] The server receives the evaluation results from the flame risk assessment module and sends them to the terminal. The evaluation results include judgments such as "no risk" or "risk present." The input is the risk assessment result, and the output is the transmission to the terminal.

[1123] Step 6:

[1124] The terminal displays the risk assessment results to the user.

[1125] The terminal displays the risk assessment results received from the server to the user. For example, a message such as "This information has the risk of causing a firestorm. We recommend that you modify it" is displayed. The input is the risk assessment result, and the output is what is displayed to the user.

[1126] Step 7:

[1127] The user modifies the information as needed.

[1128] The user checks the risk assessment results displayed on the device and modifies the information as necessary. The modified information is sent back to the server, where it undergoes the same process before finally being published as safe information. The input is the modified information, and the output is the re-evaluated information.

[1129] (Application example 3)

[1130] Next, a description will be given of Application Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1131] When companies release information on social media or their websites, they are required to assess the risk of a backlash in advance to prevent damage to their corporate image. However, conventional methods do not fully utilize data on past backlashes or effective keywords, and are insufficient in predicting economic effects. In addition, they lack the functionality to suggest specific corrections, making it difficult for companies to respond quickly and appropriately. This often results in companies taking unnecessary risks.

[1132] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 3 is realized by the following means.

[1133] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effects of disclosing information based on these data, means for assessing the flame war risk of information using a generative AI model, and means for presenting specific correction proposals if there is a flame war risk. This enables companies to assess the flame war risk before disclosing information and obtain specific correction proposals as necessary.

[1134] "Past flame war data" refers to information about past flame war cases that occurred on social media, websites, etc., and specifically includes data on the causes and effects of the flame wars, as well as related keywords.

[1135] "Data on effective words" refers to words and phrases used on social media, websites, etc. that are said to be particularly effective in eliciting positive responses.

[1136] "Artificial intelligence that predicts the economic impact of information disclosure" is an artificial intelligence technology used to predict the impact that disclosed information will have on a company's economic activity, and makes predictions based on past data and trends.

[1137] A "generative AI model" is an artificial intelligence model that generates text using natural language processing techniques and has the ability to generate sentences based on specific prompts.

[1138] "Means for assessing the risk of online outrage" refers to a means for assessing the possibility that publicly disclosed information will cause online outrage, and determines the risk using past online outrage data and generative AI models.

[1139] "Means for presenting specific revision proposals" refers to means for presenting specific revision proposals to reduce the risk of information being assessed as having a risk of causing outrage.

[1140] The system for carrying out the present invention operates in cooperation with three entities: a server, a terminal, and a user. A specific embodiment of the system will be described below.

[1141] System Configuration

[1142] The server includes means for utilizing data on past flame wars, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effects of disclosing information, means for assessing the risk of information becoming a flame war using a generative AI model, and means for presenting specific suggestions for correction if there is a risk of becoming a flame war.

[1143] A terminal is a device through which a user inputs information and communicates with a server. Specifically, this applies to smartphones and personal computers.

[1144] The users are corporate public relations personnel, marketing personnel, etc., and are responsible for inputting information and receiving feedback from the server.

[1145] Program processing

[1146] The server first receives information entered by the user, such as advertising copy and campaign details to be published on social media or on the website.

[1147] The server then evaluates the information's risk of causing an uproar based on past data on uproars and effective keywords. Specifically, it uses a generative AI model to determine whether the input information poses a risk of causing an uproar.

[1148] If the evaluation determines that there is a risk of outrage, the server generates specific revision suggestions, which use a generative AI model to suggest specific wording or phrases to reduce the risk.

[1149] Finally, the server sends the evaluation results and suggested modifications to the user's device, where the user can receive them and modify the information as needed.

[1150] Hardware and software used

[1151] Hardware: Servers, smartphones, PCs

[1152] Software: OpenAI API, Python (registered trademark)

[1153] Specific examples

[1154] For example, suppose a user enters the following ad copy:

[1155] This product is better than the competition. Buy it now!

[1156] The server receives this copy and generates the following prompt:

[1157] Evaluate the ad copy below and determine if it poses a risk of causing controversy.

[1158] Advertisement: This product is better than the competition. Buy it now!

[1159] If there is a risk of outrage, please provide specific suggestions for correction.

[1160] The generative AI model evaluates the prompt and suggests the following corrections:

[1161] This ad copy contains language that disparages other companies' products, and there is a risk of it causing outrage. We recommend that you revise it as follows:

[1162] Suggested fix: "This product has many satisfied customers. Buy it now and experience the difference!"

[1163] In this way, users can receive feedback from the server and correct their information, reducing the risk of backlash and preventing damage to the company's image.

[1164] The flow of the specific processing in Application Example 3 will be described with reference to FIG.

[1165] Step 1:

[1166] The user uses the terminal to input information such as advertising copy and campaign details.

[1167] Input: Ad copy and campaign details

[1168] Output: The entered information is sent to the server

[1169] How it works: The user enters the ad copy into the input form on their smartphone or computer and presses the submit button. The device then sends this information to the server.

[1170] Step 2:

[1171] The server compares the information it receives with past flame data and effective word data.

[1172] Input: User-submitted information, past flame data, effective word data

[1173] Output: Matching results (data necessary for assessing the risk of a firestorm)

[1174] Specific operation: The server retrieves past flame data and effective word data from the database and compares it with the information received from the user.

[1175] Step 3:

[1176] The server uses a generative AI model to assess the risk of information becoming viral.

[1177] Input: Matching results, generative AI model

[1178] Output: Flame risk assessment results

[1179] Specific operation: The server inputs the matching results into the generative AI model, generates a prompt sentence, and performs an evaluation. The generative AI model determines the risk of flame wars and returns the result to the server.

[1180] Step 4:

[1181] If the server is at risk of being burned, specific correction suggestions are generated.

[1182] Input: Flame risk assessment results

[1183] Output: Revision proposal

[1184] Specific operation: The server uses the generative AI model again to generate specific revision suggestions to reduce the risk of flame wars. The generative AI model generates revision suggestions based on the prompt sentence and returns the results to the server.

[1185] Step 5:

[1186] The server sends the evaluation results and suggested modifications to the user's device.

[1187] Input: Evaluation results, proposed corrections

[1188] Output: Evaluation results and suggested fixes sent to the user's device

[1189] Specific operation: The server compiles the evaluation results and revision suggestions and sends them to the user's device. The user can check this information on their device and revise the ad copy as necessary.

[1190] Step 6:

[1191] The user corrects the information based on the suggested corrections and sends it back to the server.

[1192] Input: Information corrected based on proposed amendment

[1193] Output: The modified information is sent to the server

[1194] Specific operation: The user modifies the ad copy based on the suggested revisions received from the server and sends it back to the server. The server receives the revised information and re-evaluates it.

[1195] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1196] "Example 1"

[1197] In one example of an invention incorporating an emotion engine, an emotion engine that recognizes user emotions is combined with a system that includes AI that utilizes data on past flame wars and effective words, and predicts the economic impact of publishing information on social media or websites based on this data. Specifically, the system analyzes emotions from text data posted by users, and corrects or creates public information based on the results. For example, if a user posts text expressing joy, that information is used to publish information that reinforces a positive image of a company's products or services.

[1198] "Example 2"

[1199] As a second example of the invention incorporating an emotion engine, in the system described in claim 2, an emotion engine that recognizes the user's emotions is combined with the means for determining the amount of loss when past flame data is used during editing. Specifically, the user's emotions are analyzed, and the risk of a flame war is assessed based on the results. For example, if a user posts text that expresses anger, that information is used to avoid publishing information that increases the risk of a flame war.

[1200] "Example 3"

[1201] As a third embodiment of the invention incorporating an emotion engine, in the system described in claim 3, an emotion engine that recognizes user emotions is combined with a means for preventing damage to the company's image in advance. Specifically, the system analyzes the user's emotions and, based on the results, publishes information to prevent damage to the company's image. For example, if a user posts text expressing sadness, that information is used to publish information about the company's response and service improvements.

[1202] The processing flow of each embodiment will be described below.

[1203] "Example 1"

[1204] Step 1: A user posts text data on a social networking site or website.

[1205] Step 2: The emotion engine analyzes the user's emotions from the text data.

[1206] Step 3: The AI ​​predicts the economic impact of the posted text data based on past flame war data and data on effective words.

[1207] Step 4: Based on the analysis results of the emotion engine and the prediction results of the AI, edit or create public information.

[1208] "Example 2"

[1209] Step 1: A user posts text data on a social networking site or website.

[1210] Step 2: The emotion engine analyzes the user's emotions from the text data.

[1211] Step 3: The AI ​​uses past flame war data to determine the amount of loss caused by the posted text data.

[1212] Step 4: Evaluate the risk of a social media outrage based on the results of the emotion engine analysis and the AI's understanding of the loss amount.

[1213] "Example 3"

[1214] Step 1: A user posts text data on a social networking site or website.

[1215] Step 2: The emotion engine analyzes the user's emotions from the text data.

[1216] Step 3: Based on past data on controversy and effective words, the AI ​​predicts the impact of the posted text data on the company's image.

[1217] Step 4: Based on the analysis results of the emotion engine and the prediction results of the AI, disclose information to prevent damage to the company's image.

[1218] Example 1

[1219] Next, a description will be given of Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1220] When companies release information on social media or websites, they are required to predict the economic impact by utilizing data on past scandals and effective keywords. However, there is a lack of systems to effectively utilize this data, making it difficult to prevent damage to corporate image and economic loss. Furthermore, there is a lack of methods to analyze user sentiment and edit or create information based on that sentiment, preventing effective corporate information disclosure strategies.

[1221] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1222] In this invention, the server includes means for collecting and storing data on past flame wars, means for collecting and storing data on effective words, means for using artificial intelligence to predict the economic impact of publishing on social media or websites based on this data, means for analyzing emotions from text data posted by users, and means for editing or creating published information based on the results of the emotion analysis.This enables companies to predict the economic impact by utilizing data on past flame wars and effective words, and to publish information based on user emotions.

[1223] "Past flame war data" refers to detailed information on past flame war cases that have occurred on social media or the internet.

[1224] "Effective word data" refers to information on words and phrases that are considered effective in marketing and advertising.

[1225] "SNS" is an abbreviation for social networking service, which refers to an online platform where users can share information and interact with each other.

[1226] A "website" is a collection of information made public on the Internet, and refers to an online page where companies and individuals disseminate information.

[1227] "Economic impact" refers to the impact that a particular action or event has on the economy, specifically including increases or decreases in sales and profits.

[1228] "Artificial intelligence" refers to technology that uses machine learning and data analysis to mimic human intelligence.

[1229] "User" refers to any individual or entity that uses the System or Services.

[1230] "Text data" refers to digital data that contains text information.

[1231] "Sentiment analysis" refers to the technology of identifying emotions from text data and evaluating the type and intensity of those emotions.

[1232] "Public information" refers to information that is made publicly available through social media or websites.

[1233] "Correction" refers to the act of revising existing text and changing it to more appropriate expression.

[1234] "Creating" refers to the act of generating new text or information.

[1235] MODE FOR CARRYING OUT THE INVENTION

[1236] This invention is an AI-based information disclosure support system for companies that uses data on past flame wars and effective keywords to predict the economic impact of information published on social media and websites. It also includes a function to analyze user sentiment and edit or create published information based on the results.

[1237] Data collection and storage

[1238] The server collects past flame war cases that have occurred on social media and the internet. Specifically, the data is obtained using a web scraping tool (e.g., Beautiful Soup). The collected data is stored in a database (e.g., MySQL (registered trademark), PostgreSQL (registered trademark)). Similarly, data on effective keywords is collected and stored in the same database.

[1239] Analyzing the data

[1240] The server analyzes the accumulated flame data using an AI model (e.g., TensorFlow (registered trademark), PyTorch (registered trademark)). Specifically, it uses natural language processing (NLP) technology to extract the causes and patterns of flame wars. Data on effective words is also analyzed using an AI model to identify which words are effective in what situations. These analysis results are stored in a database.

[1241] Economic impact forecast

[1242] The server predicts the economic impact of information published on social media and websites based on past flame war data and data on effective keywords. Specifically, it uses a machine learning model to simulate the impact the published information will have. The prediction results are stored in a database.

[1243] Combining Emotion Engines

[1244] The device sends the text data posted by the user to an emotion engine (e.g., IBM Watson or Microsoft Azure's emotion analysis API). The server receives the analysis results returned by the emotion engine and stores them in a database. Based on the results of the emotion analysis, the system edits or creates public information. Specifically, based on text that shows positive emotions, it generates information that reinforces a positive image of a company's products or services. The generated information is published on social media and websites.

[1245] Specific examples

[1246] For example, if a user posts the text "I really like the new product!", the device sends this text to the emotion engine. The server receives the emotion analysis result of "joy" from the emotion engine, and based on this analysis result generates positive information such as "Many users like the new product," which is then published on the SNS.

[1247] Prompt Sentence Examples

[1248] Examples of prompts to input to a generative AI model might include:

[1249] "Based on past data on flame wars and data on effective words, please predict the economic impact of information published on social media and websites. Also, please analyze user sentiment and correct the published information based on the results."

[1250] The above is an embodiment of the invention.

[1251] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1252] Step 1:

[1253] The server collects past incidents of online outrage that have occurred on social media and the internet. Specifically, it acquires the data using a web scraping tool (e.g., Beautiful Soup). The input is text data from social media and websites, and the output is the collected data on incidents of online outrage. This data is stored in a database (e.g., MySQL (registered trademark), PostgreSQL (registered trademark)).

[1254] Step 2:

[1255] The server collects data on words that are considered effective in marketing and advertising. This is done by utilizing existing marketing databases and expert knowledge. The input is word data from the marketing database, and the output is the collected data on effective words. This data is also stored in the same database.

[1256] Step 3:

[1257] The server analyzes the accumulated flame data using an AI model (e.g., TensorFlow (registered trademark), PyTorch (registered trademark)). Specifically, it uses natural language processing (NLP) technology to extract the causes and patterns of flames. The input is the flame data stored in the database, and the output is the analysis results of the causes and patterns of flames. These analysis results are stored in the database.

[1258] Step 4:

[1259] The server also uses an AI model to analyze the effective word data and determine which words are effective in what situations. The input is the effective word data stored in the database, and the output is the analysis results of the effective words. These analysis results are also stored in the database.

[1260] Step 5:

[1261] The server predicts the economic impact of information published on social media and websites based on past flame data and data on effective keywords. Specifically, it uses a machine learning model to simulate the extent of the reaction the published information will have. The input is the flame data and the analysis results of effective keywords, and the output is the predicted economic impact. The predicted results are stored in a database.

[1262] Step 6:

[1263] The device sends the text data posted by the user to an emotion engine (e.g., IBM Watson or Microsoft Azure's emotion analysis API). The input is the text data posted by the user, and the output is the emotion analysis result by the emotion engine.

[1264] Step 7:

[1265] The server receives the analysis results returned from the emotion engine and stores them in a database. The input is the emotion analysis results, and the output is the emotion analysis results stored in the database.

[1266] Step 8:

[1267] The server corrects and creates public information based on the results of the sentiment analysis. Specifically, it generates information that reinforces a positive image of a company's products and services based on text that indicates positive sentiment. The input is the sentiment analysis results, and the output is the generated public information. The generated information is published on social media and websites.

[1268] Step 9:

[1269] For example, if a user posts the text "I really like the new product!", the device sends this text to the emotion engine. The server receives the emotion analysis result of "joy" from the emotion engine, and based on this analysis result generates positive information such as "Many users like the new product," which is then published on the SNS.

[1270] (Application example 1)

[1271] Next, a description will be given of Application Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1272] When a company publishes information on social media or its website, it is important to predict the economic impact by utilizing data on past controversies and effective keywords. However, conventional systems lack the functionality to analyze user sentiment and generate optimal advertising copy based on that, making it difficult to improve the company's image and maximize the economic impact.

[1273] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means. In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including AI for predicting the economic effect of publishing on SNS or a website based on this data, means for analyzing user emotions, means for generating advertising copy based on the results of the emotion analysis, and means for predicting the economic effect of the generated advertising copy. This enables companies to generate optimal advertising copy based on user emotions and maximize the economic effect.

[1274] "Past flame war data" refers to data on negative reactions, critical comments, posts, etc. that have occurred in the past on social media or websites.

[1275] "Effective word data" refers to data on keywords and phrases that have generated positive responses and high economic impact on social media and websites.

[1276] "AI that predicts economic impact" is artificial intelligence that uses past data to predict the extent of economic impact that information published on social media or websites will have.

[1277] "Means for analyzing user emotions" refers to technology that analyzes emotions from text data posted by users and identifies emotional states such as positive, negative, and neutral.

[1278] "Means for generating advertising copy based on the results of sentiment analysis" refers to a technology that automatically generates optimal advertising copy using the results of user sentiment analysis.

[1279] The "means for predicting the economic effect of the generated advertising copy" is a technique for predicting the degree of economic impact that the generated advertising copy will have.

[1280] The following system configuration will be described as an embodiment of the present invention.

[1281] The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including AI for predicting the economic impact when such data is published on social media or a website, a means for analyzing user emotions, a means for generating advertising copy based on the results of the emotion analysis, and a means for predicting the economic impact of the generated advertising copy.

[1282] Hardware and software used

[1283] Hardware: Servers, smartphones

[1284] Software: Python (registered trademark), pandas, scikit-learn (registered trademark), TextBlob, OpenAI API

[1285] Data processing and calculation

[1286] 1. Data loading: The server loads past flame data and effective word data from a CSV file.

[1287] 2. Vectorization: The server vectorizes the text data using TfidfVectorizer.

[1288] 3. Model training: The server trains the economic impact prediction model using LinearRegression.

[1289] 4. Sentiment Analysis: The server uses TextBlob to analyze the sentiment of the user's posted text.

[1290] 5. Ad copy generation: The server uses the OpenAI API to generate ad copy based on sentiment.

[1291] 6. Economic impact prediction: The server uses the trained model to predict the economic impact of the generated ad copy.

[1292] Specific examples

[1293] When a user uses a smartphone to post the text "I want to create an advertisement for a new smartphone," the server performs the following process.

[1294] 1. The server performs sentiment analysis on the user's posted text using TextBlob to detect positive sentiment.

[1295] 2. The server uses the OpenAI API to generate the following prompt text and create an ad copy.

[1296] Example prompt: "Create a positive advertisement text for the following product: I want to create an advertisement for a new smartphone."

[1297] 3. The server predicts the economic effect based on the generated advertising copy.

[1298] In this way, companies can generate optimal advertising copy based on user emotions and maximize economic benefits.

[1299] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1300] Step 1:

[1301] The server reads past flame data and effective word data from a CSV file. As input, it receives the CSV file of past flame data and effective word data, and as output, it stores this data in memory in data frame format. Specifically, it uses the pandas library to read the CSV file and convert it to a data frame.

[1302] Step 2:

[1303] The server vectorizes the text data using TfidfVectorizer. It receives the text string of past flame data as input and obtains a vector representation of the text data as output. Specifically, it initializes TfidfVectorizer and vectorizes the text data using the fit_transform method.

[1304] Step 3:

[1305] The server trains an economic impact prediction model using LinearRegression. It receives vectorized text data and corresponding economic impact data as input and obtains a trained economic impact prediction model as output. Specifically, it initializes a LinearRegression model and trains the model using the fit method.

[1306] Step 4:

[1307] The server uses TextBlob to analyze the sentiment of user-posted text. It receives the text data posted by the user as input and obtains the sentiment analysis result (positive, negative, or neutral) as output. Specifically, it creates a TextBlob object and analyzes the sentiment using the sentiment property.

[1308] Step 5:

[1309] The server uses the OpenAI API to generate ad copy based on emotions. It receives the emotion analysis results and the user's posted text as input, and obtains the generated ad copy as output. Specifically, it generates a prompt based on the emotion analysis results and sends a request to the OpenAI API to generate the ad copy.

[1310] Step 6:

[1311] The server uses the trained model to predict the economic impact of the generated ad copy. It receives the generated ad copy as input and obtains the predicted economic impact value as output. Specifically, it vectorizes the generated ad copy and predicts the economic impact using the trained model's predict method.

[1312] Example 2

[1313] Next, a description will be given of Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1314] In today's information society, when companies or individuals publish information on social media or their websites, it is important to properly assess the risk of a controversy and prevent economic losses before they occur. However, conventional systems were unable to fully utilize past controversy data or user sentiment, resulting in insufficient assessment of the risk of a controversy. Furthermore, there was a lack of means to prevent damage to a company's image in advance. This made risk management difficult when disclosing information, and there was a high possibility of economic losses and damage to the company's image occurring.

[1315] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1316] In this invention, the server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic impact of information disclosure based on this data, a means for analyzing user input text in real time and recognizing emotions, and a means for assessing the risk of flame wars based on the results of the emotion analysis. This makes it possible to specifically quantify the risk of flame wars before information disclosure, thereby preventing economic losses and damage to the company's image.

[1317] "Past flame war data" refers to information about past flame war cases that occurred on social media, websites, etc., and specifically includes data such as the content of posts, comments, and related losses.

[1318] "Effective word data" refers to data on keywords and phrases that are believed to have a positive impact when information is released.

[1319] "Artificial intelligence that predicts the economic impact of information disclosure" refers to machine learning models and algorithms that use past data to predict the economic impact of information disclosure.

[1320] "Means for analyzing user-entered text in real time and recognizing emotions" refers to technology for instantly analyzing text entered by a user and identifying the emotions contained in that text.

[1321] "Means for assessing the risk of outrage based on the results of sentiment analysis" refers to technology that uses the results of sentiment analysis to assess the likelihood that input text will cause outrage.

[1322] "Means of determining the amount of loss" refers to technology for identifying and quantifying the economic losses incurred during a social media outburst based on data from past social media outbursts.

[1323] "Means to prevent damage to a company's image in advance" refers to technology that assesses the risk of a backlash before information is made public, preventing a company's reputation and image from being damaged.

[1324] MODE FOR CARRYING OUT THE INVENTION

[1325] This invention is a system that utilizes past flame data, analyzes user input text in real time to recognize emotions, and evaluates the risk of a flame war. Specific embodiments of this system are described below.

[1326] Server Processing

[1327] The server first collects data on past online scandals. It uses SQL queries from the database to retrieve data on past online scandals, including posts, comments, and associated losses. It then trains a model to predict losses based on the collected data. It preprocesses the data using Python's Pandas library and builds a regression model using Scikit-learn. The trained model is saved as a file using the Pickle library for later use.

[1328] Terminal handling

[1329] The device analyzes the text entered by the user in real time and calls the emotion analysis API to recognize emotions. Specifically, it receives the user's text input and analyzes the emotion of the text using the Google Cloud Natural Language API. The analysis results are displayed to the user on the device. For example, if the emotion "anger" is recognized, the user is notified.

[1330] User operations

[1331] Users input text into the system, which evaluates the risk of that text sparking a controversy. For example, they input the content of a social media post or blog article. The system evaluates the risk of the input text sparking a controversy based on past controversy data and the results of sentiment analysis. The evaluation results are displayed to the user, who can then modify the content of their post based on the results.

[1332] Specific examples

[1333] Example prompts to input to the generative AI model

[1334] "Design a system that uses past data on flaming incidents to assess the risk of specific text becoming flaming. Include functionality to analyze user-entered text in real time, recognize emotions using an emotion engine, and assess the risk of flaming based on the results."

[1335] This system allows users to quantify and confirm the risk of a controversy before posting, making it possible to disclose information while taking the risks into consideration.

[1336] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1337] System program processing flow

[1338] Server Processing

[1339] Step 1: Collect data on past flame wars

[1340] The server collects data on past incidents of online outrage from a database, specifically using SQL queries to retrieve data such as posts, comments, and associated losses.

[1341] Input: Database connection information, SQL query

[1342] Output: Past fire dataset

[1343] Specific operation: The server executes the SQL query "SELECT FROM EnjoData WHERE Date BETWEEN '2018-01-01' AND '2023-01-01';" to retrieve the required data from the database.

[1344] Step 2: Preprocessing the data

[1345] The server preprocesses the collected data, specifically by using the Pandas library in Python (registered trademark) to fill in missing values ​​and remove outliers.

[1346] Input: Historical fire dataset

[1347] Output: Preprocessed dataset

[1348] Specific behavior: The server uses Pandas to perform operations such as "df.fillna(0)" and "df.dropna()" to clean up the data.

[1349] Step 3: Training the loss prediction model

[1350] The server trains a model to predict loss amounts based on the preprocessed data. Specifically, it builds a regression model using Scikit-learn (registered trademark).

[1351] Input: Preprocessed dataset

[1352] Output: A trained loss prediction model

[1353] Specific operation: The server executes code such as "from sklearn.linear_model import LinearRegression" and "model = LinearRegression().fit(X_train, y_train)" to train the model.

[1354] Step 4: Save the model

[1355] The server saves the trained model as a file for later use. It uses the Pickle library to save the model.

[1356] Input: A trained loss prediction model

[1357] Output: Saved model file

[1358] Specific operation: The server executes the code "import pickle" and "with open('loss_prediction_model.pkl', 'wb') as file: pickle.dump(model, file)" to save the model.

[1359] Terminal handling

[1360] Step 5: Receiving User Input

[1361] The device receives text entered by the user, for example, from a text box in a web form or mobile app.

[1362] Input: User text input

[1363] Output: Received text data

[1364] Specific operation: The device says, "The user enters 'I can't use this product at all' into the web form."

[1365] Step 6: Call the sentiment analysis API

[1366] The device sends the input text to the sentiment analysis API to recognize emotions, and uses the Google Cloud Natural Language API to analyze the sentiment of the text.

[1367] Input: Received text data

[1368] Output: Emotion analysis results

[1369] Specific operation: The device executes the code "response = client.analyze_sentiment(document=document)" and "sentiment = response.document_sentiment" to analyze the sentiment.

[1370] Step 7: Viewing the analysis results

[1371] The device displays the emotion analysis results to the user. For example, if the emotion "anger" is recognized, the device notifies the user.

[1372] Input: Sentiment analysis results

[1373] Output: A notification message to the user

[1374] Specific behavior: The device "displays to the user 'This text contains angry sentiment.'"

[1375] User operations

[1376] Step 8: Entering Text

[1377] A user inputs text into the system, for example, the content of a social media post or blog post.

[1378] Input:Text content

[1379] Output: Send text data to the system

[1380] Specific operation: The user types "This service is completely unusable" and submits it to the system.

[1381] Step 9: Identify risk of outrage

[1382] The user checks the results of the flame risk assessment provided by the system, which assesses the risk based on past flame data and the results of sentiment analysis.

[1383] Input: Evaluation results from the system

[1384] Output: Display evaluation results to the user

[1385] Specific operation: The system displays the evaluation result, "This post has a high risk of causing a firestorm," and the user confirms it.

[1386] In this way, users can confirm the risk of a controversy in concrete terms and quantify it before posting, making it possible to disclose information while taking the risks into account.

[1387] (Application example 2)

[1388] Next, a description will be given of Application Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1389] In today's information society, when companies or individuals publish information on social media or their websites, they are required to use past data on flame wars to assess the risk of flame wars and prevent economic losses. However, conventional systems lack the means to analyze user emotions and assess the risk of flame wars, making it difficult to provide advance warnings of high-risk posts.

[1390] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1391] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic impact of posting on a social networking site or website based on this data, means for analyzing user emotions, means for assessing the risk of a flame war, and means for issuing advance warnings of high-risk posts. This makes it possible to assess the risk of a flame war taking user emotions into consideration and to issue advance warnings of high-risk posts.

[1392] "Past flame war data" refers to data on flame wars that have occurred in the past due to information that was made public on social media, websites, etc.

[1393] "Effective word data" refers to data on words and phrases that are likely to provoke a positive response when publishing information on social media or a website.

[1394] "Artificial intelligence" is the technology that enables computer systems to learn, reason, and self-correct by imitating human intelligence.

[1395] "Means for analyzing user emotions" refers to technology that identifies emotions from text and comments posted by users and evaluates the type and intensity of those emotions.

[1396] The "means for assessing the risk of a backlash" is a technology that quantifies and evaluates the possibility that the release of information will cause a backlash based on past backlash data and the results of user sentiment analysis.

[1397] "Means for providing advance warnings about high-risk posts" refers to technology that issues advance warnings to users about posts that are evaluated as having a high risk of causing a flame war.

[1398] The system for carrying out this invention consists of three main elements: a server, a terminal, and a user. The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic effects of posting on social media or a website based on this data, a means for analyzing user emotions, a means for assessing the risk of flame wars, and a means for issuing advance warnings about high-risk posts.

[1399] The server stores past flame data and effective word data in a database, and uses artificial intelligence to predict economic impact based on this data. Specifically, the server uses a text analysis engine (e.g., TextBlob) to analyze user posts and identify their sentiment. Furthermore, it uses a machine learning model (e.g., LinearRegression) to assess the risk of flame wars based on past flame war data and the results of sentiment analysis.

[1400] When a user posts to a social networking site or website, the device receives the results of a flame war risk assessment provided by the server and displays a warning for posts with a high risk, allowing the user to reconsider the content of their post and reduce the risk of a flame war.

[1401] For example, if a user tries to post "This product is completely unusable!", the device will send this text to the server. The server will then use a text analysis engine to analyze the sentiment and identify anger. It will then use a machine learning model to assess the risk of outrage and, if a high risk is detected, will display a warning on the device.

[1402] Example prompt sentence:

[1403] Enter your user post: This product is completely unusable!

[1404] This system allows users to check the risk of a controversy before posting and avoid high-risk posts, thereby preventing economic losses for companies and individuals.

[1405] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1406] Step 1:

[1407] The user uses the device to input text to post on a social networking site or website. The input text is sent from the device to the server. The input data is the content of the user's post.

[1408] Step 2:

[1409] The server passes the received text to a text analysis engine (e.g., TextBlob) for sentiment analysis. The text analysis engine identifies the emotion of the text and evaluates the type (e.g., anger, joy, sadness) and intensity of the emotion. The input data is the user's post, and the output data is the type and intensity of the emotion.

[1410] Step 3:

[1411] The server evaluates the risk of a controversy using a machine learning model (e.g., Linear Regression) based on the results of the emotion analysis. The machine learning model receives past controversy data and the emotion analysis results as input, and quantifies the risk of a controversy. The input data is the type and intensity of emotion, as well as past controversy data, and the output data is a numerical value for the risk of a controversy.

[1412] Step 4:

[1413] The server sends the results of the flame risk assessment to the terminal. The terminal displays a warning to the user if the risk is high based on the received flame risk value. The input data is the flame risk value, and the output data is a warning message.

[1414] Step 5:

[1415] The user checks the warning message displayed on the terminal and reconsiders the content of the post. If necessary, the user corrects the content of the post and sends it back to the server. The input data is the warning message, and the output data is the corrected content of the post.

[1416] This series of processes allows users to check for the risk of a controversy before posting and avoid high-risk posts, thereby preventing economic losses for companies and individuals.

[1417] Example 3

[1418] Next, a description will be given of Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1419] Companies are being asked to prevent the risk of information they disclose becoming a hot topic in advance and to prevent damage to their corporate image. However, with conventional systems, it is difficult to consistently generate and edit information, evaluate the risk of a hot topic, and analyze user sentiment, and effective countermeasures are often not possible. There is also a lack of means to predict the economic impact by utilizing data on past hot topics and effective words. This means that companies are taking on significant risks when disclosing information.

[1420] The identification process by the identification processing unit 290 of the data processing device 12 in Example 3 is realized by the following means. In this invention, the server includes means for utilizing past flame data, means for utilizing data on effective words, means for using a generative AI model for generating information based on these data, means for correcting the generated information, means for evaluating the flame risk of the generated or corrected information, means for prompting correction if the flame risk is high, means for using an emotion engine for analyzing user emotions, means for publishing information based on the analysis results, and means for predicting the economic effect of publishing the information on an SNS or website based on these data. This makes it possible to prevent a damage to the company's image by evaluating in advance the flame risk of information published by a company and prompting correction as necessary.

[1421] "Past flame war data" refers to data on past flame wars that occurred as a result of information that was made public on social media, websites, etc.

[1422] "Effective word data" refers to data on words and phrases that are considered effective in eliciting positive responses when information is released.

[1423] A "generative AI model" is an artificial intelligence model that generates natural language text based on an input prompt.

[1424] "Correction" is the process of checking the grammar and expression of the generated information for appropriateness and correcting it if necessary.

[1425] "Flame risk assessment" is the process of analyzing the possibility of a flame war occurring when generated or edited information is made public, and determining the risk.

[1426] "Methods to encourage corrections" are methods of notifying users that information that is deemed to pose a high risk of causing a backlash needs to be corrected.

[1427] An "emotion engine" is a technology that analyzes text posted by users and identifies the emotions (joy, sadness, anger, etc.) contained in that text.

[1428] "Information disclosure" is the process of publishing information based on the analysis results on social media, websites, etc. in order to prevent a company's image from being damaged.

[1429] "Economic impact forecasting" is the process of predicting the economic impact that publicly disclosed information will have on a company.

[1430] MODE FOR CARRYING OUT THE INVENTION

[1431] This invention is a system that prevents damage to a company's image by assessing in advance the risk of information that a company discloses becoming a target of criticism and prompting corrections as necessary. A specific embodiment of this system will be described below.

[1432] Hardware and software used

[1433] Hardware: Servers, terminals

[1434] Software: Generative AI model, emotion engine, database management system (DBMS)

[1435] System configuration

[1436] 1. The server stores past flame data in a database and has the means to utilize this data.

[1437] 2. The server stores data on effective words in a database and has the means to utilize this data.

[1438] 3. The server has a means to generate information using a generative AI model.

[1439] 4. The server has a means of correcting the generated information.

[1440] 5. The server has a means to assess the risk of flame wars for the information generated or edited.

[1441] 6. The server has the means to prompt corrections if there is a high risk of outrage.

[1442] 7. The server has a means to use an emotion engine to analyze the user's emotions.

[1443] 8. The server has a means to publish information based on the analysis results.

[1444] 9. The server has the means to predict the economic impact of publishing this data on social media or on a website.

[1445] System Operation

[1446] 1. The server uses the generative AI model to generate information that the company is planning to release. For example, the server inputs a prompt, "Please generate a description that explains the features of the new product," into the generative AI model to generate a description of the new product.

[1447] 2. The server uses AI to correct the generated information. Specifically, it checks grammar and expressions for appropriateness, and makes corrections as necessary.

[1448] 3. The server analyzes the corrected information and assesses the risk of a social media outrage. Natural language processing (NLP) technology is used to detect risk factors within the text.

[1449] 4. If the server is judged to be at high risk of being exposed to malicious content, a notification is sent to the device urging the user to make corrections. The user receives the notification and corrects the information as necessary.

[1450] 5. The server uses an emotion engine to analyze the user's emotions. For example, if a user posts "The price of the new product is too high," the server analyzes the text and detects the emotion of sadness.

[1451] 6. Based on the analysis results, the server publishes information to prevent damage to the company's image, such as explanations about prices and information about service improvements.

[1452] Specific examples

[1453] For example, consider the case where a company uses this system when announcing a new product.

[1454] 1. The server inputs the prompt "Please generate an introductory text that describes the features of the new product" into the generative AI model.

[1455] 2. The generative AI model generates a description such as, "This new product features high performance and an easy-to-use design."

[1456] 3. The server uses AI to check the grammar of the description it generates and corrects it to, "It features high performance and an easy-to-use design."

[1457] 4. The server analyzes the revised description using NLP technology and determines that the expression "high performance" may be implicitly criticizing other companies' products.

[1458] 5. The server sends a notification to the device prompting the correction, suggesting that the expression "high performance" be changed to "multi-functional."

[1459] 6. The user receives the notification and the message is revised to read, "It features a versatile and user-friendly design."

[1460] 7. The server uses an emotion engine to analyze the text posted by the user, "The price of the new product is too high," and detects the emotion of sadness.

[1461] 8. Based on the analysis results, the server publishes an "Explanation of the New Product's Price," explaining that "This new product uses high-quality materials, which is why it is priced higher."

[1462] In this way, the system functions to prevent damage to the company's image in advance. The flow of the identification process in the third embodiment will be described with reference to FIG.

[1463] Step 1:

[1464] The server inputs a prompt into the generative AI model to generate information. Specifically, the prompt is "Please generate an introduction that explains the features of our new product." Based on this prompt, the generative AI model generates an introduction such as "This new product is characterized by its high performance and easy-to-use design." The input is the prompt, and the output is the generated introduction.

[1465] Step 2:

[1466] The server uses AI to correct the generated information. Specifically, it checks the grammar of the generated introduction and confirms the appropriateness of the expression. For example, the sentence "It features high performance and an easy-to-use design" is corrected to "It features high performance and an easy-to-use design." The input is the generated introduction, and the output is the corrected introduction.

[1467] Step 3:

[1468] The server analyzes the corrected information and evaluates the risk of a controversy. Specifically, it uses natural language processing (NLP) technology to detect risk factors within the text. For example, it determines that the expression "high performance" may be an implicit criticism of another company's product. The input is the corrected introduction, and the output is the evaluation result of the risk of a controversy.

[1469] Step 4:

[1470] If the server determines that there is a high risk of a flame war, it sends a notification to the terminal urging the user to make corrections. Specifically, it notifies the user of the parts that need to be corrected and the reasons for the corrections. For example, it suggests changing the expression "high performance" to "multi-functional." The input is the evaluation result of the flame war risk, and the output is a notification urging the user to make the corrections.

[1471] Step 5:

[1472] The user receives the notification and corrects the information as necessary. Specifically, the description is corrected based on the notification. For example, the sentence "It features high performance and an easy-to-use design" is corrected to "It features multifunctionality and an easy-to-use design." The input is the notification prompting the correction, and the output is the corrected description.

[1473] Step 6:

[1474] The server uses an emotion engine to analyze the user's emotions. Specifically, it analyzes the text posted by the user and identifies the type of emotion (joy, sadness, anger, etc.). For example, if a user posts "The price of the new product is too high," the text is analyzed to detect the emotion of sadness. The input is the text posted by the user, and the output is the result of the emotion analysis.

[1475] Step 7:

[1476] Based on the analysis results, the server publishes information to prevent damage to the company's image. Specifically, it uses the analysis results to provide explanations about service improvements and prices. For example, it publishes an "explanation about the price of a new product" and explains that "this new product uses high-quality materials, and therefore the price is set higher." The input is the result of the sentiment analysis, and the output is the information to be published.

[1477] (Application example 3)

[1478] Next, a description will be given of Application Example 3 of Form Example 3. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1479] In today's information society, there is an increasing risk that information released by a company will cause a backlash, which could damage the company's image and lead to financial losses. Furthermore, releasing information without regard for user feelings can worsen the relationship between the company and its users. To solve these problems, it is necessary to evaluate the risk of a backlash before releasing information and optimize the information while taking user feelings into consideration.

[1480] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 3 is realized by the following means. In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means including artificial intelligence for predicting the economic effect of information disclosure based on this data, means for analyzing user emotions, and means for optimizing information based on the analysis results. This makes it possible to evaluate the risk of flame wars before information disclosure and to optimize information taking user emotions into consideration.

[1481] "Past online outrage data" refers to data on cases of online outrage caused by information that was made public in the past.

[1482] "Effective word data" is data on words and phrases that are expected to elicit a positive response when used when disclosing information.

[1483] "Artificial intelligence that predicts the economic impact of disclosure" is artificial intelligence that calculates the expected economic impact if information is made public.

[1484] "Means for analyzing user emotions" refers to technology for analyzing the emotions expressed by a user from text, voice, etc.

[1485] "Means for optimizing information" refers to technology for correcting and adjusting the information to be made public in an optimal form based on analysis results and predictive data.

[1486] The system for carrying out this invention comprises a server and a user terminal. The server includes a means for utilizing data on past flame wars, a means for utilizing data on effective words, a means including artificial intelligence for predicting the economic effects of disclosure, a means for analyzing user sentiment, and a means for optimizing information based on the analysis results.

[1487] Hardware and software used

[1488] Hardware:

[1489] Server (cloud-based)

[1490] User device (smartphone, PC)

[1491] software:

[1492] Sentiment analysis engines (e.g., IBM Watson, Google Cloud Natural Language API)

[1493] Flaming risk assessment AI model (e.g., OpenAI GPT-4 (registered trademark))

[1494] Ad generation AI model (e.g., OpenAI GPT-4 (registered trademark))

[1495] Data processing and calculation

[1496] 1. Ad generation:

[1497] The user device sends basic information about the advertising campaign (target audience, product information, etc.) to the server.

[1498] The server uses the ad generation AI model to generate ad copy.

[1499] 2. Flaming Risk Assessment:

[1500] The server inputs the generated advertising copy into an AI model for assessing the risk of a controversy and performs a risk assessment.

[1501] If the risk is high, the server will suggest a correction to the user terminal.

[1502] 3. Emotion analysis:

[1503] The server collects users' emotional data (e.g., social media posts, feedback, etc.).

[1504] A sentiment analysis engine is used to analyze the user's sentiment.

[1505] Based on the analysis results, the server optimizes the advertising content.

[1506] Specific examples

[1507] For example, if you are running an advertising campaign for a new smartphone, you can send the following prompt message from the user's device to the server:

[1508] Prompt Sentence Examples

[1509] Target Audience: Young people (18-25 years old)

[1510] Product Information: New smartphone "XPhone" with high-performance camera, long battery life, and latest OS

[1511] Tone of the ad copy: Casual and friendly

[1512] Based on this prompt, the server uses a generative AI model to generate ad copy, and an AI model for assessing the risk of flame wars evaluates the risk. If the risk is high, it suggests modifications, and finally, a sentiment analysis engine optimizes the ad content based on user sentiment.

[1513] The above is an embodiment of the present invention.

[1514] The flow of the specific processing in Application Example 3 will be described with reference to FIG.

[1515] Step 1:

[1516] The user inputs basic information about the advertising campaign (target audience, product information, tone of the advertising copy, etc.) into the user terminal, and the input information is sent to the server.

[1517] Step 2:

[1518] The server generates ad copy using an ad generation AI model based on the received basic information. The generated ad copy is temporarily stored on the server.

[1519] Step 3:

[1520] The server inputs the generated ad copy into an AI model for assessing the risk of a viral outbreak and performs a risk assessment. The results of the risk assessment are output, and if the risk is high, the server generates a revision proposal.

[1521] Step 4:

[1522] The server sends the risk assessment results and proposed revisions to the user terminal, and the user checks the proposed revisions and revises the advertisement copy as necessary.

[1523] Step 5:

[1524] The server collects users' emotional data (e.g., social media posts, feedback, etc.) and inputs the collected emotional data into a sentiment analysis engine.

[1525] Step 6:

[1526] The server uses an emotion analysis engine to analyze the user's emotions, and the analysis results are output and stored in the server.

[1527] Step 7:

[1528] The server optimizes the ad content based on the analysis results, and the optimized ad copy is stored on the server.

[1529] Step 8:

[1530] The server sends the optimized ad copy to the user terminal, and the user checks the optimized ad copy and publishes it as the final ad copy.

[1531] The above is the processing flow of this program.

[1532] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1533] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ) and other generative AIs. Data generation models 58

[1534] , obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in the data format of voice data, text data, etc. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1535] Another example of generative AI is Gemini (internet search engine). <url: https: gemini.google.com ?hl="ja">) are mentioned.

[1536] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[1537] [Third embodiment]

[1538] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[1539] 5, the data processing system 310 includes the data processing device 12 and a headset terminal 314. An example of the data processing device 12 is a server.

[1540] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1541] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[1542] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1543] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1544] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1545] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1546] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1547] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1548] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1549] Next, the specific processing by the specific processing unit 290 of the data processing device 12 will be described.

[1550] "Example 1"

[1551] One embodiment of the present invention is an information disclosure support system for companies that utilizes AI. This system utilizes data on past flame wars and data on effective words, and predicts the economic impact of publishing information on social media or websites based on this data. Specifically, past flame war data is stored in a database and analyzed by AI. Data on effective words is also stored in the same way, and the published information is edited and created based on this data.

[1552] "Example 2"

[1553] Furthermore, another embodiment of the present invention has a function to grasp the amount of loss when using past flame data during correction. Specifically, the amount of loss in the event of a flame is estimated from past flame data and used as reference information for correction. This makes it possible to specifically quantify the risk of a flame and disclose information taking that risk into account.

[1554] "Example 3"

[1555] Furthermore, a further embodiment of the present invention has a function to prevent damage to a company's image in advance. Specifically, before information created or corrected by AI is made public, it evaluates whether the information poses a risk of causing a backlash and prompts corrections as necessary. This makes it possible to prevent damage to a company's image in advance.

[1556] The processing flow of each embodiment will be described below.

[1557] "Example 1"

[1558] Step 1: Accumulate data on past controversies and effective words in a database.

[1559] Step 2: AI analyzes this data and edits or creates public information.

[1560] Step 3: Determine the optimal timing and content of information disclosure based on the economic impact predicted by AI

[1561] do.

[1562] "Example 2"

[1563] Step 1: Estimate the amount of loss in the event of a fire based on past fire data.

[1564] Step 2: Use the estimated loss amount as a reference for corrections.

[1565] Step 3: Quantify the risk of a backlash and disclose information based on that risk.

[1566] "Example 3"

[1567] Step 1: Before information created or edited by AI is made public, assess whether it poses a risk of causing outrage.

[1568] Step 2: Based on the evaluation results, encourage correction of information as necessary.

[1569] Step 3: Publish the corrected information to prevent damage to the company's image.

[1570] Example 1

[1571] Next, a description will be given of Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1572] When companies release information on social media or their websites, they face the challenge of being unable to predict the risk of a public outcry and the economic impact. In particular, there is a need for methods to utilize past examples of public outcry and effective keywords to reduce the risk of public information and maximize the economic impact. Another important issue is preventing damage to a company's image in advance.

[1573] The identification process by the identification processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means. In this invention, the server includes a means for collecting and storing past flame data, a means for collecting and storing data on effective words, a means for analyzing this data using natural language processing technology, a means for predicting the economic effect of public information based on the analysis results, and a means for supporting the editing and creation of public information. This allows companies to reduce the risk of public information becoming a flame and maximize the economic effect. It also makes it possible to prevent damage to the company's image in advance.

[1574] "Past flame war data" refers to information about past flame war cases that occurred on social media or websites, and specifically includes the content of posts, comments, number of retweets, and user reactions.

[1575] "Effective word data" refers to information about keywords and phrases that elicit positive responses on social media and websites, and specifically includes keywords and phrases that have generated positive responses from users.

[1576] "Natural language processing technology" refers to technology that enables computers to understand, analyze, and generate human language, and specifically includes sentiment analysis and keyword extraction.

[1577] "Economic impact forecasting" refers to assessing how much traffic the public information will attract on social media or websites, or how much of a risk there is of it causing a backlash.

[1578] "Support for editing and creating public information" refers to improving the quality of information created by users by reducing the risk of online outrage and suggesting effective words.

[1579] "Preventing damage to a company's image" refers to measures taken to prevent a company's reputation from being damaged due to information that the company makes public.

[1580] MODE FOR CARRYING OUT THE INVENTION

[1581] This invention is an information disclosure support system that reduces the risk of flame wars and maximizes the economic effect when companies disclose information on social media or their websites. This system collects and stores data on past flame wars and effective words, and performs analysis based on this data using natural language processing technology. Furthermore, it predicts the economic effect of the disclosed information based on the analysis results and supports the editing and creation of the disclosed information.

[1582] Data collection and storage

[1583] The server collects past flame war data from social media and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags, and obtains data such as the number of comments and retweets related to the flame war. It also uses marketing research tools to collect data on effective keywords. For example, it uses Google Analytics or SEMrush to analyze how much traffic specific keywords attracted. The collected data is stored in a database such as MySQL (registered trademark) or PostgreSQL (registered trademark).

[1584] Analyzing the data

[1585] The server uses TensorFlow (registered trademark) and PyTorch (registered trademark) to analyze the accumulated flame war data. Specifically, it uses natural language processing (NLP) technology to analyze the sentiment of posts and identify the factors that cause flame wars. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive responses. For example, it uses models such as Word2Vec and BERT to evaluate the relevance of keywords.

[1586] Economic impact forecast

[1587] Based on the analysis results, the server predicts the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. This prediction is made by combining past data with AI models.

[1588] Editing and creating public information

[1589] The user creates public information and sends it from their device to the server. For example, they may enter a company's new product introduction or campaign announcement. The server assesses the risk of a controversy based on the information sent. Specifically, it analyzes the entered text using NLP technology and compares it with past controversy data to assess the risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a controversy, so please change it to this expression." The user then amends the information based on the server's suggestions and creates the final public information. The revised information can then be sent back to the server for final confirmation.

[1590] Specific examples

[1591] Example 1: Flaming risk assessment

[1592] A user creates a description of a new product and sends it from their device to the server. The server evaluates the risk of this description causing a backlash based on past data on backlash. For example, if the expression "This product is superior to other companies' products" is judged to increase the risk of backlash, the server will suggest changing the expression to something like "This product is well-received by many users."

[1593] Example 2: Effective word suggestions

[1594] The user creates a campaign announcement and sends it from their device to the server. The server then uses a database of effective words to suggest keywords that will elicit a positive response to the announcement. For example, if keywords such as "limited," "free," and "special offer" are deemed effective, the server will suggest adding these keywords to the announcement.

[1595] Prompt Sentence Examples

[1596] "We have created an introduction for a new product. Please assess the risk of this introduction causing outrage if it is published on social media, and make any necessary revisions."

[1597] "We've created a campaign announcement. Can you suggest some effective words to elicit a positive response to this announcement?"

[1598] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1599] Step 1: Collect and store data

[1600] The server collects past flame war data from social media sites and websites. Specifically, it uses social media APIs to search for posts containing specific keywords or hashtags and obtains data such as comments and retweets related to the flame war. The input is the specific keywords or hashtags, and the output is the collected flame war data. The server stores the collected data in a database such as MySQL (registered trademark) or PostgreSQL (registered trademark).

[1601] Step 2: Effective Word Data Collection

[1602] The server uses marketing research tools to collect data on effective keywords. For example, it uses Google Analytics or SEMrush to analyze how much traffic a specific keyword attracted. The input is the specific keyword, and the output is data on effective keywords. The server stores the collected data in a database.

[1603] Step 3: Analyze the data

[1604] The server uses TensorFlow (registered trademark) and PyTorch (registered trademark) to analyze the accumulated flame data. Specifically, it uses natural language processing (NLP) technology to analyze the sentiment of the posts and identify the factors that cause the flames. The input is the accumulated flame data, and the output is the results of identifying the factors that cause the flames. Similarly, it analyzes data on effective words to identify keywords and phrases that elicit positive reactions. The input is data on effective words, and the output is keywords and phrases that elicit positive reactions.

[1605] Step 4: Forecasting economic impact

[1606] The server uses the analysis results to predict the economic impact of the information to be released. Specifically, it evaluates how much traffic the information to be released will attract and the risk of it causing a backlash. The input is the analysis results, and the output is the predicted economic impact.

[1607] Step 5: Create and submit your publication

[1608] A user creates public information and sends it to the server from their terminal. For example, they input a company's new product introduction or campaign announcement. The input is the public information created by the user, and the output is the public information sent to the server.

[1609] Step 6: Evaluate public information and make recommendations

[1610] The server evaluates the risk of a flame war based on the information sent. Specifically, it analyzes the input text using NLP technology and compares it with past flame war data to evaluate the risk. The input is public information created by the user, and the output is the evaluation result of the flame war risk. The server suggests effective words and corrects the information. For example, it makes specific suggestions such as, "This expression has a high risk of causing a flame war, so please change it to this expression." The input is the evaluation result of the flame war risk, and the output is the suggested correction.

[1611] Step 7: Revise and finalize the public information

[1612] The user modifies the information based on the server's suggestions and creates the final published information. The modified information can be sent back to the server for final confirmation. The input is the server's suggested modifications, and the output is the modified published information.

[1613] (Application example 1)

[1614] Next, a description will be given of Application Example 1 of Form Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1615] When companies publish advertisements on social media or their websites, they are required to predict the economic impact by utilizing data on past controversies and effective words. However, there is no system that effectively utilizes this data to support the editing and creation of advertisement copy. Furthermore, there is a lack of methods to prevent damage to a company's image in advance by predicting controversy risk scores and effective word scores. This creates a problem in that companies are forced to operate while taking on risks when publishing advertisements.

[1616] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1617] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, a means including AI for predicting the economic effect of publishing on social media or a website based on this data, a means for supporting the correction and creation of advertising copy, and a means for predicting a flame risk score and an effective word score, thereby enabling companies to reduce the risk of publishing advertisements and create effective advertising copy.

[1618] "Past flame war data" refers to information about past flame war cases that have occurred on social media or websites.

[1619] "Effective word data" refers to information about words and phrases that are highly effective in advertising and information disclosure.

[1620] "AI that predicts economic impact" refers to artificial intelligence that predicts the economic impact of information published on social media or websites based on past data.

[1621] "Means to support the editing and creation of advertising copy" refers to a function that provides support for companies to revise advertising copy created by them based on past data or to create new copy.

[1622] The "flame risk score" is an index that quantifies the likelihood that a particular advertisement copy or piece of information will cause a flame war on social media or a website.

[1623] "Effective word score" refers to an indicator that quantifies the likelihood that a particular ad copy or piece of information will be highly effective.

[1624] "Measures to prevent damage to a company's image in advance" refers to a function that prevents a company's image from being damaged by information that the company makes public.

[1625] The system for implementing this invention includes AI that utilizes past data on flaming and effective words to predict the economic impact of advertising copy to be published on social media and websites. It also has the function of supporting the correction and creation of advertising copy, and predicting the flaming risk score and effective word score.

[1626] System configuration

[1627] Hardware:

[1628] server

[1629] Smartphone

[1630] software:

[1631] Python (registered trademark)

[1632] pandas

[1633] scikit-learn (registered trademark)

[1634] joblib

[1635] Data Handling

[1636] The server stores past flame data and effective word data in a database. These data are saved as CSV files and loaded using pandas. The loaded data is converted into a numeric vector using TfidfVectorizer.

[1637] Training and saving the model

[1638] The server trains a Logistic Regression model based on the data converted to a numeric vector using TfidfVectorizer. The trained model is saved using joblib. This lays the foundation for predicting flame risk scores and effective word scores.

[1639] Ad copy rating

[1640] When a user enters an ad copy using their smartphone, the server uses the stored model to predict the ad copy's flame war risk score and effective word score, allowing users to understand the risk and effectiveness of the ad copy in advance.

[1641] Specific examples

[1642] For example, if a user enters the following ad text:

[1643] To celebrate the launch of our new products, we're holding a special sale!

[1644] The server predicts the flame risk score and effective word score for this ad copy. The predicted results are a flame risk score of 0.2 and an effective word score of 0.8.

[1645] Prompt Sentence Examples

[1646] Predict the flame risk score and effective word score for the following ad copy.

[1647] Ad copy: "To celebrate the launch of our new product, we're offering a special sale!"

[1648] In this way, companies can reduce the risk of advertising and create effective advertising copy.

[1649] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1650] Step 1:

[1651] The server loads past flame data and effective word data from the database. These data are stored as CSV files and are loaded using pandas. The input is a CSV file and the output is a data frame. Specifically, the data is loaded using the pandas.read_csv function.

[1652] Step 2:

[1653] The server converts the read data into a numeric vector using TfidfVectorizer. The input is a data frame, and the output is a numeric vector. Specifically, the server converts text data into a numeric vector using the fit_transform method of TfidfVectorizer.

[1654] Step 3:

[1655] The server trains a LogisticRegression model based on the converted data. The input is a numeric vector, and the output is the trained model. Specifically, the model is trained using the fit method of LogisticRegression.

[1656] Step 4:

[1657] The server saves the trained model and TfidfVectorizer using joblib. The input is the trained model and TfidfVectorizer, and the output is the saved model file. Specifically, the joblib.dump function is used to save the model and vectorizer.

[1658] Step 5:

[1659] The user inputs the ad copy using a smartphone. The input is the text of the ad copy, and the output is a request to the server. Specifically, the ad copy is input through a smartphone application and sent to the server.

[1660] Step 6:

[1661] The server converts the received ad copy into a numeric vector using TfidfVectorizer. The input is the ad copy text, and the output is a numeric vector. Specifically, the server converts the ad copy into a numeric vector using the transform method of the saved TfidfVectorizer.

[1662] Step 7:

[1663] The server predicts the flame risk score and effective word score based on the ad copy converted to a numeric vector. The input is a numeric vector, and the output is the flame risk score and effective word score. Specifically, the server predicts the scores using the predict_proba method of the saved LogisticRegression model.

[1664] Step 8:

[1665] The server returns the predicted flame risk score and effective word score to the user. The input is the predicted score, and the output is a response to the user. Specifically, the score is displayed through a smartphone application.

[1666] Example 2

[1667] Next, a description will be given of Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1668] In today's information society, when companies or individuals release information on social media or their websites, they are required to use past data on online outrage to predict risks and take appropriate measures. However, existing systems lack the means to effectively utilize past data on online outrage and to specifically estimate the amount of loss, which makes it difficult to adequately manage risk. Furthermore, they lack concrete measures to prevent damage to a company's image in advance.

[1669] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1670] In this invention, the server includes a means for utilizing past flame data, a means for utilizing data on effective words, an artificial intelligence system for predicting the economic impact of information disclosure based on this data, a means for analyzing past flame data and estimating loss amounts, a means for storing the estimated loss amounts in a database, and a means for providing risk information including the estimated loss amounts when a user discloses information. This makes it possible to estimate specific loss amounts based on past flame data, quantify risk, and take appropriate measures when disclosing information. It also provides specific means for preventing damage to a company's image in advance.

[1671] "Past flame war data" refers to data on past flame war cases that occurred on online platforms such as social media and websites.

[1672] "Effective word data" refers to data on keywords and phrases that are effective when used when publishing information on social media or websites.

[1673] The "economic impact of information disclosure" refers to the predicted economic impact or benefits when information is disclosed on social media or a website.

[1674] "Artificial intelligence" is a system that uses machine learning and data analysis techniques to learn patterns from past data and make future predictions and decisions.

[1675] "Amount of loss" refers to the estimated amount of economic loss based on past cases of online outrage.

[1676] A "database" is a system for efficiently storing, managing, and retrieving data.

[1677] "Risk information" refers to information about risks that should be taken into consideration when disclosing information, and specifically includes estimated losses.

[1678] "Damage to corporate image" refers to a decline in a company's reputation and brand value.

[1679] This invention is a system for utilizing past flame war data to quantify the risk of information disclosure and take appropriate measures. Specific embodiments of this system are described below.

[1680] Server Roles

[1681] The server retrieves past online flame data from a database, analyzes it, and estimates the amount of loss. Specifically, the server uses the following hardware and software:

[1682] Hardware: A server machine with a powerful processor and sufficient memory

[1683] Software: Relational database management systems (RDBMS) such as MySQL (registered trademark) and PostgreSQL (registered trademark), data analysis libraries such as Python (registered trademark) pandas and scikit-learn (registered trademark)

[1684] The server first executes an SQL query to retrieve past flame data from the database. Next, it converts the retrieved data into a data frame using the Python (registered trademark) pandas library and performs preprocessing. Using the preprocessed data, it estimates the loss amount using the Linear Regression model of scikit-learn (registered trademark). The estimated loss amount is then saved back to the database.

[1685] Device Role

[1686] When a user discloses information, the terminal acquires risk information from the server and displays it to the user. Specifically, the terminal uses the following hardware and software:

[1687] Hardware: Personal computers, tablets, smartphones, etc.

[1688] Software: Web browser, HTML, CSS, JavaScript

[1689] The device sends an HTTP GET request to the server and receives risk information in JSON format, which is then displayed on a web page using HTML and JavaScript.

[1690] User Roles

[1691] Users can check the risk information through their devices and adjust the content of the information they disclose. Specifically, users check the estimated loss amount displayed on the web page, and if the risk is high, they can take measures such as changing the content of the information they disclose.

[1692] Specific examples

[1693] Example 1: Acquiring and analyzing flame data

[1694] The server retrieves past social media flame data from a MySQL® database, using, for example, the following SQL query:

[1695] sql

[1696] SELECT FROM flame_data WHERE date >= '2020-01-01';

[1697] The acquired data is analyzed using the Python (registered trademark) pandas library to estimate the loss amount in the event of a fire. For example, the following code is used:

[1698] Python (registered trademark)

[1699] import pandas as pd

[1700] from sklearn.linear_model import LinearRegression

[1701] Loading data

[1702] data = pd.read_sql('SELECT FROM flame_data WHERE date >= "2020-01-01"', con=database_connection)

[1703] Estimated loss amount

[1704] model = LinearRegression()

[1705] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[1706] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[1707] Example 2: Displaying risk information

[1708] When a user makes a disclosure, the device retrieves the risk information from the server and displays it in an HTML page, for example using code like this:

[1709] html

[1710] <!DOCTYPE html>

[1711]

[1712]

[1713] <title> Risk Information< / title>

[1714]

[1715]

[1716] <h1> Risk Information< / h1>

[1717] Estimated loss amount: \

[1718] <script>

[1719] fetch('https: / / example.com / api / risk_info')

[1720] .then(response => response.json())

[1721] .then(data => {

[1722] document.getElementById('loss_amount').textContent = data.loss_amount;

[1723] });

[1724] < / script>

[1725]

[1726]

[1727] Prompt Sentence Examples

[1728] Examples of prompts to input to a generative AI model might include:

[1729] "Generate Python (registered trademark) code that uses past social media flame war data to estimate the amount of losses incurred during a flame war."

[1730] Using this prompt sentence, the generative AI model can generate the appropriate Python code.

[1731] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1732] Step 1:

[1733] The server retrieves past flame data from the database. It receives database connection information and an SQL query as input, and generates a data frame containing the flame data as output. Specifically, the server executes the SQL query to extract the past flame data. For example, the following SQL query is used:

[1734] sql

[1735] SELECT FROM flame_data WHERE date >= '2020-01-01';

[1736] This query retrieves flame data for the specified time period.

[1737] Step 2:

[1738] The server analyzes the acquired data. It receives a data frame as input and generates a preprocessed data frame as output. Specifically, the server completes missing values ​​in the data and removes unnecessary columns. For example, it preprocesses the data using the Python (registered trademark) pandas library.

[1739] Step 3:

[1740] The server uses the preprocessed data to estimate the loss amount. It receives the preprocessed data frame as input and generates the estimated loss amount as output. Specifically, the server uses a LinearRegression model from scikit-learn (registered trademark) to estimate the loss amount. For example, the model is trained as follows to predict the loss amount:

[1741] Python (registered trademark)

[1742] model = LinearRegression()

[1743] model.fit(data[['followers', 'negative_comments']], data['loss_amount'])

[1744] predicted_loss = model.predict(new_data[['followers', 'negative_comments']])

[1745] This process allows the amount of loss to be estimated.

[1746] Step 4:

[1747] The server stores the estimated loss amount in a database. It receives the estimated loss amount and database connection information as input, and generates the loss amount stored in the database as output. Specifically, the server executes an SQL query to insert the estimated result into the database. For example, it uses the following SQL query:

[1748] sql

[1749] INSERT INTO risk_info (date, predicted_loss) VALUES ('2023-10-01', 500000);

[1750] This query results in the estimated loss amount being stored in a database.

[1751] Step 5:

[1752] When a user discloses information, the device obtains risk information from the server. It receives an HTTP GET request as input and generates JSON data containing the risk information as output. Specifically, the device sends an HTTP GET request to the server and receives the risk information. For example, the following JavaScript code is used:

[1753] javascript

[1754] fetch('https: / / example.com / api / risk_info')

[1755] .then(response => response.json())

[1756] .then(data => {

[1757] document.getElementById('loss_amount').textContent = data.predicted_loss;

[1758] });

[1759] Through this process, risk information is obtained.

[1760] Step 6:

[1761] The terminal displays the acquired risk information to the user. It receives JSON data containing risk information as input and generates an HTML page displaying the risk information as output. Specifically, the terminal uses HTML and JavaScript to display the risk information on a web page. For example, it can be implemented as follows:

[1762] html

[1763] <!DOCTYPE html>

[1764]

[1765]

[1766] <title> Risk Information< / title>

[1767]

[1768]

[1769] <h1> Risk Information< / h1>

[1770] Estimated loss amount: \

[1771] <script>

[1772] fetch('https: / / example.com / api / risk_info')

[1773] .then(response => response.json())

[1774] .then(data => {

[1775] document.getElementById('loss_amount').textContent = data.predicted_loss;

[1776] });

[1777] < / script>

[1778]

[1779]

[1780] This process displays risk information to the user.

[1781] Step 7:

[1782] The user checks the risk information through the terminal. As input, the user receives the displayed risk information, and as output, adjusts the content of the information disclosure. Specifically, the user checks the estimated loss amount displayed on the web page, and if the risk is high, takes action such as changing the disclosure content.

[1783] (Application example 2)

[1784] Next, a description will be given of Application Example 2 of Form Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1785] There is a need for a system that can quantify the risk of a controversy and predict the amount of loss in advertising campaigns using past controversy data. However, current systems have the problem of being unable to specifically quantify the risk of a controversy, disclose information based on that risk, or provide specific advice for reducing the risk.

[1786] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1787] In this invention, the server includes means for utilizing past flame data, means for utilizing data on effective words, means including a generative AI model for predicting the economic impact of publishing on social media or a website based on this data, means for quantifying the risk of a flame based on past flame data when the content of an advertising campaign is input, means for estimating the amount of loss based on the flame risk, and means for providing specific advice for reducing the risk. This makes it possible to specifically quantify the risk of a flame in an advertising campaign and provide information disclosure based on that risk and specific advice for reducing the risk.

[1788] "Past flame war data" refers to data such as negative reactions, comments, and criticisms that occurred on social media or websites in the past.

[1789] "Effective word data" refers to data on keywords and phrases that are considered effective in eliciting positive responses on social media and websites.

[1790] A "generative AI model" refers to an algorithm or system that uses machine learning or artificial intelligence techniques to generate a specific output from input data.

[1791] "Means for quantifying the risk of a backlash" refers to a method or system that uses data on past backlashes to numerically express the possibility that a particular advertising campaign will cause a backlash.

[1792] "Means for estimating loss amount" refers to a method or system for calculating expected economic loss based on the risk of a fire.

[1793] "Means for providing specific advice" refers to methods or systems that suggest specific actions or measures to reduce the risk of online outrage.

[1794] The system for carrying out the present invention operates in cooperation with three entities: a server, a terminal, and a user. A specific embodiment of the system will be described below.

[1795] Server Roles

[1796] The server stores data on past flame wars and effective keywords, and uses a generative AI model based on this data to predict economic impact. The server uses the following software and hardware:

[1797] Software: Python (registered trademark), Pandas, Scikit-learn (registered trademark)

[1798] Hardware: A server with a powerful processor and large memory capacity

[1799] The server loads past flame data and converts it into a data frame. Next, it sets features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model. This allows it to quantify the flame risk and predict loss amounts by inputting the details of an advertising campaign.

[1800] Device Role

[1801] The terminal provides an interface for users to input the content of their advertising campaigns. The terminal uses the following software and hardware:

[1802] Software: Web browser, mobile application

[1803] Hardware: Smartphones, tablets, computers

[1804] When a user enters details of an ad campaign through their device, the data is sent to a server, which then calculates the risk of a potential online outrage and predicts the amount of damage. The prediction is then sent back to the device and displayed to the user.

[1805] User Roles

[1806] Users can input details of their advertising campaigns using the terminal, check the predicted results of the risk of outrage and the amount of loss, and receive specific advice from the server on how to reduce the risk and take appropriate measures.

[1807] Specific examples

[1808] For example, if a user inputs an ad campaign length of 30 days, an audience size of 100,000, and the number of negative comments as 50, the server will use this data to quantify the risk of outrage and calculate the expected loss amount. As a result, the estimated loss amount is displayed and specific advice on how to reduce the risk is provided.

[1809] Prompt Sentence Examples

[1810] "If your ad campaign is 30 days long, your audience size is 100,000, and you have 50 negative comments, what is your estimated loss?"

[1811] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[1812] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1813] Step 1:

[1814] A user uses a terminal to enter details of an advertising campaign.

[1815] Inputs: length of ad campaign, audience size, number of negative comments

[1816] Output: The input data is sent to the server

[1817] What it does: Users enter details about their advertising campaign through a smartphone or computer interface and click the "Submit" button.

[1818] Step 2:

[1819] Based on the data received by the server, past flame data is read.

[1820] Input: Details of the ad campaign submitted by the user

[1821] Output: Historical fire data converted into a data frame

[1822] Specific operation: The server uses the Python (registered trademark) Pandas library to read past fire data from a CSV file and convert it into a data frame.

[1823] Step 3:

[1824] The server sets the features and target and trains a linear regression model.

[1825] Input: Historical fire data converted into a data frame

[1826] Output: A trained linear regression model

[1827] How it works: The server uses the Scikit-learn (registered trademark) library to set features (campaign length, audience size, number of negative comments) and targets (loss amounts) and trains a linear regression model.

[1828] Step 4:

[1829] Based on the detailed data of the advertising campaign received by the server from the user, the risk of a backlash is quantified and the amount of loss is predicted.

[1830] Input: Detailed advertising campaign data received from users, a trained linear regression model

[1831] Output: Fire risk figure and predicted loss amount

[1832] How it works: The server inputs the received detailed data of the advertising campaign as features, and uses a trained linear regression model to quantify the risk of a backlash and predict the amount of loss.

[1833] Step 5:

[1834] The server returns the predicted results of the risk of a firestorm and the amount of loss to the terminal.

[1835] Input: Flame risk value and predicted loss amount

[1836] Output: The prediction results are displayed on the terminal.

[1837] Specific operation: The server sends the prediction results in JSON format to the device, which receives them and displays them to the user.

[1838] Step 6:

[1839] The server generates specific advice to reduce the risk and sends it to the terminal.

[1840] Input: Flame risk value and predicted loss amount

[1841] Output: Specific advice for reducing risk

[1842] Specific operation: The server uses the generative AI model to generate specific advice for reducing the risk of flame wars and sends it to the device, which receives it and displays it to the user.

[1843] In this way, it is possible to specifically quantify the risk of a backlash in an advertising campaign, and provide information disclosure based on that risk and specific advice for reducing the risk.

[1844] Example 3

[1845] Next, a third embodiment of the third embodiment will be described. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1846] In today's information society, there is an increasing risk of a company becoming involved in a backlash when publishing information on social media or on its website. This risk can damage a company's image and cause financial losses, so companies are required to assess the risk of a backlash in advance and revise information as necessary. However, the current system does not adequately assess the risk of a backlash or revise information, which is a major issue for companies.

[1847] The specific processing by the specific processing unit 290 of the data processing device 12 in the third embodiment is realized by the following means.

[1848] In this invention, the server includes means for utilizing past flame war data, means for utilizing data on effective words, means for using a generative AI model that generates information based on these data, means for assessing the flame war risk before the generated information is made public, means for encouraging the correction of information if there is a flame war risk, and means for predicting the economic effect of publishing the information on social media or a website based on these means. This enables companies to prevent damage to their corporate image by assessing the flame war risk before publishing information and correcting the information as necessary.

[1849] "Past flame war data" refers to data on past flame wars that occurred as a result of information that was made public on social media, websites, etc.

[1850] "Effective word data" refers to data on words and phrases that are considered effective in eliciting positive responses when publishing information on social media or websites.

[1851] A "generative AI model" is an artificial intelligence model that generates information based on prompts entered by the user.

[1852] The "means for assessing the risk of a controversy" is a means for analyzing and assessing the possibility that the generated information will cause a controversy when it is made public.

[1853] "Means to encourage correction of information" refers to means to encourage users to correct information that is assessed as having a risk of causing a flame war.

[1854] "Means for predicting economic effects" refers to a means for predicting the economic impact that information published on social media or websites will have on a company.

[1855] MODE FOR CARRYING OUT THE INVENTION

[1856] This invention is a system that prevents damage to a company's image by assessing the risk of a controversy beforehand when a company publishes information on social media or a website, and correcting the information as necessary. This system consists of three main elements: a server, a terminal, and a user.

[1857] 1. Program Generation

[1858] The server generates a program to prevent damage to a company's image in advance. This program uses a generative AI model to create or edit information and evaluate the risk of that information causing a backlash before it is made public. Specifically, OpenAI's GPT-4 (registered trademark) is used as the generative AI model.

[1859] 2. Program Processing

[1860] The server uses a generative AI model (e.g., OpenAI's GPT-4®) to generate information based on the prompt entered by the user. The generated information is then sent to an inflammability risk assessment module. This module uses natural language processing (NLP) techniques to analyze the content of the information and assess whether it poses an inflammability risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information poses an inflammability risk.

[1861] The terminal sends the prompt sentence entered by the user to the server, and receives the generated information and the risk assessment result returned from the server. The user checks the risk assessment result on the terminal and corrects the information as necessary.

[1862] 3. Examples of concrete examples and prompts

[1863] As a concrete example, consider a user creating a press release for a new product. The user types the following prompt into the terminal:

[1864] Example prompt sentence:

[1865] "Write a press release for your new product. Call it 'EcoSmart' and highlight its environmentally friendly features."

[1866] The server inputs this prompt into a generative AI model and generates the following information:

[1867] Examples of generated information:

[1868] "EcoSmart is a new product that utilizes the latest environmental technology. It is energy efficient and made from recyclable materials."

[1869] The server then sends the generated information to the flame risk assessment module, which performs a risk assessment. If the assessment results in a "risk," the server sends a message to the user urging them to make corrections. The user can then check the suggested corrections on their device and make any necessary corrections to prevent damage to the company's image.

[1870] This system enables companies to assess the risk of information being exposed to controversy before it is released and correct the information as necessary, thereby preventing damage to the company's image in advance. The flow of the identification process in the third embodiment will be described with reference to FIG.

[1871] Step 1:

[1872] The user enters a prompt statement.

[1873] The user inputs a prompt into the input field of the terminal, for example, "Please write a press release for a new product. The product name should be 'EcoSmart' and the product's environmentally friendly features should be emphasized." The input prompt is temporarily stored in the terminal's memory.

[1874] Step 2:

[1875] The terminal sends the prompt to the server.

[1876] The terminal converts the prompt text entered by the user into an appropriate format and sends a request to the server's API endpoint. Specifically, it sends the prompt text to the server using an HTTP POST request. The input is the prompt text, and the output is the request sent to the server.

[1877] Step 3:

[1878] The server generates information using a generative AI model.

[1879] The server inputs the received prompt into a generative AI model (e.g., OpenAI's GPT-4 (registered trademark)) to generate information. The generative AI model analyzes the prompt and generates information based on the appropriate context. The input is the prompt, and the output is the generated information. For example, the information generated is, "EcoSmart is a new product that utilizes the latest environmental technology. It is highly energy efficient and uses recyclable materials."

[1880] Step 4:

[1881] The server sends the generated information to a flame risk assessment module.

[1882] The server sends the generated information to the flame risk assessment module for risk assessment. This module uses natural language processing (NLP) technology to analyze the content of the information and assess whether it poses a flame risk. Specifically, it uses a risk assessment algorithm (e.g., a BERT-based classification model) to determine whether the information contains a flame risk. The input is the generated information, and the output is the risk assessment result.

[1883] Step 5:

[1884] The server generates the risk assessment results and sends them to the terminal.

[1885] The server receives the evaluation results from the flame risk assessment module and sends them to the terminal. The evaluation results include judgments such as "no risk" or "risk present." The input is the risk assessment result, and the output is the transmission to the terminal.

[1886] Step 6:

[1887] The terminal displays the risk assessment results to the user.

[1888] The terminal displays the risk assessment results received from the server to the user. For example, a message such as "This information has the risk of causing a firestorm. We recommend that you modify it" is displayed. The input is the risk assessment result, and the output is what is displayed to the user.

[1889] Step 7:

[1890] The user modifies the information as needed.

[1891] The user checks the risk assessment results displayed on the device and modifies the information as necessary. The modified information is sent back to the server, where it undergoes the same process before finally being published as safe informa...

Claims

1. A means of collecting and storing past flame war data, A means for collecting and storing effective word data; A means for analyzing the flame war data and the effective words using natural language processing technology; A means of predicting the economic impact of public information based on the analysis results; A means for supporting at least one of correcting and creating public information; A system including:

2. A way to utilize past flame war data, How to use effective word data, A means for using a generative AI model to generate information based on the flame data and the effective words; A means of assessing the risk of outrage before the information generated is made public; and A method to prompt correction of information when there is a risk of outrage; A means for predicting the economic impact of releasing the generated information; and A system including:

3. By inputting the details of your ad campaign, you can get a numerical value for the risk of a backlash based on past backlash data, A means for estimating the amount of loss based on the risk of a firestorm; measures that provide specific advice on reducing risks; A system including:

Images