System

Abstract

A system is provided.SOLUTION: A system comprising: means for collecting textual data generated on the Internet; means for training a generative AI model that learns slang from the collected textual data; means for detecting slang in new textual data using the trained generative AI model; means for analyzing patterns of use and context of the slang and assessing risk of an account being analyzed; and means for identifying risky users based on the assessment.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] On the Internet, recruitment for criminal activities is often carried out using code words, and seemingly harmless communications using code words are frequently used, especially on online platforms such as social media. This code words can evade surveillance, making it difficult to detect and identify criminal activity. Therefore, a reliable system is needed to continuously understand the evolving code words and identify criminal activity. [Means for solving the problem]

[0005] The present invention solves this problem by providing a system that includes a means for collecting text data generated on the Internet, a means for training a generative AI model that learns slang from the collected text data, a means for detecting slang in new text data using the trained generative AI model, a means for analyzing slang usage patterns and contexts and assessing the risk level of the account being analyzed, and a means for identifying risky users based on the assessment results. The system also includes a means for collecting new text data in real time and saving it as an analysis target, and a means for saving the learned slang usage patterns as a dictionary and using the dictionary to detect slang. This system enables the identification of criminal activity on the Internet and provides a safe online environment.

[0006] The "Internet" is a global collection of computer networks and a digital platform that enables information exchange and communication.

[0007] "Text data" is a collection of information made up of characters and symbols, and is digital data structured in the form of sentences, words, etc.

[0008] "Slang" is a word or phrase that has a special meaning that is not generally recognized and is used by a particular group or for a particular purpose.

[0009] A "generative AI model" is an artificial intelligence algorithm that has the ability to generate new data based on input data or analyze existing data.

[0010] "Training" is the process of providing an artificial intelligence model with specific data and having it learn the patterns and characteristics of that data.

[0011] "Detecting jargon" is the process of identifying and confirming the presence of jargon contained within text data.

[0012] A "usage pattern" is, for example, a statistical distribution or tendency that indicates in what contexts or situations a particular jargon is frequently used.

[0013] "Context" is the information or circumstances surrounding the use of a particular word or phrase that determine its meaning.

[0014] An "account" is a unit that holds the identification information required for a user to use a service on the Internet.

[0015] Evaluating "risk" is the process of expressing the likelihood that an account is related to criminal activity in a numerical or categorical way based on the context of the text data and usage patterns.

[0016] "Identifying" is the process of accurately identifying and recognizing what an object is based on the analysis results.

[0017] "Real-time" is a time concept that involves immediate processing and response to events that occur or acquired data.

[0018] A "dictionary" is a database that compiles a set of words and phrases, their meanings, and their uses. [Brief explanation of the drawings]

[0019] [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. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0020] 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.

[0021] First, the terms used in the following description will be explained.

[0022] 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, a 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), and an APU (Accelerated Processing Unit).

[0023] 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.

[0024] 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.

[0025] 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.

[0026] 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."

[0027] [First embodiment]

[0028] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0029] 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.

[0030] 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).

[0031] 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.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0036] 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.

[0037] 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.

[0038] 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.

[0039] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. 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."

[0040] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[0041] 1. Data Collection Phase

[0042] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[0043] 2. Secret Language Learning Phase

[0044] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[0045] 3. The jargon analysis phase

[0046] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[0047] 4. Identifying risky users

[0048] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[0049] Specific examples

[0050] For example, suppose there is a social media post like the one below.

[0051] User A: "Can you bring me some crystals tonight?"

[0052] User B: "Okay. In the usual place."

[0053] In this case, the server performs the following process.

[0054] 1. Data Collection:

[0055] The server retrieves these posts from the internet and stores them in a database.

[0056] 2. Learn the secret language:

[0057] A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[0058] 3. Analysis of the jargon:

[0059] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[0060] 4. Identifying risky users:

[0061] The server analyzes the past posting history and interaction patterns of User A and User B and assesses that there is a high possibility that they are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[0062] In this way, this system can quickly identify accounts that may be involved in criminal activities through a series of processes that collect and analyze text data on the Internet, thereby improving the safety of the Internet environment.

[0063] The processing flow will be explained below.

[0064] Step 1:

[0065] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[0066] Step 2:

[0067] The server stores text data obtained from various internet sources such as social networking sites, message boards, and blogs in a temporary buffer. The collected data includes information about the poster (user ID, posting time, etc.).

[0068] Step 3:

[0069] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[0070] Step 4:

[0071] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[0072] Step 5:

[0073] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[0074] Step 6:

[0075] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[0076] Step 7:

[0077] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[0078] Step 8:

[0079] The server uses the generative AI model and the slang dictionary to analyze new data for the presence of slang, and once identified, analyzes the context in which the slang is used.

[0080] Step 9:

[0081] The server evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[0082] Step 10:

[0083] The server creates a profile of each poster using the code, including their posting history, the content they post, and their patterns of interaction with other users.

[0084] Step 11:

[0085] The server uses a generative AI model to assess the riskiness of user profiles, taking into account factors such as posting frequency, use of slang, and association with other high-risk users.

[0086] Step 12:

[0087] Based on the evaluation results, the server identifies high-risk users. If necessary, the administrator is notified of the identified risky users and appropriate measures are taken.

[0088] Example 1

[0089] Next, a description will be given of 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."

[0090] In recent years, the use of slang suggesting criminal activity on social networking sites, blogs, and bulletin boards on the Internet has led to an increase in illegal transactions and activities. However, these slang terms cannot be easily detected using conventional filtering technology, making it difficult to quickly and accurately identify criminal activity using slang terms. Therefore, there is a need for a system that can effectively detect slang terms on the Internet and prevent criminal activity by analyzing the context of their use.

[0091] 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.

[0092] In this invention, the server includes means for collecting text data generated on the Internet, means for temporarily saving the collected text data in a buffer and storing it in a database, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing the usage context of the slang in the newly collected text data, means for analyzing the usage patterns and context of the slang and evaluating the risk level of the account being analyzed, and means for identifying risky users based on the evaluation results and notifying an administrator as necessary. This makes it possible to quickly and accurately detect slang suggestive of criminal activity on the Internet and analyze the usage context to identify potentially risky users.

[0093] The "Internet" is a collection of globally connected computer networks and an infrastructure for information exchange and communication.

[0094] "Text data" refers to data composed of characters, such as sentences, words, and phrases, in a format that humans can read and write.

[0095] "Collecting" refers to the act of selecting and acquiring data based on specific criteria or filters and putting it in one place.

[0096] A "buffer" is a storage area for temporarily storing data, and is used to improve the efficiency of data transfer and processing.

[0097] A "database" is a system for efficiently storing, searching, and managing structured data, and includes software such as MySQL and MongoDB.

[0098] A "generative AI model" is an artificial intelligence model that is trained using machine learning algorithms to perform specific tasks, such as natural language processing or image recognition.

[0099] "Training" refers to the act of using collected data to train a generative AI model and improve its ability to perform a specific task.

[0100] "Arguments" are special words or phrases that are only understood by certain groups or in certain contexts, and often have meanings that are not known to the general public.

[0101] "Detect" refers to the act of finding specific patterns, words, phrases, etc. in data.

[0102] "Context" refers to the context or situation in which a word or phrase is used, including background information needed to understand its meaning or intent.

[0103] "Behavioral patterns" refer to the actions and tendencies that a particular user repeats on the Internet, and by analyzing these, it is possible to predict that person's behavior.

[0104] "Risk level" is an evaluation indicator that indicates the likelihood that a particular account or behavior is related to criminal or illegal activity.

[0105] "Notifying" refers to the act of communicating specific information or results to a third party, including alerting a system administrator.

[0106] MODE FOR CARRYING OUT THE INVENTION

[0107] This invention is a system that collects text data generated on the Internet, detects slang in new text data using a generative AI model that learns slang, and analyzes the context of the slang to assess its risk level. Specific embodiments are described below.

[0108] System Configuration

[0109] The system of the present invention consists of the following main components:

[0110] 1. Data Collection Methods

[0111] 2. Data Buffers and Databases

[0112] 3. Generative AI models and their training methods

[0113] 4. Idiom detection and context analysis methods

[0114] 5. Risk Assessment Tools

[0115] 6. Administrator notification method

[0116] Hardware and software used

[0117] Hardware: Servers (general computer servers), network infrastructure

[0118] Software: SNS API, database software (e.g., MySQL, MongoDB), generative AI models (e.g., Transformer, BERT)

[0119] Specific operation explanation

[0120] Data collection

[0121] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. Specifically, it uses social media APIs to set specific keyword filters and retrieves posts that match the criteria. For example, it uses prompts like the following:

[0122] "Get recent posts that match a specific keyword"

[0123] The acquired data is temporarily stored in a buffer and then stored in a database.

[0124] Learning slang

[0125] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques such as Transformer and BERT, to learn specific jargon and its context. For example, it learns that the word "crystal" means stimulant in a specific context.

[0126] Code detection and context analysis

[0127] The trained generative AI model is then used to analyze newly collected text data. The server scans the text for words and phrases to detect whether they contain slang. Specifically, it uses prompts like:

[0128] "Scan specific words and phrases in new data to detect slang"

[0129] Additionally, the context in which the slang is used is analyzed to determine whether it is related to illegal activity.

[0130] Risk assessment and administrator notification

[0131] The server evaluates the behavioral patterns of accounts that use a lot of slang based on the analysis results. The evaluation includes the frequency of posts, interaction patterns with other accounts, and past posting history. This allows it to identify potentially dangerous users and evaluate their risk level using a number or category. Based on the evaluation results, it notifies administrators as necessary. The notification includes details such as potential illegal activity, slang used, and past behavioral patterns. For example, the following notifications may be sent:

[0132] "User A frequently uses the code word 'crystal' and has been identified as a high-risk user."

[0133] Specific examples

[0134] For example, consider the following social media post:

[0135] User A: "Can you bring me some crystals tonight?"

[0136] User B: "Okay. In the usual place."

[0137] The process flow in this case is as follows:

[0138] 1. Data collection: The server retrieves these posts, stores them in a buffer, and then stores them in a database.

[0139] 2. Slang learning: The generative AI model learns that the word "crystal" is slang.

[0140] 3. Slang analysis: Detect the word "crystal" in the newly collected data and analyze its context.

[0141] 4. Identifying risky users: The server analyzes the behavioral patterns of users A and B and notifies the administrator that they are high-risk users.

[0142] In this way, this system can quickly and accurately analyze text data generated on the Internet and identify potentially dangerous users, thereby improving the safety of the Internet environment.

[0143] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0144] Step 1: Data collection

[0145] The server uses the APIs of social media, bulletin boards, blogs, etc. to set specific keyword filters and periodically collect posts that match the conditions. For example, posts that match the keyword "crystal" are obtained through the social media API. The input is text data obtained through the social media API, and the output is data temporarily stored in a buffer. Specifically, the server makes an API call and saves the obtained data in a buffer in memory.

[0146] Step 2: Temporarily save to the data buffer

[0147] The server temporarily stores the collected text data in a buffer. The input is text data collected from social media and bulletin boards, and the output is the data in the buffer that has been temporarily stored. This temporary storage allows the integrity of the data to be checked before it is subsequently stored in the database. Specifically, the server performs a process to insert the data into a data structure in memory.

[0148] Step 3: Store in the database

[0149] The server takes the text data from the buffer and stores it in the database. The input is the text data in the buffer, and the output is the data stored in the database. Specifically, it issues an INSERT query to a database system such as MySQL or MongoDB.

[0150] Step 4: Training the generative AI model

[0151] The server uses the text data collected from the database to train a generative AI model. The input is the text data in the database, and the output is the trained generative AI model. Specifically, it uses AI models such as Transformer and BERT to learn specific jargon and its context through supervised learning.

[0152] Step 5: Collect new data

[0153] The server processes newly collected text data in real time. The input is text data from social media and bulletin boards collected in real time, and the output is real-time data in a buffer. Specifically, the server continuously calls the API to collect new post data.

[0154] Step 6: Scan for jargon

[0155] The server uses a trained generative AI model to scan new data for slang. The input is text data collected in real time, and the output is a list of detected slang words. Specifically, the inference function of the generative AI model is used to identify slang words within the text.

[0156] Step 7: Parsing the Context

[0157] The server analyzes the context of slang usage. The input is a list of detected slang words and their surrounding text, and the output is the context analysis results. Specifically, it uses natural language processing technology to analyze the context in which the slang words are used.

[0158] Step 8: Evaluate behavioral patterns

[0159] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang. The input is the context analysis results and the user's posting history data, and the output is a risk assessment result. Specifically, the server analyzes posting frequency, interaction patterns, past posting history, etc. as evaluation indicators.

[0160] Step 9: Risk Assessment

[0161] The server evaluates the user's risk level using a numerical value or category based on the behavioral pattern evaluation. The input is the behavioral pattern evaluation result, and the output is the risk evaluation result and a list of high-risk users. Specifically, a risk scoring algorithm is used to quantify the user's risk level.

[0162] Step 10: Notify administrators

[0163] The server identifies high-risk users and notifies administrators as necessary. The input is the risk assessment result, and the output is a notification message sent to the administrator. Specifically, email and dashboard notifications are sent when certain conditions are met.

[0164] (Application example 1)

[0165] Next, a description will be given of Application 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."

[0166] Conventional systems detect and assess the risk of slang only in text data on the Internet, but do not address slang contained in audio or image data. This makes it difficult to monitor dangerous behavior in real time, and is insufficient in situations where a rapid response is required, such as in public places or events.

[0167] 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.

[0168] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for notifying users of the analyzed information in real time, means for collecting and analyzing audio data, and means for collecting and analyzing image data. This makes it possible to detect dangerous slang in real time from various data including audio and image data.

[0169] A "server" is a computer system that collects and analyzes data over a network.

[0170] The "Internet" is a huge information network that spans the entire world and is a platform for sending and receiving various types of information.

[0171] "Text data" refers to information composed of character strings, including social media posts, blog articles, and message board posts on the Internet.

[0172] A "generative AI model" is a model that uses artificial intelligence technology to learn patterns and features from text data.

[0173] "Training" is the process of feeding a generative AI model large amounts of data to teach it specific patterns and characteristics.

[0174] "Argot" is a word or phrase that has a different meaning from its usual meaning among a particular group or in a particular situation.

[0175] "Usage patterns" refer to how frequently and in what contexts particular words or phrases are used.

[0176] "Context" refers to the context or situation in which a particular word or phrase is used.

[0177] "Danger level" is an indicator of the likelihood that a particular account is involved in criminal activity, derived from the pattern and context of code word usage.

[0178] "User" is a term that refers to a person or account that posts information on the Internet.

[0179] "Real-time" is a term that indicates the extremely short time between the occurrence of an event and the information being processed and analyzed.

[0180] "Notification" is a means by which the system informs the user of analysis results and important information.

[0181] "Voice data" refers to data that records human speech or sound in digital format.

[0182] "Image data" is visual data that records light information in digital format.

[0183] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[0184] Data Collection Phase

[0185] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[0186] Secret language learning phase

[0187] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[0188] The jargon analysis phase

[0189] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[0190] Risky user identification phase

[0191] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[0192] Real-time notification and multi-data analysis functions

[0193] The analyzed information is then sent to users in real time via smart glasses or other wearable devices. The system also collects audio and image data for similar analysis. For example, speech recognition technology is used to convert speech at events into text and analyze high-risk slang. It also uses Google Cloud Vision API to extract text from images and perform similar analysis.

[0194] Hardware and Software Used

[0195] This system uses the following hardware and software:

[0196] Server: A central processing unit that performs data collection and analysis.

[0197] SNS API: An API used to collect data.

[0198] Generative AI model: A model for learning and analyzing jargon from data.

[0199] Speech recognition libraries (e.g. speech_recognition): Used to convert voice data into text.

[0200] Google Cloud Vision API: Used to extract text from image data.

[0201] Google Cloud Natural Language API: Used to parse text data and extract entities.

[0202] Specific examples

[0203] For example, consider the following social media post:

[0204] User A: "Can you bring me some crystals tonight?"

[0205] User B: "Okay. In the usual place."

[0206] In this case, the server performs the following process.

[0207] Data collection: The server retrieves these posts from the internet and stores them in a database.

[0208] Slang learning: A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[0209] Code word analysis: The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[0210] Identifying risky users: Analyzing the past posting history and interaction patterns of User A and User B, the server assesses the likelihood that the two are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[0211] Example prompt for a generative AI model:

[0212] Check to see if any of the following words appear in today's conversations: crystal, popcorn, white circle.

[0213] If you find any, assess whether they may be related to illegal activity.

[0214] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0215] Step 1:

[0216] The server collects text data generated on the Internet.

[0217] Input: Text data from the internet

[0218] Output: Collected text data

[0219] How it works: The server uses the SNS API to set up a specific keyword filter and periodically retrieves posts that match the filter. This data is temporarily stored in a buffer and then stored in a database.

[0220] Step 2:

[0221] The server uses the collected text data to train a generative AI model.

[0222] Input: Collected text data

[0223] Output: A trained generative AI model

[0224] How it works: The server extracts text data from a database and uses natural language processing (NLP) techniques to train a generative AI model to understand jargon and its context.

[0225] Step 3:

[0226] The server uses the trained generative AI model to detect jargon in new text data.

[0227] Input: New text data, trained generative AI model

[0228] Output: Detected slang

[0229] How it works: The server analyzes newly collected text data in real time and uses a generative AI model to detect whether it contains slang.

[0230] Step 4:

[0231] The server analyzes the usage patterns and context of slang and assesses the risk level of the account being analyzed.

[0232] Input: Detected slang, new text data

[0233] Output: Account risk assessment results

[0234] Specific operation: The server evaluates the frequency of slang use, the context of use, interaction patterns with other accounts, etc., and evaluates the risk level of each account using a number or category.

[0235] Step 5:

[0236] The server identifies risky users based on the evaluation results.

[0237] Input: Account risk assessment results

[0238] Output: Identified risky users

[0239] Specific operation: The server lists accounts that are assessed as high risk and prepares to notify the administrator of the list.

[0240] Step 6:

[0241] The server notifies the user of the analyzed information in real time.

[0242] Input: Analysis results

[0243] Output: User notification

[0244] Specific operation: The server generates a warning message based on the analysis results and notifies wearable devices such as smart glasses in real time.

[0245] Step 7:

[0246] The device collects and analyzes audio data at events and public places.

[0247] Input: Audio data

[0248] Output: Text data

[0249] Specific operation: The device (smart glasses, etc.) converts audio data recorded at events or public places into text using the SpeechRecognition library.

[0250] Step 8:

[0251] The terminal collects and analyzes image data.

[0252] Input: Image data

[0253] Output: Text data

[0254] Specific operation: The device (smart glasses, etc.) converts the captured image into text using the Google Cloud Vision API and sends the result to the server.

[0255] The above are the processing steps of the present invention, which make it possible to detect and notify dangerous slang from a variety of data sources in real time.

[0256] 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.

[0257] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is built around a server and performs a series of processes, including data collection, analysis, evaluation, and notification.

[0258] 1. Data Collection Phase

[0259] The server collects text data from various sources on the Internet (e.g., social media, bulletin boards, blogs, etc.). The collected data is temporarily stored in a buffer and then stored in a database. A regular scheduling process is performed to collect new data in real time.

[0260] 2. Secret Language Learning Phase

[0261] The server uses the collected text data to train a generative AI model, which learns specific slang terms and the context in which they are used. For example, it learns from the training data that "crystal" means stimulant in certain contexts.

[0262] 3. The jargon analysis phase

[0263] Using the trained generative AI model, the server can detect code words in newly collected text data, and if code words are present, analyze their context in detail to determine whether they are related to illegal activity.

[0264] 4. Analysis phase using emotion engine

[0265] The server uses an emotion engine to recognize emotions from users' text data. This engine can analyze the emotional tone and changes in the content of posts. For example, if a post contains anger or stress, it can determine that this is likely related to the use of slang.

[0266] 5. Risky user identification phase

[0267] The server evaluates the user's risk level based on the analysis results of the emotion engine and the usage pattern of slang. The evaluation takes into account the frequency of posts, frequency of slang use, and emotional fluctuations. For example, if the context in which slang is used is accompanied by tension or excitement, the risk level is evaluated as high.

[0268] 6. Notification Phase

[0269] The server identifies users who are assessed as high risk and notifies the administrator, who uses this information to guide appropriate action.

[0270] Specific examples

[0271] For example, suppose there is a social media post like the one below.

[0272] User A: "Bring me the crystal tonight, at the place I told you about."

[0273] User B: "Be careful. Come early. I'm getting annoyed."

[0274] In this case, the processing flow is as follows.

[0275] 1. Data Collection:

[0276] The server collects these posts and stores them in a database.

[0277] 2. Learn the secret language:

[0278] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0279] 3. Analysis of the jargon:

[0280] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0281] 4. Analysis by emotion engine:

[0282] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0283] 5. Identifying risky users:

[0284] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0285] 6. Notice:

[0286] The server notifies the administrator of the results and prompts them to take appropriate action.

[0287] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[0288] The processing flow will be explained below.

[0289] Step 1:

[0290] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[0291] Step 2:

[0292] The server stores text data obtained from various internet sources such as social media, message boards, and blogs in a temporary buffer. The collected data also includes information about the poster (user ID, posting time, etc.).

[0293] Step 3:

[0294] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[0295] Step 4:

[0296] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[0297] Step 5:

[0298] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[0299] Step 6:

[0300] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[0301] Step 7:

[0302] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[0303] Step 8:

[0304] The server uses the generative AI model and slang dictionary to analyze new data to determine whether it contains slang. If slang is identified, it analyzes the context and surrounding words.

[0305] Step 9:

[0306] The server then carefully evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[0307] Step 10:

[0308] The server uses an emotion engine to recognize the poster's emotions from the text data. This emotion analysis includes the tone of the text, the intensity of the emotion, and the type of emotion (joy, anger, sadness, etc.).

[0309] Step 11:

[0310] The server combines the identified emotional information with the slang usage patterns and stores them in a database, making the correlation between emotional fluctuations and slang usage available for future analysis.

[0311] Step 12:

[0312] The server evaluates the results of the emotion engine and the usage patterns of slang comprehensively to assess the risk level of each user. For example, if the context in which slang is used is emotionally charged, the risk level is evaluated as high.

[0313] Step 13:

[0314] The server identifies high-risk users based on the evaluation results, which include the behavioral patterns, past posts, and emotional fluctuations of the identified risky users.

[0315] Step 14:

[0316] The server notifies the administrator of users who are rated as highly dangerous. The notification includes the user's ID, the content of their posts, the risk assessment result, and the related sentiment analysis result.

[0317] Specific examples

[0318] For example, suppose there is a social media post like the one below.

[0319] User A: "Bring me the crystal tonight, at the place I told you about."

[0320] User B: "Be careful. Come early. I'm getting annoyed."

[0321] In this case, the processing flow is as follows.

[0322] 1. Data Collection:

[0323] The server collects these posts and stores them in a database.

[0324] 2. Learn the secret language:

[0325] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0326] 3. Analysis of the jargon:

[0327] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0328] 4. Analysis by emotion engine:

[0329] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0330] 5. Identifying risky users:

[0331] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0332] 6. Notice:

[0333] The server notifies the administrator of the results and prompts them to take appropriate action.

[0334] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[0335] Example 2

[0336] Next, a description will be given of 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."

[0337] It is difficult to detect in real time when users are using slang on the Internet, and to recognize the users' emotions and evaluate their overall risk level. In addition, existing systems cannot integrate slang detection and emotion analysis, identify risky users based on the results, and promptly notify administrators, which results in delayed response.

[0338] 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.

[0339] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for recognizing emotions from user text data using an emotion engine, means for evaluating user risk levels based on the analysis results of the emotion engine and slang usage patterns, and means for notifying an administrator of information about high-risk users. This makes it possible to effectively detect fraudulent activities and risky behavior on the Internet and respond quickly.

[0340] "Internet-generated text data" refers to text information posted by users on online platforms such as websites, social networking sites, message boards, and blogs.

[0341] "Collection methods" refer to the processes and tools used to obtain text data from multiple sources on the Internet.

[0342] A "generative AI model" is an artificial intelligence model that has been trained using machine learning algorithms to perform a specific task, in this case code word detection.

[0343] "Training" is the process of using collected data to train a generative AI model and improve the model's performance.

[0344] A "jargon" is a word or expression used within a particular group or community that is only understood by a select few.

[0345] "Detection" is the process of using trained generative AI models to find jargon and specific patterns in text data.

[0346] "Usage patterns" are data that show how frequently and in what contexts a particular jargon is used.

[0347] "Context" refers to the context or situation before and after the use of jargon in text data.

[0348] "Means of analysis" refers to the process of closely examining the collected data, jargon usage patterns, and context to understand their meaning and intent.

[0349] "Danger level" is an assessment that indicates how dangerous the target user is based on text data, slang usage patterns, and the results of sentiment analysis.

[0350] An "emotion engine" is a tool or model for identifying user emotions from text data and analyzing their emotional tone and variance.

[0351] "Means of recognition" refers to the process of identifying emotions in text data using an emotion engine and obtaining analysis results.

[0352] "High-risk users" are users who are deemed to be at high risk based on the use of slang and the results of sentiment analysis.

[0353] "Means of notification" refers to methods and tools used to inform administrators about high-risk users.

[0354] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is mainly composed of a server, and performs a series of processes including data collection, analysis, evaluation, and notification.

[0355] First, the server collects text data from various sources on the Internet (social media, bulletin boards, blogs, etc.). This collection is done using the Python language to obtain data via API. For example, a request is sent to the social media API to collect text data. This is done using the Python requests library. The collected data is temporarily saved in a buffer and then stored in a MySQL database. A regular scheduling process (e.g., a cron job) is performed to collect new data in real time.

[0356] Next, the server uses the collected text data to train a generative AI model. It uses OpenAI's GPT-3 to train the model, which uses a process to learn specific slang terms and their contexts. For example, it learns that the slang term "crystal" means stimulant. It uses OpenAI's libraries for this training.

[0357] Using a trained generative AI model, the server detects slang in newly collected text data. The model analyzes specific words and phrases to determine whether they are slang, and then performs detailed analysis of the context of the detected slang to determine whether it is related to illegal activity.

[0358] Furthermore, the server uses an emotion engine to recognize emotions from the user's text data. This emotion engine uses Azure's Text Analytics API and can analyze the emotional tone and changes in the posted content. For example, it can recognize the user's excited state from the word "irritated."

[0359] The server comprehensively evaluates the emotion engine's analysis results and the usage patterns of slang to assess the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of use of slang, and emotional fluctuations. Based on this, it is determined that a particular user is high risk.

[0360] Finally, the server notifies the administrator of users who are assessed as high risk via email or the administration screen, allowing the administrator to take prompt action.

[0361] Specific examples

[0362] For example, suppose there is a social media post like the one below.

[0363] User A: "Bring me the crystal tonight, at the place I told you about."

[0364] User B: "Be careful. Come early. I'm getting annoyed."

[0365] In this case, the processing flow is as follows.

[0366] 1. Data Collection:

[0367] The server collects these posts and stores them in a database.

[0368] 2. Learn the secret language:

[0369] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0370] 3. Analysis of the jargon:

[0371] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0372] 4. Analysis by emotion engine:

[0373] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0374] 5. Identifying risky users:

[0375] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0376] 6. Notice:

[0377] The server notifies the administrator of the results and prompts them to take appropriate action.

[0378] Prompt Sentence Examples

[0379] For example, use the following prompt for your generative AI model:

[0380] "Detect whether the text data contains slang and determine its context. Furthermore, analyze the sentiment of the post and assess its overall risk level. The text data is shown below."

[0381] In this way, this system combines code word detection and sentiment analysis to comprehensively analyze risky behavior on the Internet and respond quickly.

[0382] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0383] Step 1:

[0384] Data collection

[0385] The server collects text data from social networking sites, message boards, blogs, etc. on the Internet.

[0386] As a concrete example, we will use the Python language and the requests library to retrieve data from a social networking service API.

[0387] Input: Text data from the internet

[0388] Output: Collected text data (saved in a buffer)

[0389] Specific behavior:

[0390] The server sends a request to the SNS API and retrieves the text data.

[0391] The server temporarily stores the acquired data in a buffer.

[0392] Step 2:

[0393] Database storage

[0394] The server periodically stores the data in the buffer in a database.

[0395] Input: Text data in a buffer

[0396] Output: Text data stored in a database

[0397] Specific behavior:

[0398] The server takes the data from the buffer and stores it in a MySQL database.

[0399] The server uses the mysql-connector-python library for database connections.

[0400] Step 3:

[0401] Learning slang

[0402] The server feeds the collected text data back into a generative AI model (e.g., GPT-3) to learn slang and the context in which it is used.

[0403] Input: Collected text data

[0404] Output: A trained generative AI model

[0405] Specific behavior:

[0406] The server uses OpenAI's API to input text data into the generative AI model.

[0407] The generative AI model learns specific jargon and its context.

[0408] Step 4:

[0409] Detecting slang

[0410] The server uses a trained generative AI model to detect slang in newly acquired text data.

[0411] Input: New text data

[0412] Output: Detected slang words and their context information

[0413] Specific behavior:

[0414] The server uses a generative AI model to analyze new text data and detect slang.

[0415] The server also analyzes the context of the detected code words to determine if they relate to illegal activity.

[0416] Step 5:

[0417] Emotion analysis

[0418] The server uses an emotion engine to recognize emotions from the user's text data.

[0419] Input: User's text data

[0420] Output: Emotion analysis results (e.g. anger, stress, etc.)

[0421] Specific behavior:

[0422] The server uses the Azure Text Analytics API and other tools to analyze the emotional tone and fluctuations of the posts.

[0423] The server evaluates the user's emotional state based on the acquired emotional data.

[0424] Step 6:

[0425] Risk Assessment

[0426] The server combines the analysis results of the emotion engine with the usage patterns of slang to assess the user's risk level.

[0427] Input: Sentiment analysis results, slang usage patterns

[0428] Output: User risk assessment

[0429] Specific behavior:

[0430] The server calculates the risk level by combining the results of the emotion analysis with the frequency of appearance of slang.

[0431] The server uses this assessment to identify high-risk users.

[0432] Step 7:

[0433] notification

[0434] The server notifies the administrator of information about users who are assessed as high risk.

[0435] Input: High-risk user information

[0436] Output:Notify administrator

[0437] Specific behavior:

[0438] The server uses SMTP to send email notifications to the administrator.

[0439] The server also displays the information on the management screen so that the administrator can respond quickly.

[0440] (Application example 2)

[0441] Next, a description will be given of Application 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."

[0442] There is a need to detect signs of illegal or fraudulent activity on the Internet early and take appropriate action. However, the use of slang or coded language makes detection extremely difficult. Furthermore, when identifying potentially dangerous users, it is necessary not only to detect slang but also to consider the user's emotions and the context of their behavior. Currently, there is a lack of systems that can perform such comprehensive analysis in real time.

[0443] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for detecting slang and performing sentiment analysis using the generative AI model, means for comprehensively evaluating the results of the sentiment analysis and the slang usage patterns to calculate the risk level, means for identifying risky users based on the evaluation results, and means for notifying an administrator of information about identified risky users. This makes it possible to comprehensively evaluate potentially risky users by combining slang detection and sentiment analysis and quickly address them.

[0444] "Internet-generated text data" refers to text content posted by users on online platforms such as websites, social media platforms, chat applications, blogs, and message boards.

[0445] "Means of collection" refers to collection mechanisms that use functions for efficiently capturing text data from various sources on the Internet, such as web crawlers and APIs (application programming interfaces).

[0446] A "generative AI model" is an algorithm that has been trained using machine learning and deep learning techniques to perform a specific task (in this case, slang detection and sentiment analysis).

[0447] "Training means" refers to the process of training a generative AI model using large amounts of text data, which includes data preprocessing, model training, evaluation, and tuning.

[0448] "Code word detection" refers to the ability to use a trained generative AI model to identify specific coded words or slang within text.

[0449] "Means for performing emotion analysis" refers to the analytical process for determining a user's emotional state (e.g., joy, anger, sadness, surprise, etc.) from text data and the generative AI model used for that analysis.

[0450] "Means for analyzing context" refers to the ability to perform a detailed analysis of the usage of slang and the surrounding context, and determine what that context indicates.

[0451] "Means for assessing risk" refers to a function that comprehensively considers slang usage patterns and the results of sentiment analysis, and expresses in numerical terms or ratings how potentially dangerous a particular user is.

[0452] "Means for identifying risky users" refers to a function that identifies which users pose a high risk based on the assessed risk level and lists those users.

[0453] "Means of notifying administrators" refers to functions for promptly communicating information about identified risky users to administrators, including email notifications, alert systems, dashboard displays, etc.

[0454] "Means of collection in real time" refers to the ability to instantly capture newly generated text data and keep it in a state where it can be analyzed in near real time.

[0455] "Means for saving as a dictionary" refers to the function of saving the usage patterns and meanings of slang that have been learned in a database or file format, and reusing them for subsequent analysis.

[0456] The present invention provides a system for identifying risky users through the detection of slang and analysis of user sentiment, and for dealing with such users appropriately. The following describes in detail the embodiments of the present invention.

[0457] 1. Overall structure

[0458] The server runs a series of processes to collect, analyze, and evaluate text data generated on the internet. The required hardware includes a web crawler and a high-performance server with APIs to collect the data. It also uses deep learning frameworks such as TensorFlow and PyTorch to run generative AI models for sentiment analysis and slang detection.

[0459] 2. Data Collection Phase

[0460] The server collects text data from social media, bulletin boards, blogs, etc. In this process, data is collected periodically using a web crawler or API, temporarily saved in a buffer, and then stored in a database. A scheduling process is set up to collect new data in real time.

[0461] 3. Secret Language Learning Phase

[0462] The server uses the collected text data to train a generative AI model, which then learns specific slang terms and the context in which they are used. For example, if "crystal" is a slang term for stimulants, the model learns their usage patterns from the dataset.

[0463] 4. The jargon analysis phase

[0464] Using a trained generative AI model, the server detects slang in newly collected text data. The model identifies slang used in specific contexts and determines whether the context is related to illegal or fraudulent activity.

[0465] 5. Analysis phase using emotion engine

[0466] The server uses a sentiment analysis engine to recognize emotions from the user's text data. This engine analyzes the emotional tone and fluctuations of the posts. For example, if the post contains the expression "irritated," it will recognize that the user is in an excited state.

[0467] 6. Risky user identification phase

[0468] The server comprehensively evaluates the results of the emotion analysis and the usage patterns of slang to calculate the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of slang use, and emotional fluctuations. If slang is used with tension or excitement, the risk level is rated high.

[0469] 7. Notification Phase

[0470] The server identifies users who are assessed as high risk and notifies administrators of this information in the form of email, dashboard display, alerts, etc.

[0471] Specific examples

[0472] For example, suppose there is a social media post like the one below.

[0473] "Bring me some crystal tonight, at the place I told you about. I'm getting annoyed."

[0474] When this post is entered into the system, the server processes it as follows:

[0475] 1. The server collects the posts and stores them in a database.

[0476] 2. A trained generative AI model recognizes that "crystal" is slang for stimulants.

[0477] 3. The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0478] 4. The server recognizes from the text "I'm irritated" that the user is in an excited state.

[0479] 5. The server evaluates the user's risk level based on the results of emotion analysis and the pattern of slang usage, and determines that the user is at high risk.

[0480] 6. The server notifies the administrator of the result and prompts them to take appropriate measures.

[0481] Prompt Sentence Examples

[0482] "Analyze the emotions and jargon contained in the given text: 'Bring me some crystal tonight, in the place I mentioned earlier. I'm frustrated.'"

[0483] This system will enable early identification of potentially dangerous users on the Internet and enable rapid response.

[0484] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0485] Step 1:

[0486] The server collects text data generated on the Internet. Specifically, it uses a web crawler or API to obtain text data from social media sites, message boards, blogs, etc. The input is the target URL or API endpoint, and the output is the obtained text data. The obtained text data is temporarily saved in a buffer and then stored in a database.

[0487] Step 2:

[0488] The server uses the collected text data to train a generative AI model that learns slang. Specifically, the collected data is preprocessed (text cleaning, tokenization, etc.) and then input into the generative AI model. The input is the preprocessed text data, and the output is a trained generative AI model. This model learns specific slang and its usage context.

[0489] Step 3:

[0490] The server uses the trained generative AI model to detect slang words in newly collected text data. Specifically, the collected text data is input into the trained model to determine whether it contains slang words. The input is the new text data, and the output is the detected slang words and their location information.

[0491] Step 4:

[0492] The server analyzes the usage patterns and context of the detected slang to evaluate the meaning behind it. Specifically, it performs contextual analysis to determine whether the detected slang is related to illegal or fraudulent activity. The input is the slang and its surrounding context, and the output is the analysis result.

[0493] Step 5:

[0494] The server uses an emotion analysis engine to recognize emotions from the user's text data. Specifically, the text data is input into the emotion analysis engine to determine the user's emotional state (e.g., anger, joy, fear, etc.). The input is the text data, and the output is the determined emotion label and its score.

[0495] Step 6:

[0496] The server comprehensively evaluates the results of the sentiment analysis and the usage patterns of slang to calculate the user's risk level. Specifically, it calculates an overall risk score by taking into account the frequency of slang usage, emotional fluctuations, frequency of posting, etc. The input is the results of the sentiment analysis and slang analysis, and the output is a risk score.

[0497] Step 7:

[0498] The server identifies risky users based on the assessed risk level and notifies the administrator of this information. Specifically, the server communicates the information to the administrator via email or an alert on the dashboard. The input is the risk level score, and the output is a notification message.

[0499] 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.

[0500] 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> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) 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.

[0501] 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.

[0502] [Second embodiment]

[0503] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0504] 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.

[0505] 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).

[0506] 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.

[0507] 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.

[0508] 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).

[0509] 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.

[0510] 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.

[0511] 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.

[0512] 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.

[0513] 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.

[0514] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0515] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[0516] 1. Data Collection Phase

[0517] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[0518] 2. Secret Language Learning Phase

[0519] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[0520] 3. The jargon analysis phase

[0521] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[0522] 4. Identifying risky users

[0523] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[0524] Specific examples

[0525] For example, suppose there is a social media post like the one below.

[0526] User A: "Can you bring me some crystals tonight?"

[0527] User B: "Okay. In the usual place."

[0528] In this case, the server performs the following process.

[0529] 1. Data Collection:

[0530] The server retrieves these posts from the internet and stores them in a database.

[0531] 2. Learn the secret language:

[0532] A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[0533] 3. Analysis of the jargon:

[0534] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[0535] 4. Identifying risky users:

[0536] The server analyzes the past posting history and interaction patterns of User A and User B and assesses that there is a high possibility that they are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[0537] In this way, this system can quickly identify accounts that may be involved in criminal activities through a series of processes that collect and analyze text data on the Internet, thereby improving the safety of the Internet environment.

[0538] The processing flow will be explained below.

[0539] Step 1:

[0540] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[0541] Step 2:

[0542] The server stores text data obtained from various internet sources such as social networking sites, message boards, and blogs in a temporary buffer. The collected data includes information about the poster (user ID, posting time, etc.).

[0543] Step 3:

[0544] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[0545] Step 4:

[0546] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[0547] Step 5:

[0548] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[0549] Step 6:

[0550] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[0551] Step 7:

[0552] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[0553] Step 8:

[0554] The server uses the generative AI model and the slang dictionary to analyze new data for the presence of slang, and once identified, analyzes the context in which the slang is used.

[0555] Step 9:

[0556] The server evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[0557] Step 10:

[0558] The server creates a profile of each poster using the code, including their posting history, the content they post, and their patterns of interaction with other users.

[0559] Step 11:

[0560] The server uses a generative AI model to assess the riskiness of user profiles, taking into account factors such as posting frequency, use of slang, and association with other high-risk users.

[0561] Step 12:

[0562] Based on the evaluation results, the server identifies high-risk users. If necessary, the administrator is notified of the identified risky users and appropriate measures are taken.

[0563] Example 1

[0564] Next, a description will be given of 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."

[0565] In recent years, the use of slang suggesting criminal activity on social networking sites, blogs, and bulletin boards on the Internet has led to an increase in illegal transactions and activities. However, these slang terms cannot be easily detected using conventional filtering technology, making it difficult to quickly and accurately identify criminal activity using slang terms. Therefore, there is a need for a system that can effectively detect slang terms on the Internet and prevent criminal activity by analyzing the context of their use.

[0566] 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.

[0567] In this invention, the server includes means for collecting text data generated on the Internet, means for temporarily saving the collected text data in a buffer and storing it in a database, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing the usage context of the slang in the newly collected text data, means for analyzing the usage patterns and context of the slang and evaluating the risk level of the account being analyzed, and means for identifying risky users based on the evaluation results and notifying an administrator as necessary. This makes it possible to quickly and accurately detect slang suggestive of criminal activity on the Internet and analyze the usage context to identify potentially risky users.

[0568] The "Internet" is a collection of globally connected computer networks and an infrastructure for information exchange and communication.

[0569] "Text data" refers to data composed of characters, such as sentences, words, and phrases, in a format that humans can read and write.

[0570] "Collecting" refers to the act of selecting and acquiring data based on specific criteria or filters and putting it in one place.

[0571] A "buffer" is a storage area for temporarily storing data, and is used to improve the efficiency of data transfer and processing.

[0572] A "database" is a system for efficiently storing, searching, and managing structured data, and includes software such as MySQL and MongoDB.

[0573] A "generative AI model" is an artificial intelligence model that is trained using machine learning algorithms to perform specific tasks, such as natural language processing or image recognition.

[0574] "Training" refers to the act of using collected data to train a generative AI model and improve its ability to perform a specific task.

[0575] "Arguments" are special words or phrases that are only understood by certain groups or in certain contexts, and often have meanings that are not known to the general public.

[0576] "Detect" refers to the act of finding specific patterns, words, phrases, etc. in data.

[0577] "Context" refers to the context or situation in which a word or phrase is used, including background information needed to understand its meaning or intent.

[0578] "Behavioral patterns" refer to the actions and tendencies that a particular user repeats on the Internet, and by analyzing these, it is possible to predict that person's behavior.

[0579] "Risk level" is an evaluation indicator that indicates the likelihood that a particular account or behavior is related to criminal or illegal activity.

[0580] "Notifying" refers to the act of communicating specific information or results to a third party, including alerting a system administrator.

[0581] MODE FOR CARRYING OUT THE INVENTION

[0582] This invention is a system that collects text data generated on the Internet, detects slang in new text data using a generative AI model that learns slang, and analyzes the context of the slang to assess its risk level. Specific embodiments are described below.

[0583] System Configuration

[0584] The system of the present invention consists of the following main components:

[0585] 1. Data Collection Methods

[0586] 2. Data Buffers and Databases

[0587] 3. Generative AI models and their training methods

[0588] 4. Idiom detection and context analysis methods

[0589] 5. Risk Assessment Tools

[0590] 6. Administrator notification method

[0591] Hardware and software used

[0592] Hardware: Servers (general computer servers), network infrastructure

[0593] Software: SNS API, database software (e.g., MySQL, MongoDB), generative AI models (e.g., Transformer, BERT)

[0594] Specific operation explanation

[0595] Data collection

[0596] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. Specifically, it uses social media APIs to set specific keyword filters and retrieves posts that match the criteria. For example, it uses prompts like the following:

[0597] "Get recent posts that match a specific keyword"

[0598] The acquired data is temporarily stored in a buffer and then stored in a database.

[0599] Learning slang

[0600] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques such as Transformer and BERT, to learn specific jargon and its context. For example, it learns that the word "crystal" means stimulant in a specific context.

[0601] Code detection and context analysis

[0602] The trained generative AI model is then used to analyze newly collected text data. The server scans the text for words and phrases to detect whether they contain slang. Specifically, it uses prompts like:

[0603] "Scan specific words and phrases in new data to detect slang"

[0604] Additionally, the context in which the slang is used is analyzed to determine whether it is related to illegal activity.

[0605] Risk assessment and administrator notification

[0606] The server evaluates the behavioral patterns of accounts that use a lot of slang based on the analysis results. The evaluation includes the frequency of posts, interaction patterns with other accounts, and past posting history. This allows it to identify potentially dangerous users and evaluate their risk level using a number or category. Based on the evaluation results, it notifies administrators as necessary. The notification includes details such as potential illegal activity, slang used, and past behavioral patterns. For example, the following notifications may be sent:

[0607] "User A frequently uses the code word 'crystal' and has been identified as a high-risk user."

[0608] Specific examples

[0609] For example, consider the following social media post:

[0610] User A: "Can you bring me some crystals tonight?"

[0611] User B: "Okay. In the usual place."

[0612] The process flow in this case is as follows:

[0613] 1. Data collection: The server retrieves these posts, stores them in a buffer, and then stores them in a database.

[0614] 2. Slang learning: The generative AI model learns that the word "crystal" is slang.

[0615] 3. Slang analysis: Detect the word "crystal" in the newly collected data and analyze its context.

[0616] 4. Identifying risky users: The server analyzes the behavioral patterns of users A and B and notifies the administrator that they are high-risk users.

[0617] In this way, this system can quickly and accurately analyze text data generated on the Internet and identify potentially dangerous users, thereby improving the safety of the Internet environment.

[0618] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0619] Step 1: Data collection

[0620] The server uses the APIs of social media, bulletin boards, blogs, etc. to set specific keyword filters and periodically collect posts that match the conditions. For example, posts that match the keyword "crystal" are obtained through the social media API. The input is text data obtained through the social media API, and the output is data temporarily stored in a buffer. Specifically, the server makes an API call and saves the obtained data in a buffer in memory.

[0621] Step 2: Temporarily save to the data buffer

[0622] The server temporarily stores the collected text data in a buffer. The input is text data collected from social media and bulletin boards, and the output is the data in the buffer that has been temporarily stored. This temporary storage allows the integrity of the data to be checked before it is subsequently stored in the database. Specifically, the server performs a process to insert the data into a data structure in memory.

[0623] Step 3: Store in the database

[0624] The server takes the text data from the buffer and stores it in the database. The input is the text data in the buffer, and the output is the data stored in the database. Specifically, it issues an INSERT query to a database system such as MySQL or MongoDB.

[0625] Step 4: Training the generative AI model

[0626] The server uses the text data collected from the database to train a generative AI model. The input is the text data in the database, and the output is the trained generative AI model. Specifically, it uses AI models such as Transformer and BERT to learn specific jargon and its context through supervised learning.

[0627] Step 5: Collect new data

[0628] The server processes newly collected text data in real time. The input is text data from social media and bulletin boards collected in real time, and the output is real-time data in a buffer. Specifically, the server continuously calls the API to collect new post data.

[0629] Step 6: Scan for jargon

[0630] The server uses a trained generative AI model to scan new data for slang. The input is text data collected in real time, and the output is a list of detected slang words. Specifically, the inference function of the generative AI model is used to identify slang words within the text.

[0631] Step 7: Parsing the Context

[0632] The server analyzes the context of slang usage. The input is a list of detected slang words and their surrounding text, and the output is the context analysis results. Specifically, it uses natural language processing technology to analyze the context in which the slang words are used.

[0633] Step 8: Evaluate behavioral patterns

[0634] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang. The input is the context analysis results and the user's posting history data, and the output is a risk assessment result. Specifically, the server analyzes posting frequency, interaction patterns, past posting history, etc. as evaluation indicators.

[0635] Step 9: Risk Assessment

[0636] The server evaluates the user's risk level using a numerical value or category based on the behavioral pattern evaluation. The input is the behavioral pattern evaluation result, and the output is the risk evaluation result and a list of high-risk users. Specifically, a risk scoring algorithm is used to quantify the user's risk level.

[0637] Step 10: Notify administrators

[0638] The server identifies high-risk users and notifies administrators as necessary. The input is the risk assessment result, and the output is a notification message sent to the administrator. Specifically, email and dashboard notifications are sent when certain conditions are met.

[0639] (Application example 1)

[0640] Next, a description will be given of Application 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."

[0641] Conventional systems detect and assess the risk of slang only in text data on the Internet, but do not address slang contained in audio or image data. This makes it difficult to monitor dangerous behavior in real time, and is insufficient in situations where a rapid response is required, such as in public places or events.

[0642] 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.

[0643] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for notifying users of the analyzed information in real time, means for collecting and analyzing audio data, and means for collecting and analyzing image data. This makes it possible to detect dangerous slang in real time from various data including audio and image data.

[0644] A "server" is a computer system that collects and analyzes data over a network.

[0645] The "Internet" is a huge information network that spans the entire world and is a platform for sending and receiving various types of information.

[0646] "Text data" refers to information composed of character strings, including social media posts, blog articles, and message board posts on the Internet.

[0647] A "generative AI model" is a model that uses artificial intelligence technology to learn patterns and features from text data.

[0648] "Training" is the process of feeding a generative AI model large amounts of data to teach it specific patterns and characteristics.

[0649] "Argot" is a word or phrase that has a different meaning from its usual meaning among a particular group or in a particular situation.

[0650] "Usage patterns" refer to how frequently and in what contexts particular words or phrases are used.

[0651] "Context" refers to the context or situation in which a particular word or phrase is used.

[0652] "Danger level" is an indicator of the likelihood that a particular account is involved in criminal activity, derived from the pattern and context of code word usage.

[0653] "User" is a term that refers to a person or account that posts information on the Internet.

[0654] "Real-time" is a term that indicates the extremely short time between the occurrence of an event and the information being processed and analyzed.

[0655] "Notification" is a means by which the system informs the user of analysis results and important information.

[0656] "Voice data" refers to data that records human speech or sound in digital format.

[0657] "Image data" is visual data that records light information in digital format.

[0658] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[0659] Data Collection Phase

[0660] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[0661] Secret language learning phase

[0662] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[0663] The jargon analysis phase

[0664] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[0665] Risky user identification phase

[0666] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[0667] Real-time notification and multi-data analysis functions

[0668] The analyzed information is then sent to users in real time via smart glasses or other wearable devices. The system also collects audio and image data for similar analysis. For example, speech recognition technology is used to convert speech at events into text and analyze high-risk slang. It also uses Google Cloud Vision API to extract text from images and perform similar analysis.

[0669] Hardware and Software Used

[0670] This system uses the following hardware and software:

[0671] Server: A central processing unit that performs data collection and analysis.

[0672] SNS API: An API used to collect data.

[0673] Generative AI model: A model for learning and analyzing jargon from data.

[0674] Speech recognition libraries (e.g. speech_recognition): Used to convert voice data into text.

[0675] Google Cloud Vision API: Used to extract text from image data.

[0676] Google Cloud Natural Language API: Used to parse text data and extract entities.

[0677] Specific examples

[0678] For example, consider the following social media post:

[0679] User A: "Can you bring me some crystals tonight?"

[0680] User B: "Okay. In the usual place."

[0681] In this case, the server performs the following process.

[0682] Data collection: The server retrieves these posts from the internet and stores them in a database.

[0683] Slang learning: A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[0684] Code word analysis: The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[0685] Identifying risky users: Analyzing the past posting history and interaction patterns of User A and User B, the server assesses the likelihood that the two are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[0686] Example prompt for a generative AI model:

[0687] Check to see if any of the following words appear in today's conversations: crystal, popcorn, white circle.

[0688] If you find any, assess whether they may be related to illegal activity.

[0689] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0690] Step 1:

[0691] The server collects text data generated on the Internet.

[0692] Input: Text data from the internet

[0693] Output: Collected text data

[0694] How it works: The server uses the SNS API to set up a specific keyword filter and periodically retrieves posts that match the filter. This data is temporarily stored in a buffer and then stored in a database.

[0695] Step 2:

[0696] The server uses the collected text data to train a generative AI model.

[0697] Input: Collected text data

[0698] Output: A trained generative AI model

[0699] How it works: The server extracts text data from a database and uses natural language processing (NLP) techniques to train a generative AI model to understand jargon and its context.

[0700] Step 3:

[0701] The server uses the trained generative AI model to detect jargon in new text data.

[0702] Input: New text data, trained generative AI model

[0703] Output: Detected slang

[0704] How it works: The server analyzes newly collected text data in real time and uses a generative AI model to detect whether it contains slang.

[0705] Step 4:

[0706] The server analyzes the usage patterns and context of slang and assesses the risk level of the account being analyzed.

[0707] Input: Detected slang, new text data

[0708] Output: Account risk assessment results

[0709] Specific operation: The server evaluates the frequency of slang use, the context of use, interaction patterns with other accounts, etc., and evaluates the risk level of each account using a number or category.

[0710] Step 5:

[0711] The server identifies risky users based on the evaluation results.

[0712] Input: Account risk assessment results

[0713] Output: Identified risky users

[0714] Specific operation: The server lists accounts that are assessed as high risk and prepares to notify the administrator of the list.

[0715] Step 6:

[0716] The server notifies the user of the analyzed information in real time.

[0717] Input: Analysis results

[0718] Output: User notification

[0719] Specific operation: The server generates a warning message based on the analysis results and notifies wearable devices such as smart glasses in real time.

[0720] Step 7:

[0721] The device collects and analyzes audio data at events and public places.

[0722] Input: Audio data

[0723] Output: Text data

[0724] Specific operation: The device (smart glasses, etc.) converts audio data recorded at events or public places into text using the SpeechRecognition library.

[0725] Step 8:

[0726] The terminal collects and analyzes image data.

[0727] Input: Image data

[0728] Output: Text data

[0729] Specific operation: The device (smart glasses, etc.) converts the captured image into text using the Google Cloud Vision API and sends the result to the server.

[0730] The above are the processing steps of the present invention, which make it possible to detect and notify dangerous slang from a variety of data sources in real time.

[0731] 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.

[0732] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is built around a server and performs a series of processes, including data collection, analysis, evaluation, and notification.

[0733] 1. Data Collection Phase

[0734] The server collects text data from various sources on the Internet (e.g., social media, bulletin boards, blogs, etc.). The collected data is temporarily stored in a buffer and then stored in a database. A regular scheduling process is performed to collect new data in real time.

[0735] 2. Secret Language Learning Phase

[0736] The server uses the collected text data to train a generative AI model, which learns specific slang terms and the context in which they are used. For example, it learns from the training data that "crystal" means stimulant in certain contexts.

[0737] 3. The jargon analysis phase

[0738] Using the trained generative AI model, the server can detect code words in newly collected text data, and if code words are present, analyze their context in detail to determine whether they are related to illegal activity.

[0739] 4. Analysis phase using emotion engine

[0740] The server uses an emotion engine to recognize emotions from users' text data. This engine can analyze the emotional tone and changes in the content of posts. For example, if a post contains anger or stress, it can determine that this is likely related to the use of slang.

[0741] 5. Risky user identification phase

[0742] The server evaluates the user's risk level based on the analysis results of the emotion engine and the usage pattern of slang. The evaluation takes into account the frequency of posts, frequency of slang use, and emotional fluctuations. For example, if the context in which slang is used is accompanied by tension or excitement, the risk level is evaluated as high.

[0743] 6. Notification Phase

[0744] The server identifies users who are assessed as high risk and notifies the administrator, who uses this information to guide appropriate action.

[0745] Specific examples

[0746] For example, suppose there is a social media post like the one below.

[0747] User A: "Bring me the crystal tonight, at the place I told you about."

[0748] User B: "Be careful. Come early. I'm getting annoyed."

[0749] In this case, the processing flow is as follows.

[0750] 1. Data Collection:

[0751] The server collects these posts and stores them in a database.

[0752] 2. Learn the secret language:

[0753] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0754] 3. Analysis of the jargon:

[0755] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0756] 4. Analysis by emotion engine:

[0757] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0758] 5. Identifying risky users:

[0759] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0760] 6. Notice:

[0761] The server notifies the administrator of the results and prompts them to take appropriate action.

[0762] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[0763] The processing flow will be explained below.

[0764] Step 1:

[0765] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[0766] Step 2:

[0767] The server stores text data obtained from various internet sources such as social media, message boards, and blogs in a temporary buffer. The collected data also includes information about the poster (user ID, posting time, etc.).

[0768] Step 3:

[0769] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[0770] Step 4:

[0771] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[0772] Step 5:

[0773] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[0774] Step 6:

[0775] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[0776] Step 7:

[0777] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[0778] Step 8:

[0779] The server uses the generative AI model and slang dictionary to analyze new data to determine whether it contains slang. If slang is identified, it analyzes the context and surrounding words.

[0780] Step 9:

[0781] The server then carefully evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[0782] Step 10:

[0783] The server uses an emotion engine to recognize the poster's emotions from the text data. This emotion analysis includes the tone of the text, the intensity of the emotion, and the type of emotion (joy, anger, sadness, etc.).

[0784] Step 11:

[0785] The server combines the identified emotional information with the slang usage patterns and stores them in a database, making the correlation between emotional fluctuations and slang usage available for future analysis.

[0786] Step 12:

[0787] The server evaluates the results of the emotion engine and the usage patterns of slang comprehensively to assess the risk level of each user. For example, if the context in which slang is used is emotionally charged, the risk level is evaluated as high.

[0788] Step 13:

[0789] The server identifies high-risk users based on the evaluation results, which include the behavioral patterns, past posts, and emotional fluctuations of the identified risky users.

[0790] Step 14:

[0791] The server notifies the administrator of users who are rated as highly dangerous. The notification includes the user's ID, the content of their posts, the risk assessment result, and the related sentiment analysis result.

[0792] Specific examples

[0793] For example, suppose there is a social media post like the one below.

[0794] User A: "Bring me the crystal tonight, at the place I told you about."

[0795] User B: "Be careful. Come early. I'm getting annoyed."

[0796] In this case, the processing flow is as follows.

[0797] 1. Data Collection:

[0798] The server collects these posts and stores them in a database.

[0799] 2. Learn the secret language:

[0800] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0801] 3. Analysis of the jargon:

[0802] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0803] 4. Analysis by emotion engine:

[0804] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0805] 5. Identifying risky users:

[0806] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0807] 6. Notice:

[0808] The server notifies the administrator of the results and prompts them to take appropriate action.

[0809] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[0810] Example 2

[0811] Next, a description will be given of 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."

[0812] It is difficult to detect in real time when users are using slang on the Internet, and to recognize the users' emotions and evaluate their overall risk level. In addition, existing systems cannot integrate slang detection and emotion analysis, identify risky users based on the results, and promptly notify administrators, which results in delayed response.

[0813] 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.

[0814] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for recognizing emotions from user text data using an emotion engine, means for evaluating user risk levels based on the analysis results of the emotion engine and slang usage patterns, and means for notifying an administrator of information about high-risk users. This makes it possible to effectively detect fraudulent activities and risky behavior on the Internet and respond quickly.

[0815] "Internet-generated text data" refers to text information posted by users on online platforms such as websites, social networking sites, message boards, and blogs.

[0816] "Collection methods" refer to the processes and tools used to obtain text data from multiple sources on the Internet.

[0817] A "generative AI model" is an artificial intelligence model that has been trained using machine learning algorithms to perform a specific task, in this case code word detection.

[0818] "Training" is the process of using collected data to train a generative AI model and improve the model's performance.

[0819] A "jargon" is a word or expression used within a particular group or community that is only understood by a select few.

[0820] "Detection" is the process of using trained generative AI models to find jargon and specific patterns in text data.

[0821] "Usage patterns" are data that show how frequently and in what contexts a particular jargon is used.

[0822] "Context" refers to the context or situation before and after the use of jargon in text data.

[0823] "Means of analysis" refers to the process of closely examining the collected data, jargon usage patterns, and context to understand their meaning and intent.

[0824] "Danger level" is an assessment that indicates how dangerous the target user is based on text data, slang usage patterns, and the results of sentiment analysis.

[0825] An "emotion engine" is a tool or model for identifying user emotions from text data and analyzing their emotional tone and variance.

[0826] "Means of recognition" refers to the process of identifying emotions in text data using an emotion engine and obtaining analysis results.

[0827] "High-risk users" are users who are deemed to be at high risk based on the use of slang and the results of sentiment analysis.

[0828] "Means of notification" refers to methods and tools used to inform administrators about high-risk users.

[0829] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is mainly composed of a server, and performs a series of processes including data collection, analysis, evaluation, and notification.

[0830] First, the server collects text data from various sources on the Internet (social media, bulletin boards, blogs, etc.). This collection is done using the Python language to obtain data via API. For example, a request is sent to the social media API to collect text data. This is done using the Python requests library. The collected data is temporarily saved in a buffer and then stored in a MySQL database. A regular scheduling process (e.g., a cron job) is performed to collect new data in real time.

[0831] Next, the server uses the collected text data to train a generative AI model. It uses OpenAI's GPT-3 to train the model, which uses a process to learn specific slang terms and their contexts. For example, it learns that the slang term "crystal" means stimulant. It uses OpenAI's libraries for this training.

[0832] Using a trained generative AI model, the server detects slang in newly collected text data. The model analyzes specific words and phrases to determine whether they are slang, and then performs detailed analysis of the context of the detected slang to determine whether it is related to illegal activity.

[0833] Furthermore, the server uses an emotion engine to recognize emotions from the user's text data. This emotion engine uses Azure's Text Analytics API and can analyze the emotional tone and changes in the posted content. For example, it can recognize the user's excited state from the word "irritated."

[0834] The server comprehensively evaluates the emotion engine's analysis results and the usage patterns of slang to assess the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of use of slang, and emotional fluctuations. Based on this, it is determined that a particular user is high risk.

[0835] Finally, the server notifies the administrator of users who are assessed as high risk via email or the administration screen, allowing the administrator to take prompt action.

[0836] Specific examples

[0837] For example, suppose there is a social media post like the one below.

[0838] User A: "Bring me the crystal tonight, at the place I told you about."

[0839] User B: "Be careful. Come early. I'm getting annoyed."

[0840] In this case, the processing flow is as follows.

[0841] 1. Data Collection:

[0842] The server collects these posts and stores them in a database.

[0843] 2. Learn the secret language:

[0844] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[0845] 3. Analysis of the jargon:

[0846] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0847] 4. Analysis by emotion engine:

[0848] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[0849] 5. Identifying risky users:

[0850] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[0851] 6. Notice:

[0852] The server notifies the administrator of the results and prompts them to take appropriate action.

[0853] Prompt Sentence Examples

[0854] For example, use the following prompt for your generative AI model:

[0855] "Detect whether the text data contains slang and determine its context. Furthermore, analyze the sentiment of the post and assess its overall risk level. The text data is shown below."

[0856] In this way, this system combines code word detection and sentiment analysis to comprehensively analyze risky behavior on the Internet and respond quickly.

[0857] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0858] Step 1:

[0859] Data collection

[0860] The server collects text data from social networking sites, message boards, blogs, etc. on the Internet.

[0861] As a concrete example, we will use the Python language and the requests library to retrieve data from a social networking service API.

[0862] Input: Text data from the internet

[0863] Output: Collected text data (saved in a buffer)

[0864] Specific behavior:

[0865] The server sends a request to the SNS API and retrieves the text data.

[0866] The server temporarily stores the acquired data in a buffer.

[0867] Step 2:

[0868] Database storage

[0869] The server periodically stores the data in the buffer in a database.

[0870] Input: Text data in a buffer

[0871] Output: Text data stored in a database

[0872] Specific behavior:

[0873] The server takes the data from the buffer and stores it in a MySQL database.

[0874] The server uses the mysql-connector-python library for database connections.

[0875] Step 3:

[0876] Learning slang

[0877] The server feeds the collected text data back into a generative AI model (e.g., GPT-3) to learn slang and the context in which it is used.

[0878] Input: Collected text data

[0879] Output: A trained generative AI model

[0880] Specific behavior:

[0881] The server uses OpenAI's API to input text data into the generative AI model.

[0882] The generative AI model learns specific jargon and its context.

[0883] Step 4:

[0884] Detecting slang

[0885] The server uses a trained generative AI model to detect slang in newly acquired text data.

[0886] Input: New text data

[0887] Output: Detected slang words and their context information

[0888] Specific behavior:

[0889] The server uses a generative AI model to analyze new text data and detect slang.

[0890] The server also analyzes the context of the detected code words to determine if they relate to illegal activity.

[0891] Step 5:

[0892] Emotion analysis

[0893] The server uses an emotion engine to recognize emotions from the user's text data.

[0894] Input: User's text data

[0895] Output: Emotion analysis results (e.g. anger, stress, etc.)

[0896] Specific behavior:

[0897] The server uses the Azure Text Analytics API and other tools to analyze the emotional tone and fluctuations of the posts.

[0898] The server evaluates the user's emotional state based on the acquired emotional data.

[0899] Step 6:

[0900] Risk Assessment

[0901] The server combines the analysis results of the emotion engine with the usage patterns of slang to assess the user's risk level.

[0902] Input: Sentiment analysis results, slang usage patterns

[0903] Output: User risk assessment

[0904] Specific behavior:

[0905] The server calculates the risk level by combining the results of the emotion analysis with the frequency of appearance of slang.

[0906] The server uses this assessment to identify high-risk users.

[0907] Step 7:

[0908] notification

[0909] The server notifies the administrator of information about users who are assessed as high risk.

[0910] Input: High-risk user information

[0911] Output:Notify administrator

[0912] Specific behavior:

[0913] The server uses SMTP to send email notifications to the administrator.

[0914] The server also displays the information on the management screen so that the administrator can respond quickly.

[0915] (Application example 2)

[0916] Next, a description will be given of Application 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."

[0917] There is a need to detect signs of illegal or fraudulent activity on the Internet early and take appropriate action. However, the use of slang or coded language makes detection extremely difficult. Furthermore, when identifying potentially dangerous users, it is necessary not only to detect slang but also to consider the user's emotions and the context of their behavior. Currently, there is a lack of systems that can perform such comprehensive analysis in real time.

[0918] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for detecting slang and performing sentiment analysis using the generative AI model, means for comprehensively evaluating the results of the sentiment analysis and the slang usage patterns to calculate the risk level, means for identifying risky users based on the evaluation results, and means for notifying an administrator of information about identified risky users. This makes it possible to comprehensively evaluate potentially risky users by combining slang detection and sentiment analysis and quickly address them.

[0919] "Internet-generated text data" refers to text content posted by users on online platforms such as websites, social media platforms, chat applications, blogs, and message boards.

[0920] "Means of collection" refers to collection mechanisms that use functions for efficiently capturing text data from various sources on the Internet, such as web crawlers and APIs (application programming interfaces).

[0921] A "generative AI model" is an algorithm that has been trained using machine learning and deep learning techniques to perform a specific task (in this case, slang detection and sentiment analysis).

[0922] "Training means" refers to the process of training a generative AI model using large amounts of text data, which includes data preprocessing, model training, evaluation, and tuning.

[0923] "Code word detection" refers to the ability to use a trained generative AI model to identify specific coded words or slang within text.

[0924] "Means for performing emotion analysis" refers to the analytical process for determining a user's emotional state (e.g., joy, anger, sadness, surprise, etc.) from text data and the generative AI model used for that analysis.

[0925] "Means for analyzing context" refers to the ability to perform a detailed analysis of the usage of slang and the surrounding context, and determine what that context indicates.

[0926] "Means for assessing risk" refers to a function that comprehensively considers slang usage patterns and the results of sentiment analysis, and expresses in numerical terms or ratings how potentially dangerous a particular user is.

[0927] "Means for identifying risky users" refers to a function that identifies which users pose a high risk based on the assessed risk level and lists those users.

[0928] "Means of notifying administrators" refers to functions for promptly communicating information about identified risky users to administrators, including email notifications, alert systems, dashboard displays, etc.

[0929] "Means of collection in real time" refers to the ability to instantly capture newly generated text data and keep it in a state where it can be analyzed in near real time.

[0930] "Means for saving as a dictionary" refers to the function of saving the usage patterns and meanings of slang that have been learned in a database or file format, and reusing them for subsequent analysis.

[0931] The present invention provides a system for identifying risky users through the detection of slang and analysis of user sentiment, and for dealing with such users appropriately. The following describes in detail the embodiments of the present invention.

[0932] 1. Overall structure

[0933] The server runs a series of processes to collect, analyze, and evaluate text data generated on the internet. The required hardware includes a web crawler and a high-performance server with APIs to collect the data. It also uses deep learning frameworks such as TensorFlow and PyTorch to run generative AI models for sentiment analysis and slang detection.

[0934] 2. Data Collection Phase

[0935] The server collects text data from social media, bulletin boards, blogs, etc. In this process, data is collected periodically using a web crawler or API, temporarily saved in a buffer, and then stored in a database. A scheduling process is set up to collect new data in real time.

[0936] 3. Secret Language Learning Phase

[0937] The server uses the collected text data to train a generative AI model, which then learns specific slang terms and the context in which they are used. For example, if "crystal" is a slang term for stimulants, the model learns their usage patterns from the dataset.

[0938] 4. The jargon analysis phase

[0939] Using a trained generative AI model, the server detects slang in newly collected text data. The model identifies slang used in specific contexts and determines whether the context is related to illegal or fraudulent activity.

[0940] 5. Analysis phase using emotion engine

[0941] The server uses a sentiment analysis engine to recognize emotions from the user's text data. This engine analyzes the emotional tone and fluctuations of the posts. For example, if the post contains the expression "irritated," it will recognize that the user is in an excited state.

[0942] 6. Risky user identification phase

[0943] The server comprehensively evaluates the results of the emotion analysis and the usage patterns of slang to calculate the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of slang use, and emotional fluctuations. If slang is used with tension or excitement, the risk level is rated high.

[0944] 7. Notification Phase

[0945] The server identifies users who are assessed as high risk and notifies administrators of this information in the form of email, dashboard display, alerts, etc.

[0946] Specific examples

[0947] For example, suppose there is a social media post like the one below.

[0948] "Bring me some crystal tonight, at the place I told you about. I'm getting annoyed."

[0949] When this post is entered into the system, the server processes it as follows:

[0950] 1. The server collects the posts and stores them in a database.

[0951] 2. A trained generative AI model recognizes that "crystal" is slang for stimulants.

[0952] 3. The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[0953] 4. The server recognizes from the text "I'm irritated" that the user is in an excited state.

[0954] 5. The server evaluates the user's risk level based on the results of emotion analysis and the pattern of slang usage, and determines that the user is at high risk.

[0955] 6. The server notifies the administrator of the result and prompts them to take appropriate measures.

[0956] Prompt Sentence Examples

[0957] "Analyze the emotions and jargon contained in the given text: 'Bring me some crystal tonight, in the place I mentioned earlier. I'm frustrated.'"

[0958] This system will enable early identification of potentially dangerous users on the Internet and enable rapid response.

[0959] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0960] Step 1:

[0961] The server collects text data generated on the Internet. Specifically, it uses a web crawler or API to obtain text data from social media sites, message boards, blogs, etc. The input is the target URL or API endpoint, and the output is the obtained text data. The obtained text data is temporarily saved in a buffer and then stored in a database.

[0962] Step 2:

[0963] The server uses the collected text data to train a generative AI model that learns slang. Specifically, the collected data is preprocessed (text cleaning, tokenization, etc.) and then input into the generative AI model. The input is the preprocessed text data, and the output is a trained generative AI model. This model learns specific slang and its usage context.

[0964] Step 3:

[0965] The server uses the trained generative AI model to detect slang words in newly collected text data. Specifically, the collected text data is input into the trained model to determine whether it contains slang words. The input is the new text data, and the output is the detected slang words and their location information.

[0966] Step 4:

[0967] The server analyzes the usage patterns and context of the detected slang to evaluate the meaning behind it. Specifically, it performs contextual analysis to determine whether the detected slang is related to illegal or fraudulent activity. The input is the slang and its surrounding context, and the output is the analysis result.

[0968] Step 5:

[0969] The server uses an emotion analysis engine to recognize emotions from the user's text data. Specifically, the text data is input into the emotion analysis engine to determine the user's emotional state (e.g., anger, joy, fear, etc.). The input is the text data, and the output is the determined emotion label and its score.

[0970] Step 6:

[0971] The server comprehensively evaluates the results of the sentiment analysis and the usage patterns of slang to calculate the user's risk level. Specifically, it calculates an overall risk score by taking into account the frequency of slang usage, emotional fluctuations, frequency of posting, etc. The input is the results of the sentiment analysis and slang analysis, and the output is a risk score.

[0972] Step 7:

[0973] The server identifies risky users based on the assessed risk level and notifies the administrator of this information. Specifically, the server communicates the information to the administrator via email or an alert on the dashboard. The input is the risk level score, and the output is a notification message.

[0974] 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.

[0975] 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> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) 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.

[0976] 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.

[0977] [Third embodiment]

[0978] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0979] 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.

[0980] 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).

[0981] 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.

[0982] 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.

[0983] 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).

[0984] 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.

[0985] 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.

[0986] 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.

[0987] 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.

[0988] 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.

[0989] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[0990] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[0991] 1. Data Collection Phase

[0992] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[0993] 2. Secret Language Learning Phase

[0994] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[0995] 3. The jargon analysis phase

[0996] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[0997] 4. Identifying risky users

[0998] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[0999] Specific examples

[1000] For example, suppose there is a social media post like the one below.

[1001] User A: "Can you bring me some crystals tonight?"

[1002] User B: "Okay. In the usual place."

[1003] In this case, the server performs the following process.

[1004] 1. Data Collection:

[1005] The server retrieves these posts from the internet and stores them in a database.

[1006] 2. Learn the secret language:

[1007] A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[1008] 3. Analysis of the jargon:

[1009] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[1010] 4. Identifying risky users:

[1011] The server analyzes the past posting history and interaction patterns of User A and User B and assesses that there is a high possibility that they are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[1012] In this way, this system can quickly identify accounts that may be involved in criminal activities through a series of processes that collect and analyze text data on the Internet, thereby improving the safety of the Internet environment.

[1013] The processing flow will be explained below.

[1014] Step 1:

[1015] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[1016] Step 2:

[1017] The server stores text data obtained from various internet sources such as social networking sites, message boards, and blogs in a temporary buffer. The collected data includes information about the poster (user ID, posting time, etc.).

[1018] Step 3:

[1019] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[1020] Step 4:

[1021] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[1022] Step 5:

[1023] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[1024] Step 6:

[1025] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[1026] Step 7:

[1027] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[1028] Step 8:

[1029] The server uses the generative AI model and the slang dictionary to analyze new data for the presence of slang, and once identified, analyzes the context in which the slang is used.

[1030] Step 9:

[1031] The server evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[1032] Step 10:

[1033] The server creates a profile of each poster using the code, including their posting history, the content they post, and their patterns of interaction with other users.

[1034] Step 11:

[1035] The server uses a generative AI model to assess the riskiness of user profiles, taking into account factors such as posting frequency, use of slang, and association with other high-risk users.

[1036] Step 12:

[1037] Based on the evaluation results, the server identifies high-risk users. If necessary, the administrator is notified of the identified risky users and appropriate measures are taken.

[1038] Example 1

[1039] Next, a description will be given of 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."

[1040] In recent years, the use of slang suggesting criminal activity on social networking sites, blogs, and bulletin boards on the Internet has led to an increase in illegal transactions and activities. However, these slang terms cannot be easily detected using conventional filtering technology, making it difficult to quickly and accurately identify criminal activity using slang terms. Therefore, there is a need for a system that can effectively detect slang terms on the Internet and prevent criminal activity by analyzing the context of their use.

[1041] 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.

[1042] In this invention, the server includes means for collecting text data generated on the Internet, means for temporarily saving the collected text data in a buffer and storing it in a database, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing the usage context of the slang in the newly collected text data, means for analyzing the usage patterns and context of the slang and evaluating the risk level of the account being analyzed, and means for identifying risky users based on the evaluation results and notifying an administrator as necessary. This makes it possible to quickly and accurately detect slang suggestive of criminal activity on the Internet and analyze the usage context to identify potentially risky users.

[1043] The "Internet" is a collection of globally connected computer networks and an infrastructure for information exchange and communication.

[1044] "Text data" refers to data composed of characters, such as sentences, words, and phrases, in a format that humans can read and write.

[1045] "Collecting" refers to the act of selecting and acquiring data based on specific criteria or filters and putting it in one place.

[1046] A "buffer" is a storage area for temporarily storing data, and is used to improve the efficiency of data transfer and processing.

[1047] A "database" is a system for efficiently storing, searching, and managing structured data, and includes software such as MySQL and MongoDB.

[1048] A "generative AI model" is an artificial intelligence model that is trained using machine learning algorithms to perform specific tasks, such as natural language processing or image recognition.

[1049] "Training" refers to the act of using collected data to train a generative AI model and improve its ability to perform a specific task.

[1050] "Arguments" are special words or phrases that are only understood by certain groups or in certain contexts, and often have meanings that are not known to the general public.

[1051] "Detect" refers to the act of finding specific patterns, words, phrases, etc. in data.

[1052] "Context" refers to the context or situation in which a word or phrase is used, including background information needed to understand its meaning or intent.

[1053] "Behavioral patterns" refer to the actions and tendencies that a particular user repeats on the Internet, and by analyzing these, it is possible to predict that person's behavior.

[1054] "Risk level" is an evaluation indicator that indicates the likelihood that a particular account or behavior is related to criminal or illegal activity.

[1055] "Notifying" refers to the act of communicating specific information or results to a third party, including alerting a system administrator.

[1056] MODE FOR CARRYING OUT THE INVENTION

[1057] This invention is a system that collects text data generated on the Internet, detects slang in new text data using a generative AI model that learns slang, and analyzes the context of the slang to assess its risk level. Specific embodiments are described below.

[1058] System Configuration

[1059] The system of the present invention consists of the following main components:

[1060] 1. Data Collection Methods

[1061] 2. Data Buffers and Databases

[1062] 3. Generative AI models and their training methods

[1063] 4. Idiom detection and context analysis methods

[1064] 5. Risk Assessment Tools

[1065] 6. Administrator notification method

[1066] Hardware and software used

[1067] Hardware: Servers (general computer servers), network infrastructure

[1068] Software: SNS API, database software (e.g., MySQL, MongoDB), generative AI models (e.g., Transformer, BERT)

[1069] Specific operation explanation

[1070] Data collection

[1071] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. Specifically, it uses social media APIs to set specific keyword filters and retrieves posts that match the criteria. For example, it uses prompts like the following:

[1072] "Get recent posts that match a specific keyword"

[1073] The acquired data is temporarily stored in a buffer and then stored in a database.

[1074] Learning slang

[1075] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques such as Transformer and BERT, to learn specific jargon and its context. For example, it learns that the word "crystal" means stimulant in a specific context.

[1076] Code detection and context analysis

[1077] The trained generative AI model is then used to analyze newly collected text data. The server scans the text for words and phrases to detect whether they contain slang. Specifically, it uses prompts like:

[1078] "Scan specific words and phrases in new data to detect slang"

[1079] Additionally, the context in which the slang is used is analyzed to determine whether it is related to illegal activity.

[1080] Risk assessment and administrator notification

[1081] The server evaluates the behavioral patterns of accounts that use a lot of slang based on the analysis results. The evaluation includes the frequency of posts, interaction patterns with other accounts, and past posting history. This allows it to identify potentially dangerous users and evaluate their risk level using a number or category. Based on the evaluation results, it notifies administrators as necessary. The notification includes details such as potential illegal activity, slang used, and past behavioral patterns. For example, the following notifications may be sent:

[1082] "User A frequently uses the code word 'crystal' and has been identified as a high-risk user."

[1083] Specific examples

[1084] For example, consider the following social media post:

[1085] User A: "Can you bring me some crystals tonight?"

[1086] User B: "Okay. In the usual place."

[1087] The process flow in this case is as follows:

[1088] 1. Data collection: The server retrieves these posts, stores them in a buffer, and then stores them in a database.

[1089] 2. Slang learning: The generative AI model learns that the word "crystal" is slang.

[1090] 3. Slang analysis: Detect the word "crystal" in the newly collected data and analyze its context.

[1091] 4. Identifying risky users: The server analyzes the behavioral patterns of users A and B and notifies the administrator that they are high-risk users.

[1092] In this way, this system can quickly and accurately analyze text data generated on the Internet and identify potentially dangerous users, thereby improving the safety of the Internet environment.

[1093] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1094] Step 1: Data collection

[1095] The server uses the APIs of social media, bulletin boards, blogs, etc. to set specific keyword filters and periodically collect posts that match the conditions. For example, posts that match the keyword "crystal" are obtained through the social media API. The input is text data obtained through the social media API, and the output is data temporarily stored in a buffer. Specifically, the server makes an API call and saves the obtained data in a buffer in memory.

[1096] Step 2: Temporarily save to the data buffer

[1097] The server temporarily stores the collected text data in a buffer. The input is text data collected from social media and bulletin boards, and the output is the data in the buffer that has been temporarily stored. This temporary storage allows the integrity of the data to be checked before it is subsequently stored in the database. Specifically, the server performs a process to insert the data into a data structure in memory.

[1098] Step 3: Store in the database

[1099] The server takes the text data from the buffer and stores it in the database. The input is the text data in the buffer, and the output is the data stored in the database. Specifically, it issues an INSERT query to a database system such as MySQL or MongoDB.

[1100] Step 4: Training the generative AI model

[1101] The server uses the text data collected from the database to train a generative AI model. The input is the text data in the database, and the output is the trained generative AI model. Specifically, it uses AI models such as Transformer and BERT to learn specific jargon and its context through supervised learning.

[1102] Step 5: Collect new data

[1103] The server processes newly collected text data in real time. The input is text data from social media and bulletin boards collected in real time, and the output is real-time data in a buffer. Specifically, the server continuously calls the API to collect new post data.

[1104] Step 6: Scan for jargon

[1105] The server uses a trained generative AI model to scan new data for slang. The input is text data collected in real time, and the output is a list of detected slang words. Specifically, the inference function of the generative AI model is used to identify slang words within the text.

[1106] Step 7: Parsing the Context

[1107] The server analyzes the context of slang usage. The input is a list of detected slang words and their surrounding text, and the output is the context analysis results. Specifically, it uses natural language processing technology to analyze the context in which the slang words are used.

[1108] Step 8: Evaluate behavioral patterns

[1109] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang. The input is the context analysis results and the user's posting history data, and the output is a risk assessment result. Specifically, the server analyzes posting frequency, interaction patterns, past posting history, etc. as evaluation indicators.

[1110] Step 9: Risk Assessment

[1111] The server evaluates the user's risk level using a numerical value or category based on the behavioral pattern evaluation. The input is the behavioral pattern evaluation result, and the output is the risk evaluation result and a list of high-risk users. Specifically, a risk scoring algorithm is used to quantify the user's risk level.

[1112] Step 10: Notify administrators

[1113] The server identifies high-risk users and notifies administrators as necessary. The input is the risk assessment result, and the output is a notification message sent to the administrator. Specifically, email and dashboard notifications are sent when certain conditions are met.

[1114] (Application example 1)

[1115] Next, a description will be given of Application 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."

[1116] Conventional systems detect and assess the risk of slang only in text data on the Internet, but do not address slang contained in audio or image data. This makes it difficult to monitor dangerous behavior in real time, and is insufficient in situations where a rapid response is required, such as in public places or events.

[1117] 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.

[1118] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for notifying users of the analyzed information in real time, means for collecting and analyzing audio data, and means for collecting and analyzing image data. This makes it possible to detect dangerous slang in real time from various data including audio and image data.

[1119] A "server" is a computer system that collects and analyzes data over a network.

[1120] The "Internet" is a huge information network that spans the entire world and is a platform for sending and receiving various types of information.

[1121] "Text data" refers to information composed of character strings, including social media posts, blog articles, and message board posts on the Internet.

[1122] A "generative AI model" is a model that uses artificial intelligence technology to learn patterns and features from text data.

[1123] "Training" is the process of feeding a generative AI model large amounts of data to teach it specific patterns and characteristics.

[1124] "Argot" is a word or phrase that has a different meaning from its usual meaning among a particular group or in a particular situation.

[1125] "Usage patterns" refer to how frequently and in what contexts particular words or phrases are used.

[1126] "Context" refers to the context or situation in which a particular word or phrase is used.

[1127] "Danger level" is an indicator of the likelihood that a particular account is involved in criminal activity, derived from the pattern and context of code word usage.

[1128] "User" is a term that refers to a person or account that posts information on the Internet.

[1129] "Real-time" is a term that indicates the extremely short time between the occurrence of an event and the information being processed and analyzed.

[1130] "Notification" is a means by which the system informs the user of analysis results and important information.

[1131] "Voice data" refers to data that records human speech or sound in digital format.

[1132] "Image data" is visual data that records light information in digital format.

[1133] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[1134] Data Collection Phase

[1135] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[1136] Secret language learning phase

[1137] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[1138] The jargon analysis phase

[1139] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[1140] Risky user identification phase

[1141] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[1142] Real-time notification and multi-data analysis functions

[1143] The analyzed information is then sent to users in real time via smart glasses or other wearable devices. The system also collects audio and image data for similar analysis. For example, speech recognition technology is used to convert speech at events into text and analyze high-risk slang. It also uses Google Cloud Vision API to extract text from images and perform similar analysis.

[1144] Hardware and Software Used

[1145] This system uses the following hardware and software:

[1146] Server: A central processing unit that performs data collection and analysis.

[1147] SNS API: An API used to collect data.

[1148] Generative AI model: A model for learning and analyzing jargon from data.

[1149] Speech recognition libraries (e.g. speech_recognition): Used to convert voice data into text.

[1150] Google Cloud Vision API: Used to extract text from image data.

[1151] Google Cloud Natural Language API: Used to parse text data and extract entities.

[1152] Specific examples

[1153] For example, consider the following social media post:

[1154] User A: "Can you bring me some crystals tonight?"

[1155] User B: "Okay. In the usual place."

[1156] In this case, the server performs the following process.

[1157] Data collection: The server retrieves these posts from the internet and stores them in a database.

[1158] Slang learning: A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[1159] Code word analysis: The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[1160] Identifying risky users: Analyzing the past posting history and interaction patterns of User A and User B, the server assesses the likelihood that the two are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[1161] Example prompt for a generative AI model:

[1162] Check to see if any of the following words appear in today's conversations: crystal, popcorn, white circle.

[1163] If you find any, assess whether they may be related to illegal activity.

[1164] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1165] Step 1:

[1166] The server collects text data generated on the Internet.

[1167] Input: Text data from the internet

[1168] Output: Collected text data

[1169] How it works: The server uses the SNS API to set up a specific keyword filter and periodically retrieves posts that match the filter. This data is temporarily stored in a buffer and then stored in a database.

[1170] Step 2:

[1171] The server uses the collected text data to train a generative AI model.

[1172] Input: Collected text data

[1173] Output: A trained generative AI model

[1174] How it works: The server extracts text data from a database and uses natural language processing (NLP) techniques to train a generative AI model to understand jargon and its context.

[1175] Step 3:

[1176] The server uses the trained generative AI model to detect jargon in new text data.

[1177] Input: New text data, trained generative AI model

[1178] Output: Detected slang

[1179] How it works: The server analyzes newly collected text data in real time and uses a generative AI model to detect whether it contains slang.

[1180] Step 4:

[1181] The server analyzes the usage patterns and context of slang and assesses the risk level of the account being analyzed.

[1182] Input: Detected slang, new text data

[1183] Output: Account risk assessment results

[1184] Specific operation: The server evaluates the frequency of slang use, the context of use, interaction patterns with other accounts, etc., and evaluates the risk level of each account using a number or category.

[1185] Step 5:

[1186] The server identifies risky users based on the evaluation results.

[1187] Input: Account risk assessment results

[1188] Output: Identified risky users

[1189] Specific operation: The server lists accounts that are assessed as high risk and prepares to notify the administrator of the list.

[1190] Step 6:

[1191] The server notifies the user of the analyzed information in real time.

[1192] Input: Analysis results

[1193] Output: User notification

[1194] Specific operation: The server generates a warning message based on the analysis results and notifies wearable devices such as smart glasses in real time.

[1195] Step 7:

[1196] The device collects and analyzes audio data at events and public places.

[1197] Input: Audio data

[1198] Output: Text data

[1199] Specific operation: The device (smart glasses, etc.) converts audio data recorded at events or public places into text using the SpeechRecognition library.

[1200] Step 8:

[1201] The terminal collects and analyzes image data.

[1202] Input: Image data

[1203] Output: Text data

[1204] Specific operation: The device (smart glasses, etc.) converts the captured image into text using the Google Cloud Vision API and sends the result to the server.

[1205] The above are the processing steps of the present invention, which make it possible to detect and notify dangerous slang from a variety of data sources in real time.

[1206] 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.

[1207] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is built around a server and performs a series of processes, including data collection, analysis, evaluation, and notification.

[1208] 1. Data Collection Phase

[1209] The server collects text data from various sources on the Internet (e.g., social media, bulletin boards, blogs, etc.). The collected data is temporarily stored in a buffer and then stored in a database. A regular scheduling process is performed to collect new data in real time.

[1210] 2. Secret Language Learning Phase

[1211] The server uses the collected text data to train a generative AI model, which learns specific slang terms and the context in which they are used. For example, it learns from the training data that "crystal" means stimulant in certain contexts.

[1212] 3. The jargon analysis phase

[1213] Using the trained generative AI model, the server can detect code words in newly collected text data, and if code words are present, analyze their context in detail to determine whether they are related to illegal activity.

[1214] 4. Analysis phase using emotion engine

[1215] The server uses an emotion engine to recognize emotions from users' text data. This engine can analyze the emotional tone and changes in the content of posts. For example, if a post contains anger or stress, it can determine that this is likely related to the use of slang.

[1216] 5. Risky user identification phase

[1217] The server evaluates the user's risk level based on the analysis results of the emotion engine and the usage pattern of slang. The evaluation takes into account the frequency of posts, frequency of slang use, and emotional fluctuations. For example, if the context in which slang is used is accompanied by tension or excitement, the risk level is evaluated as high.

[1218] 6. Notification Phase

[1219] The server identifies users who are assessed as high risk and notifies the administrator, who uses this information to guide appropriate action.

[1220] Specific examples

[1221] For example, suppose there is a social media post like the one below.

[1222] User A: "Bring me the crystal tonight, at the place I told you about."

[1223] User B: "Be careful. Come early. I'm getting annoyed."

[1224] In this case, the processing flow is as follows.

[1225] 1. Data Collection:

[1226] The server collects these posts and stores them in a database.

[1227] 2. Learn the secret language:

[1228] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1229] 3. Analysis of the jargon:

[1230] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1231] 4. Analysis by emotion engine:

[1232] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1233] 5. Identifying risky users:

[1234] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1235] 6. Notice:

[1236] The server notifies the administrator of the results and prompts them to take appropriate action.

[1237] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[1238] The processing flow will be explained below.

[1239] Step 1:

[1240] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[1241] Step 2:

[1242] The server stores text data obtained from various internet sources such as social media, message boards, and blogs in a temporary buffer. The collected data also includes information about the poster (user ID, posting time, etc.).

[1243] Step 3:

[1244] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[1245] Step 4:

[1246] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[1247] Step 5:

[1248] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[1249] Step 6:

[1250] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[1251] Step 7:

[1252] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[1253] Step 8:

[1254] The server uses the generative AI model and slang dictionary to analyze new data to determine whether it contains slang. If slang is identified, it analyzes the context and surrounding words.

[1255] Step 9:

[1256] The server then carefully evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[1257] Step 10:

[1258] The server uses an emotion engine to recognize the poster's emotions from the text data. This emotion analysis includes the tone of the text, the intensity of the emotion, and the type of emotion (joy, anger, sadness, etc.).

[1259] Step 11:

[1260] The server combines the identified emotional information with the slang usage patterns and stores them in a database, making the correlation between emotional fluctuations and slang usage available for future analysis.

[1261] Step 12:

[1262] The server evaluates the results of the emotion engine and the usage patterns of slang comprehensively to assess the risk level of each user. For example, if the context in which slang is used is emotionally charged, the risk level is evaluated as high.

[1263] Step 13:

[1264] The server identifies high-risk users based on the evaluation results, which include the behavioral patterns, past posts, and emotional fluctuations of the identified risky users.

[1265] Step 14:

[1266] The server notifies the administrator of users who are rated as highly dangerous. The notification includes the user's ID, the content of their posts, the risk assessment result, and the related sentiment analysis result.

[1267] Specific examples

[1268] For example, suppose there is a social media post like the one below.

[1269] User A: "Bring me the crystal tonight, at the place I told you about."

[1270] User B: "Be careful. Come early. I'm getting annoyed."

[1271] In this case, the processing flow is as follows.

[1272] 1. Data Collection:

[1273] The server collects these posts and stores them in a database.

[1274] 2. Learn the secret language:

[1275] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1276] 3. Analysis of the jargon:

[1277] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1278] 4. Analysis by emotion engine:

[1279] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1280] 5. Identifying risky users:

[1281] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1282] 6. Notice:

[1283] The server notifies the administrator of the results and prompts them to take appropriate action.

[1284] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[1285] Example 2

[1286] Next, a description will be given of 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."

[1287] It is difficult to detect in real time when users are using slang on the Internet, and to recognize the users' emotions and evaluate their overall risk level. In addition, existing systems cannot integrate slang detection and emotion analysis, identify risky users based on the results, and promptly notify administrators, which results in delayed response.

[1288] 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.

[1289] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for recognizing emotions from user text data using an emotion engine, means for evaluating user risk levels based on the analysis results of the emotion engine and slang usage patterns, and means for notifying an administrator of information about high-risk users. This makes it possible to effectively detect fraudulent activities and risky behavior on the Internet and respond quickly.

[1290] "Internet-generated text data" refers to text information posted by users on online platforms such as websites, social networking sites, message boards, and blogs.

[1291] "Collection methods" refer to the processes and tools used to obtain text data from multiple sources on the Internet.

[1292] A "generative AI model" is an artificial intelligence model that has been trained using machine learning algorithms to perform a specific task, in this case code word detection.

[1293] "Training" is the process of using collected data to train a generative AI model and improve the model's performance.

[1294] A "jargon" is a word or expression used within a particular group or community that is only understood by a select few.

[1295] "Detection" is the process of using trained generative AI models to find jargon and specific patterns in text data.

[1296] "Usage patterns" are data that show how frequently and in what contexts a particular jargon is used.

[1297] "Context" refers to the context or situation before and after the use of jargon in text data.

[1298] "Means of analysis" refers to the process of closely examining the collected data, jargon usage patterns, and context to understand their meaning and intent.

[1299] "Danger level" is an assessment that indicates how dangerous the target user is based on text data, slang usage patterns, and the results of sentiment analysis.

[1300] An "emotion engine" is a tool or model for identifying user emotions from text data and analyzing their emotional tone and variance.

[1301] "Means of recognition" refers to the process of identifying emotions in text data using an emotion engine and obtaining analysis results.

[1302] "High-risk users" are users who are deemed to be at high risk based on the use of slang and the results of sentiment analysis.

[1303] "Means of notification" refers to methods and tools used to inform administrators about high-risk users.

[1304] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is mainly composed of a server, and performs a series of processes including data collection, analysis, evaluation, and notification.

[1305] First, the server collects text data from various sources on the Internet (social media, bulletin boards, blogs, etc.). This collection is done using the Python language to obtain data via API. For example, a request is sent to the social media API to collect text data. This is done using the Python requests library. The collected data is temporarily saved in a buffer and then stored in a MySQL database. A regular scheduling process (e.g., a cron job) is performed to collect new data in real time.

[1306] Next, the server uses the collected text data to train a generative AI model. It uses OpenAI's GPT-3 to train the model, which uses a process to learn specific slang terms and their contexts. For example, it learns that the slang term "crystal" means stimulant. It uses OpenAI's libraries for this training.

[1307] Using a trained generative AI model, the server detects slang in newly collected text data. The model analyzes specific words and phrases to determine whether they are slang, and then performs detailed analysis of the context of the detected slang to determine whether it is related to illegal activity.

[1308] Furthermore, the server uses an emotion engine to recognize emotions from the user's text data. This emotion engine uses Azure's Text Analytics API and can analyze the emotional tone and changes in the posted content. For example, it can recognize the user's excited state from the word "irritated."

[1309] The server comprehensively evaluates the emotion engine's analysis results and the usage patterns of slang to assess the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of use of slang, and emotional fluctuations. Based on this, it is determined that a particular user is high risk.

[1310] Finally, the server notifies the administrator of users who are assessed as high risk via email or the administration screen, allowing the administrator to take prompt action.

[1311] Specific examples

[1312] For example, suppose there is a social media post like the one below.

[1313] User A: "Bring me the crystal tonight, at the place I told you about."

[1314] User B: "Be careful. Come early. I'm getting annoyed."

[1315] In this case, the processing flow is as follows.

[1316] 1. Data Collection:

[1317] The server collects these posts and stores them in a database.

[1318] 2. Learn the secret language:

[1319] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1320] 3. Analysis of the jargon:

[1321] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1322] 4. Analysis by emotion engine:

[1323] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1324] 5. Identifying risky users:

[1325] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1326] 6. Notice:

[1327] The server notifies the administrator of the results and prompts them to take appropriate action.

[1328] Prompt Sentence Examples

[1329] For example, use the following prompt for your generative AI model:

[1330] "Detect whether the text data contains slang and determine its context. Furthermore, analyze the sentiment of the post and assess its overall risk level. The text data is shown below."

[1331] In this way, this system combines code word detection and sentiment analysis to comprehensively analyze risky behavior on the Internet and respond quickly.

[1332] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1333] Step 1:

[1334] Data collection

[1335] The server collects text data from social networking sites, message boards, blogs, etc. on the Internet.

[1336] As a concrete example, we will use the Python language and the requests library to retrieve data from a social networking service API.

[1337] Input: Text data from the internet

[1338] Output: Collected text data (saved in a buffer)

[1339] Specific behavior:

[1340] The server sends a request to the SNS API and retrieves the text data.

[1341] The server temporarily stores the acquired data in a buffer.

[1342] Step 2:

[1343] Database storage

[1344] The server periodically stores the data in the buffer in a database.

[1345] Input: Text data in a buffer

[1346] Output: Text data stored in a database

[1347] Specific behavior:

[1348] The server takes the data from the buffer and stores it in a MySQL database.

[1349] The server uses the mysql-connector-python library for database connections.

[1350] Step 3:

[1351] Learning slang

[1352] The server feeds the collected text data back into a generative AI model (e.g., GPT-3) to learn slang and the context in which it is used.

[1353] Input: Collected text data

[1354] Output: A trained generative AI model

[1355] Specific behavior:

[1356] The server uses OpenAI's API to input text data into the generative AI model.

[1357] The generative AI model learns specific jargon and its context.

[1358] Step 4:

[1359] Detecting slang

[1360] The server uses a trained generative AI model to detect slang in newly acquired text data.

[1361] Input: New text data

[1362] Output: Detected slang words and their context information

[1363] Specific behavior:

[1364] The server uses a generative AI model to analyze new text data and detect slang.

[1365] The server also analyzes the context of the detected code words to determine if they relate to illegal activity.

[1366] Step 5:

[1367] Emotion analysis

[1368] The server uses an emotion engine to recognize emotions from the user's text data.

[1369] Input: User's text data

[1370] Output: Emotion analysis results (e.g. anger, stress, etc.)

[1371] Specific behavior:

[1372] The server uses the Azure Text Analytics API and other tools to analyze the emotional tone and fluctuations of the posts.

[1373] The server evaluates the user's emotional state based on the acquired emotional data.

[1374] Step 6:

[1375] Risk Assessment

[1376] The server combines the analysis results of the emotion engine with the usage patterns of slang to assess the user's risk level.

[1377] Input: Sentiment analysis results, slang usage patterns

[1378] Output: User risk assessment

[1379] Specific behavior:

[1380] The server calculates the risk level by combining the results of the emotion analysis with the frequency of appearance of slang.

[1381] The server uses this assessment to identify high-risk users.

[1382] Step 7:

[1383] notification

[1384] The server notifies the administrator of information about users who are assessed as high risk.

[1385] Input: High-risk user information

[1386] Output:Notify administrator

[1387] Specific behavior:

[1388] The server uses SMTP to send email notifications to the administrator.

[1389] The server also displays the information on the management screen so that the administrator can respond quickly.

[1390] (Application example 2)

[1391] Next, a description will be given of Application 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."

[1392] There is a need to detect signs of illegal or fraudulent activity on the Internet early and take appropriate action. However, the use of slang or coded language makes detection extremely difficult. Furthermore, when identifying potentially dangerous users, it is necessary not only to detect slang but also to consider the user's emotions and the context of their behavior. Currently, there is a lack of systems that can perform such comprehensive analysis in real time.

[1393] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for detecting slang and performing sentiment analysis using the generative AI model, means for comprehensively evaluating the results of the sentiment analysis and the slang usage patterns to calculate the risk level, means for identifying risky users based on the evaluation results, and means for notifying an administrator of information about identified risky users. This makes it possible to comprehensively evaluate potentially risky users by combining slang detection and sentiment analysis and quickly address them.

[1394] "Internet-generated text data" refers to text content posted by users on online platforms such as websites, social media platforms, chat applications, blogs, and message boards.

[1395] "Means of collection" refers to collection mechanisms that use functions for efficiently capturing text data from various sources on the Internet, such as web crawlers and APIs (application programming interfaces).

[1396] A "generative AI model" is an algorithm that has been trained using machine learning and deep learning techniques to perform a specific task (in this case, slang detection and sentiment analysis).

[1397] "Training means" refers to the process of training a generative AI model using large amounts of text data, which includes data preprocessing, model training, evaluation, and tuning.

[1398] "Code word detection" refers to the ability to use a trained generative AI model to identify specific coded words or slang within text.

[1399] "Means for performing emotion analysis" refers to the analytical process for determining a user's emotional state (e.g., joy, anger, sadness, surprise, etc.) from text data and the generative AI model used for that analysis.

[1400] "Means for analyzing context" refers to the ability to perform a detailed analysis of the usage of slang and the surrounding context, and determine what that context indicates.

[1401] "Means for assessing risk" refers to a function that comprehensively considers slang usage patterns and the results of sentiment analysis, and expresses in numerical terms or ratings how potentially dangerous a particular user is.

[1402] "Means for identifying risky users" refers to a function that identifies which users pose a high risk based on the assessed risk level and lists those users.

[1403] "Means of notifying administrators" refers to functions for promptly communicating information about identified risky users to administrators, including email notifications, alert systems, dashboard displays, etc.

[1404] "Means of collection in real time" refers to the ability to instantly capture newly generated text data and keep it in a state where it can be analyzed in near real time.

[1405] "Means for saving as a dictionary" refers to the function of saving the usage patterns and meanings of slang that have been learned in a database or file format, and reusing them for subsequent analysis.

[1406] The present invention provides a system for identifying risky users through the detection of slang and analysis of user sentiment, and for dealing with such users appropriately. The following describes in detail the embodiments of the present invention.

[1407] 1. Overall structure

[1408] The server runs a series of processes to collect, analyze, and evaluate text data generated on the internet. The required hardware includes a web crawler and a high-performance server with APIs to collect the data. It also uses deep learning frameworks such as TensorFlow and PyTorch to run generative AI models for sentiment analysis and slang detection.

[1409] 2. Data Collection Phase

[1410] The server collects text data from social media, bulletin boards, blogs, etc. In this process, data is collected periodically using a web crawler or API, temporarily saved in a buffer, and then stored in a database. A scheduling process is set up to collect new data in real time.

[1411] 3. Secret Language Learning Phase

[1412] The server uses the collected text data to train a generative AI model, which then learns specific slang terms and the context in which they are used. For example, if "crystal" is a slang term for stimulants, the model learns their usage patterns from the dataset.

[1413] 4. The jargon analysis phase

[1414] Using a trained generative AI model, the server detects slang in newly collected text data. The model identifies slang used in specific contexts and determines whether the context is related to illegal or fraudulent activity.

[1415] 5. Analysis phase using emotion engine

[1416] The server uses a sentiment analysis engine to recognize emotions from the user's text data. This engine analyzes the emotional tone and fluctuations of the posts. For example, if the post contains the expression "irritated," it will recognize that the user is in an excited state.

[1417] 6. Risky user identification phase

[1418] The server comprehensively evaluates the results of the emotion analysis and the usage patterns of slang to calculate the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of slang use, and emotional fluctuations. If slang is used with tension or excitement, the risk level is rated high.

[1419] 7. Notification Phase

[1420] The server identifies users who are assessed as high risk and notifies administrators of this information in the form of email, dashboard display, alerts, etc.

[1421] Specific examples

[1422] For example, suppose there is a social media post like the one below.

[1423] "Bring me some crystal tonight, at the place I told you about. I'm getting annoyed."

[1424] When this post is entered into the system, the server processes it as follows:

[1425] 1. The server collects the posts and stores them in a database.

[1426] 2. A trained generative AI model recognizes that "crystal" is slang for stimulants.

[1427] 3. The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1428] 4. The server recognizes from the text "I'm irritated" that the user is in an excited state.

[1429] 5. The server evaluates the user's risk level based on the results of emotion analysis and the pattern of slang usage, and determines that the user is at high risk.

[1430] 6. The server notifies the administrator of the result and prompts them to take appropriate measures.

[1431] Prompt Sentence Examples

[1432] "Analyze the emotions and jargon contained in the given text: 'Bring me some crystal tonight, in the place I mentioned earlier. I'm frustrated.'"

[1433] This system will enable early identification of potentially dangerous users on the Internet and enable rapid response.

[1434] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1435] Step 1:

[1436] The server collects text data generated on the Internet. Specifically, it uses a web crawler or API to obtain text data from social media sites, message boards, blogs, etc. The input is the target URL or API endpoint, and the output is the obtained text data. The obtained text data is temporarily saved in a buffer and then stored in a database.

[1437] Step 2:

[1438] The server uses the collected text data to train a generative AI model that learns slang. Specifically, the collected data is preprocessed (text cleaning, tokenization, etc.) and then input into the generative AI model. The input is the preprocessed text data, and the output is a trained generative AI model. This model learns specific slang and its usage context.

[1439] Step 3:

[1440] The server uses the trained generative AI model to detect slang words in newly collected text data. Specifically, the collected text data is input into the trained model to determine whether it contains slang words. The input is the new text data, and the output is the detected slang words and their location information.

[1441] Step 4:

[1442] The server analyzes the usage patterns and context of the detected slang to evaluate the meaning behind it. Specifically, it performs contextual analysis to determine whether the detected slang is related to illegal or fraudulent activity. The input is the slang and its surrounding context, and the output is the analysis result.

[1443] Step 5:

[1444] The server uses an emotion analysis engine to recognize emotions from the user's text data. Specifically, the text data is input into the emotion analysis engine to determine the user's emotional state (e.g., anger, joy, fear, etc.). The input is the text data, and the output is the determined emotion label and its score.

[1445] Step 6:

[1446] The server comprehensively evaluates the results of the sentiment analysis and the usage patterns of slang to calculate the user's risk level. Specifically, it calculates an overall risk score by taking into account the frequency of slang usage, emotional fluctuations, frequency of posting, etc. The input is the results of the sentiment analysis and slang analysis, and the output is a risk score.

[1447] Step 7:

[1448] The server identifies risky users based on the assessed risk level and notifies the administrator of this information. Specifically, the server communicates the information to the administrator via email or an alert on the dashboard. The input is the risk level score, and the output is a notification message.

[1449] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 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.

[1450] 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> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) 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.

[1451] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1452] [Fourth embodiment]

[1453] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1454] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1455] 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).

[1456] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. 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 control target 443 are also connected to the bus 52.

[1457] 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.

[1458] 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).

[1459] 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.

[1460] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1461] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, 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.

[1462] 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.

[1463] 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.

[1464] In the robot 414, the processor 46 performs the reception output process. 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.

[1465] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1466] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[1467] 1. Data Collection Phase

[1468] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[1469] 2. Secret Language Learning Phase

[1470] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[1471] 3. The jargon analysis phase

[1472] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[1473] 4. Identifying risky users

[1474] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[1475] Specific examples

[1476] For example, suppose there is a social media post like the one below.

[1477] User A: "Can you bring me some crystals tonight?"

[1478] User B: "Okay. In the usual place."

[1479] In this case, the server performs the following process.

[1480] 1. Data Collection:

[1481] The server retrieves these posts from the internet and stores them in a database.

[1482] 2. Learn the secret language:

[1483] A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[1484] 3. Analysis of the jargon:

[1485] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[1486] 4. Identifying risky users:

[1487] The server analyzes the past posting history and interaction patterns of User A and User B and assesses that there is a high possibility that they are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[1488] In this way, this system can quickly identify accounts that may be involved in criminal activities through a series of processes that collect and analyze text data on the Internet, thereby improving the safety of the Internet environment.

[1489] The processing flow will be explained below.

[1490] Step 1:

[1491] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[1492] Step 2:

[1493] The server stores text data obtained from various internet sources such as social networking sites, message boards, and blogs in a temporary buffer. The collected data includes information about the poster (user ID, posting time, etc.).

[1494] Step 3:

[1495] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[1496] Step 4:

[1497] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[1498] Step 5:

[1499] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[1500] Step 6:

[1501] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[1502] Step 7:

[1503] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[1504] Step 8:

[1505] The server uses the generative AI model and the slang dictionary to analyze new data for the presence of slang, and once identified, analyzes the context in which the slang is used.

[1506] Step 9:

[1507] The server evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[1508] Step 10:

[1509] The server creates a profile of each poster using the code, including their posting history, the content they post, and their patterns of interaction with other users.

[1510] Step 11:

[1511] The server uses a generative AI model to assess the riskiness of user profiles, taking into account factors such as posting frequency, use of slang, and association with other high-risk users.

[1512] Step 12:

[1513] Based on the evaluation results, the server identifies high-risk users. If necessary, the administrator is notified of the identified risky users and appropriate measures are taken.

[1514] Example 1

[1515] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1516] In recent years, the use of slang suggesting criminal activity on social networking sites, blogs, and bulletin boards on the Internet has led to an increase in illegal transactions and activities. However, these slang terms cannot be easily detected using conventional filtering technology, making it difficult to quickly and accurately identify criminal activity using slang terms. Therefore, there is a need for a system that can effectively detect slang terms on the Internet and prevent criminal activity by analyzing the context of their use.

[1517] 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.

[1518] In this invention, the server includes means for collecting text data generated on the Internet, means for temporarily saving the collected text data in a buffer and storing it in a database, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing the usage context of the slang in the newly collected text data, means for analyzing the usage patterns and context of the slang and evaluating the risk level of the account being analyzed, and means for identifying risky users based on the evaluation results and notifying an administrator as necessary. This makes it possible to quickly and accurately detect slang suggestive of criminal activity on the Internet and analyze the usage context to identify potentially risky users.

[1519] The "Internet" is a collection of globally connected computer networks and an infrastructure for information exchange and communication.

[1520] "Text data" refers to data composed of characters, such as sentences, words, and phrases, in a format that humans can read and write.

[1521] "Collecting" refers to the act of selecting and acquiring data based on specific criteria or filters and putting it in one place.

[1522] A "buffer" is a storage area for temporarily storing data, and is used to improve the efficiency of data transfer and processing.

[1523] A "database" is a system for efficiently storing, searching, and managing structured data, and includes software such as MySQL and MongoDB.

[1524] A "generative AI model" is an artificial intelligence model that is trained using machine learning algorithms to perform specific tasks, such as natural language processing or image recognition.

[1525] "Training" refers to the act of using collected data to train a generative AI model and improve its ability to perform a specific task.

[1526] "Arguments" are special words or phrases that are only understood by certain groups or in certain contexts, and often have meanings that are not known to the general public.

[1527] "Detect" refers to the act of finding specific patterns, words, phrases, etc. in data.

[1528] "Context" refers to the context or situation in which a word or phrase is used, including background information needed to understand its meaning or intent.

[1529] "Behavioral patterns" refer to the actions and tendencies that a particular user repeats on the Internet, and by analyzing these, it is possible to predict that person's behavior.

[1530] "Risk level" is an evaluation indicator that indicates the likelihood that a particular account or behavior is related to criminal or illegal activity.

[1531] "Notifying" refers to the act of communicating specific information or results to a third party, including alerting a system administrator.

[1532] MODE FOR CARRYING OUT THE INVENTION

[1533] This invention is a system that collects text data generated on the Internet, detects slang in new text data using a generative AI model that learns slang, and analyzes the context of the slang to assess its risk level. Specific embodiments are described below.

[1534] System Configuration

[1535] The system of the present invention consists of the following main components:

[1536] 1. Data Collection Methods

[1537] 2. Data Buffers and Databases

[1538] 3. Generative AI models and their training methods

[1539] 4. Idiom detection and context analysis methods

[1540] 5. Risk Assessment Tools

[1541] 6. Administrator notification method

[1542] Hardware and software used

[1543] Hardware: Servers (general computer servers), network infrastructure

[1544] Software: SNS API, database software (e.g., MySQL, MongoDB), generative AI models (e.g., Transformer, BERT)

[1545] Specific operation explanation

[1546] Data collection

[1547] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. Specifically, it uses social media APIs to set specific keyword filters and retrieves posts that match the criteria. For example, it uses prompts like the following:

[1548] "Get recent posts that match a specific keyword"

[1549] The acquired data is temporarily stored in a buffer and then stored in a database.

[1550] Learning slang

[1551] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques such as Transformer and BERT, to learn specific jargon and its context. For example, it learns that the word "crystal" means stimulant in a specific context.

[1552] Code detection and context analysis

[1553] The trained generative AI model is then used to analyze newly collected text data. The server scans the text for words and phrases to detect whether they contain slang. Specifically, it uses prompts like:

[1554] "Scan specific words and phrases in new data to detect slang"

[1555] Additionally, the context in which the slang is used is analyzed to determine whether it is related to illegal activity.

[1556] Risk assessment and administrator notification

[1557] The server evaluates the behavioral patterns of accounts that use a lot of slang based on the analysis results. The evaluation includes the frequency of posts, interaction patterns with other accounts, and past posting history. This allows it to identify potentially dangerous users and evaluate their risk level using a number or category. Based on the evaluation results, it notifies administrators as necessary. The notification includes details such as potential illegal activity, slang used, and past behavioral patterns. For example, the following notifications may be sent:

[1558] "User A frequently uses the code word 'crystal' and has been identified as a high-risk user."

[1559] Specific examples

[1560] For example, consider the following social media post:

[1561] User A: "Can you bring me some crystals tonight?"

[1562] User B: "Okay. In the usual place."

[1563] The process flow in this case is as follows:

[1564] 1. Data collection: The server retrieves these posts, stores them in a buffer, and then stores them in a database.

[1565] 2. Slang learning: The generative AI model learns that the word "crystal" is slang.

[1566] 3. Slang analysis: Detect the word "crystal" in the newly collected data and analyze its context.

[1567] 4. Identifying risky users: The server analyzes the behavioral patterns of users A and B and notifies the administrator that they are high-risk users.

[1568] In this way, this system can quickly and accurately analyze text data generated on the Internet and identify potentially dangerous users, thereby improving the safety of the Internet environment.

[1569] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1570] Step 1: Data collection

[1571] The server uses the APIs of social media, bulletin boards, blogs, etc. to set specific keyword filters and periodically collect posts that match the conditions. For example, posts that match the keyword "crystal" are obtained through the social media API. The input is text data obtained through the social media API, and the output is data temporarily stored in a buffer. Specifically, the server makes an API call and saves the obtained data in a buffer in memory.

[1572] Step 2: Temporarily save to the data buffer

[1573] The server temporarily stores the collected text data in a buffer. The input is text data collected from social media and bulletin boards, and the output is the data in the buffer that has been temporarily stored. This temporary storage allows the integrity of the data to be checked before it is subsequently stored in the database. Specifically, the server performs a process to insert the data into a data structure in memory.

[1574] Step 3: Store in the database

[1575] The server takes the text data from the buffer and stores it in the database. The input is the text data in the buffer, and the output is the data stored in the database. Specifically, it issues an INSERT query to a database system such as MySQL or MongoDB.

[1576] Step 4: Training the generative AI model

[1577] The server uses the text data collected from the database to train a generative AI model. The input is the text data in the database, and the output is the trained generative AI model. Specifically, it uses AI models such as Transformer and BERT to learn specific jargon and its context through supervised learning.

[1578] Step 5: Collect new data

[1579] The server processes newly collected text data in real time. The input is text data from social media and bulletin boards collected in real time, and the output is real-time data in a buffer. Specifically, the server continuously calls the API to collect new post data.

[1580] Step 6: Scan for jargon

[1581] The server uses a trained generative AI model to scan new data for slang. The input is text data collected in real time, and the output is a list of detected slang words. Specifically, the inference function of the generative AI model is used to identify slang words within the text.

[1582] Step 7: Parsing the Context

[1583] The server analyzes the context of slang usage. The input is a list of detected slang words and their surrounding text, and the output is the context analysis results. Specifically, it uses natural language processing technology to analyze the context in which the slang words are used.

[1584] Step 8: Evaluate behavioral patterns

[1585] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang. The input is the context analysis results and the user's posting history data, and the output is a risk assessment result. Specifically, the server analyzes posting frequency, interaction patterns, past posting history, etc. as evaluation indicators.

[1586] Step 9: Risk Assessment

[1587] The server evaluates the user's risk level using a numerical value or category based on the behavioral pattern evaluation. The input is the behavioral pattern evaluation result, and the output is the risk evaluation result and a list of high-risk users. Specifically, a risk scoring algorithm is used to quantify the user's risk level.

[1588] Step 10: Notify administrators

[1589] The server identifies high-risk users and notifies administrators as necessary. The input is the risk assessment result, and the output is a notification message sent to the administrator. Specifically, email and dashboard notifications are sent when certain conditions are met.

[1590] (Application example 1)

[1591] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1592] Conventional systems detect and assess the risk of slang only in text data on the Internet, but do not address slang contained in audio or image data. This makes it difficult to monitor dangerous behavior in real time, and is insufficient in situations where a rapid response is required, such as in public places or events.

[1593] 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.

[1594] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for notifying users of the analyzed information in real time, means for collecting and analyzing audio data, and means for collecting and analyzing image data. This makes it possible to detect dangerous slang in real time from various data including audio and image data.

[1595] A "server" is a computer system that collects and analyzes data over a network.

[1596] The "Internet" is a huge information network that spans the entire world and is a platform for sending and receiving various types of information.

[1597] "Text data" refers to information composed of character strings, including social media posts, blog articles, and message board posts on the Internet.

[1598] A "generative AI model" is a model that uses artificial intelligence technology to learn patterns and features from text data.

[1599] "Training" is the process of feeding a generative AI model large amounts of data to teach it specific patterns and characteristics.

[1600] "Argot" is a word or phrase that has a different meaning from its usual meaning among a particular group or in a particular situation.

[1601] "Usage patterns" refer to how frequently and in what contexts particular words or phrases are used.

[1602] "Context" refers to the context or situation in which a particular word or phrase is used.

[1603] "Danger level" is an indicator of the likelihood that a particular account is involved in criminal activity, derived from the pattern and context of code word usage.

[1604] "User" is a term that refers to a person or account that posts information on the Internet.

[1605] "Real-time" is a term that indicates the extremely short time between the occurrence of an event and the information being processed and analyzed.

[1606] "Notification" is a means by which the system informs the user of analysis results and important information.

[1607] "Voice data" refers to data that records human speech or sound in digital format.

[1608] "Image data" is visual data that records light information in digital format.

[1609] This invention is a system that detects code words used for recruiting for criminal activities, analyzes behavioral patterns, and evaluates the risk level. The system functions mainly on a server, collecting, analyzing, and evaluating data on the Internet, and based on the results, identifies potentially dangerous users.

[1610] Data Collection Phase

[1611] The server periodically crawls the web and collects text data from social media sites, message boards, blogs, etc. For example, the server uses the API of a social media site to set a keyword filter and retrieve posts that match specific criteria. This data is temporarily stored in a buffer and then stored in a database.

[1612] Secret language learning phase

[1613] The server uses text data collected from the database to train a generative AI model, which uses natural language processing techniques to learn specific slang terms and the context in which they are used, for example, learning that the word "crystal" means stimulant in certain contexts.

[1614] The jargon analysis phase

[1615] The trained generative AI model is then used to analyze newly collected text data. The server scans the text data for words and phrases to detect whether they contain slang, and then analyzes the context in which those slang words are used to determine whether they are related to illegal activity.

[1616] Risky user identification phase

[1617] Based on the analysis results, the server evaluates the behavioral patterns of accounts that use a lot of slang, including the frequency of posts, patterns of interaction with other accounts, and past posting history. This allows the server to identify potentially dangerous users and rate their risk level using a number or category.

[1618] Real-time notification and multi-data analysis functions

[1619] The analyzed information is then sent to users in real time via smart glasses or other wearable devices. The system also collects audio and image data for similar analysis. For example, speech recognition technology is used to convert speech at events into text and analyze high-risk slang. It also uses Google Cloud Vision API to extract text from images and perform similar analysis.

[1620] Hardware and Software Used

[1621] This system uses the following hardware and software:

[1622] Server: A central processing unit that performs data collection and analysis.

[1623] SNS API: An API used to collect data.

[1624] Generative AI model: A model for learning and analyzing jargon from data.

[1625] Speech recognition libraries (e.g. speech_recognition): Used to convert voice data into text.

[1626] Google Cloud Vision API: Used to extract text from image data.

[1627] Google Cloud Natural Language API: Used to parse text data and extract entities.

[1628] Specific examples

[1629] For example, consider the following social media post:

[1630] User A: "Can you bring me some crystals tonight?"

[1631] User B: "Okay. In the usual place."

[1632] In this case, the server performs the following process.

[1633] Data collection: The server retrieves these posts from the internet and stores them in a database.

[1634] Slang learning: A trained generative AI model recognizes that the word "crystal" is slang for stimulant drugs.

[1635] Code word analysis: The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at the usual place" suggests an illegal transaction.

[1636] Identifying risky users: Analyzing the past posting history and interaction patterns of User A and User B, the server assesses the likelihood that the two are involved in illegal activities. Based on this information, the server marks User A and User B as high-risk users and notifies the administrator if necessary.

[1637] Example prompt for a generative AI model:

[1638] Check to see if any of the following words appear in today's conversations: crystal, popcorn, white circle.

[1639] If you find any, assess whether they may be related to illegal activity.

[1640] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1641] Step 1:

[1642] The server collects text data generated on the Internet.

[1643] Input: Text data from the internet

[1644] Output: Collected text data

[1645] How it works: The server uses the SNS API to set up a specific keyword filter and periodically retrieves posts that match the filter. This data is temporarily stored in a buffer and then stored in a database.

[1646] Step 2:

[1647] The server uses the collected text data to train a generative AI model.

[1648] Input: Collected text data

[1649] Output: A trained generative AI model

[1650] How it works: The server extracts text data from a database and uses natural language processing (NLP) techniques to train a generative AI model to understand jargon and its context.

[1651] Step 3:

[1652] The server uses the trained generative AI model to detect jargon in new text data.

[1653] Input: New text data, trained generative AI model

[1654] Output: Detected slang

[1655] How it works: The server analyzes newly collected text data in real time and uses a generative AI model to detect whether it contains slang.

[1656] Step 4:

[1657] The server analyzes the usage patterns and context of slang and assesses the risk level of the account being analyzed.

[1658] Input: Detected slang, new text data

[1659] Output: Account risk assessment results

[1660] Specific operation: The server evaluates the frequency of slang use, the context of use, interaction patterns with other accounts, etc., and evaluates the risk level of each account using a number or category.

[1661] Step 5:

[1662] The server identifies risky users based on the evaluation results.

[1663] Input: Account risk assessment results

[1664] Output: Identified risky users

[1665] Specific operation: The server lists accounts that are assessed as high risk and prepares to notify the administrator of the list.

[1666] Step 6:

[1667] The server notifies the user of the analyzed information in real time.

[1668] Input: Analysis results

[1669] Output: User notification

[1670] Specific operation: The server generates a warning message based on the analysis results and notifies wearable devices such as smart glasses in real time.

[1671] Step 7:

[1672] The device collects and analyzes audio data at events and public places.

[1673] Input: Audio data

[1674] Output: Text data

[1675] Specific operation: The device (smart glasses, etc.) converts audio data recorded at events or public places into text using the SpeechRecognition library.

[1676] Step 8:

[1677] The terminal collects and analyzes image data.

[1678] Input: Image data

[1679] Output: Text data

[1680] Specific operation: The device (smart glasses, etc.) converts the captured image into text using the Google Cloud Vision API and sends the result to the server.

[1681] The above are the processing steps of the present invention, which make it possible to detect and notify dangerous slang from a variety of data sources in real time.

[1682] 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.

[1683] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is built around a server and performs a series of processes, including data collection, analysis, evaluation, and notification.

[1684] 1. Data Collection Phase

[1685] The server collects text data from various sources on the Internet (e.g., social media, bulletin boards, blogs, etc.). The collected data is temporarily stored in a buffer and then stored in a database. A regular scheduling process is performed to collect new data in real time.

[1686] 2. Secret Language Learning Phase

[1687] The server uses the collected text data to train a generative AI model, which learns specific slang terms and the context in which they are used. For example, it learns from the training data that "crystal" means stimulant in certain contexts.

[1688] 3. The jargon analysis phase

[1689] Using the trained generative AI model, the server can detect code words in newly collected text data, and if code words are present, analyze their context in detail to determine whether they are related to illegal activity.

[1690] 4. Analysis phase using emotion engine

[1691] The server uses an emotion engine to recognize emotions from users' text data. This engine can analyze the emotional tone and changes in the content of posts. For example, if a post contains anger or stress, it can determine that this is likely related to the use of slang.

[1692] 5. Risky user identification phase

[1693] The server evaluates the user's risk level based on the analysis results of the emotion engine and the usage pattern of slang. The evaluation takes into account the frequency of posts, frequency of slang use, and emotional fluctuations. For example, if the context in which slang is used is accompanied by tension or excitement, the risk level is evaluated as high.

[1694] 6. Notification Phase

[1695] The server identifies users who are assessed as high risk and notifies the administrator, who uses this information to guide appropriate action.

[1696] Specific examples

[1697] For example, suppose there is a social media post like the one below.

[1698] User A: "Bring me the crystal tonight, at the place I told you about."

[1699] User B: "Be careful. Come early. I'm getting annoyed."

[1700] In this case, the processing flow is as follows.

[1701] 1. Data Collection:

[1702] The server collects these posts and stores them in a database.

[1703] 2. Learn the secret language:

[1704] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1705] 3. Analysis of the jargon:

[1706] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1707] 4. Analysis by emotion engine:

[1708] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1709] 5. Identifying risky users:

[1710] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1711] 6. Notice:

[1712] The server notifies the administrator of the results and prompts them to take appropriate action.

[1713] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[1714] The processing flow will be explained below.

[1715] Step 1:

[1716] The server periodically launches a scheduler to crawl the web. This scheduler runs at specified intervals and collects text data using SNS APIs and web scraping technology.

[1717] Step 2:

[1718] The server stores text data obtained from various internet sources such as social media, message boards, and blogs in a temporary buffer. The collected data also includes information about the poster (user ID, posting time, etc.).

[1719] Step 3:

[1720] The server stores the buffered text data in a database, where it is organized and indexed for later analysis and learning.

[1721] Step 4:

[1722] The server retrieves the text data from the database and performs noise removal and formatting of the document, such as removing HTML tags and meaningless spaces.

[1723] Step 5:

[1724] The server uses the preprocessed data to train a generative AI model, which in the training process learns the jargon contained in the text data and the context in which it is used.

[1725] Step 6:

[1726] Based on the training results, the server generates a dictionary of specific slang terms and their usage patterns, which is used to determine whether new data contains slang terms.

[1727] Step 7:

[1728] The server collects newly generated posts and comments in real time and stores them in a buffer as text data, which can then be analyzed.

[1729] Step 8:

[1730] The server uses the generative AI model and slang dictionary to analyze new data to determine whether it contains slang. If slang is identified, it analyzes the context and surrounding words.

[1731] Step 9:

[1732] The server then carefully evaluates the context in which the slang is used, specifically analyzing the combination of related keywords and phrases to determine whether the word is related to illegal activity.

[1733] Step 10:

[1734] The server uses an emotion engine to recognize the poster's emotions from the text data. This emotion analysis includes the tone of the text, the intensity of the emotion, and the type of emotion (joy, anger, sadness, etc.).

[1735] Step 11:

[1736] The server combines the identified emotional information with the slang usage patterns and stores them in a database, making the correlation between emotional fluctuations and slang usage available for future analysis.

[1737] Step 12:

[1738] The server evaluates the results of the emotion engine and the usage patterns of slang comprehensively to assess the risk level of each user. For example, if the context in which slang is used is emotionally charged, the risk level is evaluated as high.

[1739] Step 13:

[1740] The server identifies high-risk users based on the evaluation results, which include the behavioral patterns, past posts, and emotional fluctuations of the identified risky users.

[1741] Step 14:

[1742] The server notifies the administrator of users who are rated as highly dangerous. The notification includes the user's ID, the content of their posts, the risk assessment result, and the related sentiment analysis result.

[1743] Specific examples

[1744] For example, suppose there is a social media post like the one below.

[1745] User A: "Bring me the crystal tonight, at the place I told you about."

[1746] User B: "Be careful. Come early. I'm getting annoyed."

[1747] In this case, the processing flow is as follows.

[1748] 1. Data Collection:

[1749] The server collects these posts and stores them in a database.

[1750] 2. Learn the secret language:

[1751] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1752] 3. Analysis of the jargon:

[1753] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1754] 4. Analysis by emotion engine:

[1755] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1756] 5. Identifying risky users:

[1757] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1758] 6. Notice:

[1759] The server notifies the administrator of the results and prompts them to take appropriate action.

[1760] In this way, by combining code word detection with user sentiment analysis, the system can comprehensively analyze and quickly respond to potentially criminal activity.

[1761] Example 2

[1762] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1763] It is difficult to detect in real time when users are using slang on the Internet, and to recognize the users' emotions and evaluate their overall risk level. In addition, existing systems cannot integrate slang detection and emotion analysis, identify risky users based on the results, and promptly notify administrators, which results in delayed response.

[1764] 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.

[1765] In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for identifying risky users based on the evaluation results, means for recognizing emotions from user text data using an emotion engine, means for evaluating user risk levels based on the analysis results of the emotion engine and slang usage patterns, and means for notifying an administrator of information about high-risk users. This makes it possible to effectively detect fraudulent activities and risky behavior on the Internet and respond quickly.

[1766] "Internet-generated text data" refers to text information posted by users on online platforms such as websites, social networking sites, message boards, and blogs.

[1767] "Collection methods" refer to the processes and tools used to obtain text data from multiple sources on the Internet.

[1768] A "generative AI model" is an artificial intelligence model that has been trained using machine learning algorithms to perform a specific task, in this case code word detection.

[1769] "Training" is the process of using collected data to train a generative AI model and improve the model's performance.

[1770] A "jargon" is a word or expression used within a particular group or community that is only understood by a select few.

[1771] "Detection" is the process of using trained generative AI models to find jargon and specific patterns in text data.

[1772] "Usage patterns" are data that show how frequently and in what contexts a particular jargon is used.

[1773] "Context" refers to the context or situation before and after the use of jargon in text data.

[1774] "Means of analysis" refers to the process of closely examining the collected data, jargon usage patterns, and context to understand their meaning and intent.

[1775] "Danger level" is an assessment that indicates how dangerous the target user is based on text data, slang usage patterns, and the results of sentiment analysis.

[1776] An "emotion engine" is a tool or model for identifying user emotions from text data and analyzing their emotional tone and variance.

[1777] "Means of recognition" refers to the process of identifying emotions in text data using an emotion engine and obtaining analysis results.

[1778] "High-risk users" are users who are deemed to be at high risk based on the use of slang and the results of sentiment analysis.

[1779] "Means of notification" refers to methods and tools used to inform administrators about high-risk users.

[1780] This invention is a system that detects situations in which users on the Internet use slang and recognizes their emotions to assess the overall risk level. This system is mainly composed of a server, and performs a series of processes including data collection, analysis, evaluation, and notification.

[1781] First, the server collects text data from various sources on the Internet (social media, bulletin boards, blogs, etc.). This collection is done using the Python language to obtain data via API. For example, a request is sent to the social media API to collect text data. This is done using the Python requests library. The collected data is temporarily saved in a buffer and then stored in a MySQL database. A regular scheduling process (e.g., a cron job) is performed to collect new data in real time.

[1782] Next, the server uses the collected text data to train a generative AI model. It uses OpenAI's GPT-3 to train the model, which uses a process to learn specific slang terms and their contexts. For example, it learns that the slang term "crystal" means stimulant. It uses OpenAI's libraries for this training.

[1783] Using a trained generative AI model, the server detects slang in newly collected text data. The model analyzes specific words and phrases to determine whether they are slang, and then performs detailed analysis of the context of the detected slang to determine whether it is related to illegal activity.

[1784] Furthermore, the server uses an emotion engine to recognize emotions from the user's text data. This emotion engine uses Azure's Text Analytics API and can analyze the emotional tone and changes in the posted content. For example, it can recognize the user's excited state from the word "irritated."

[1785] The server comprehensively evaluates the emotion engine's analysis results and the usage patterns of slang to assess the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of use of slang, and emotional fluctuations. Based on this, it is determined that a particular user is high risk.

[1786] Finally, the server notifies the administrator of users who are assessed as high risk via email or the administration screen, allowing the administrator to take prompt action.

[1787] Specific examples

[1788] For example, suppose there is a social media post like the one below.

[1789] User A: "Bring me the crystal tonight, at the place I told you about."

[1790] User B: "Be careful. Come early. I'm getting annoyed."

[1791] In this case, the processing flow is as follows.

[1792] 1. Data Collection:

[1793] The server collects these posts and stores them in a database.

[1794] 2. Learn the secret language:

[1795] A trained generative AI model recognizes that "crystal" is a slang term for stimulants.

[1796] 3. Analysis of the jargon:

[1797] The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1798] 4. Analysis by emotion engine:

[1799] The server recognizes from the text "I'm irritated" that User B is in an excited state.

[1800] 5. Identifying risky users:

[1801] The server evaluates the risk level of users A and B based on the results of the emotion engine and their usage patterns of slang, thereby identifying them as high-risk users.

[1802] 6. Notice:

[1803] The server notifies the administrator of the results and prompts them to take appropriate action.

[1804] Prompt Sentence Examples

[1805] For example, use the following prompt for your generative AI model:

[1806] "Detect whether the text data contains slang and determine its context. Furthermore, analyze the sentiment of the post and assess its overall risk level. The text data is shown below."

[1807] In this way, this system combines code word detection and sentiment analysis to comprehensively analyze risky behavior on the Internet and respond quickly.

[1808] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1809] Step 1:

[1810] Data collection

[1811] The server collects text data from social networking sites, message boards, blogs, etc. on the Internet.

[1812] As a concrete example, we will use the Python language and the requests library to retrieve data from a social networking service API.

[1813] Input: Text data from the internet

[1814] Output: Collected text data (saved in a buffer)

[1815] Specific behavior:

[1816] The server sends a request to the SNS API and retrieves the text data.

[1817] The server temporarily stores the acquired data in a buffer.

[1818] Step 2:

[1819] Database storage

[1820] The server periodically stores the data in the buffer in a database.

[1821] Input: Text data in a buffer

[1822] Output: Text data stored in a database

[1823] Specific behavior:

[1824] The server takes the data from the buffer and stores it in a MySQL database.

[1825] The server uses the mysql-connector-python library for database connections.

[1826] Step 3:

[1827] Learning slang

[1828] The server feeds the collected text data back into a generative AI model (e.g., GPT-3) to learn slang and the context in which it is used.

[1829] Input: Collected text data

[1830] Output: A trained generative AI model

[1831] Specific behavior:

[1832] The server uses OpenAI's API to input text data into the generative AI model.

[1833] The generative AI model learns specific jargon and its context.

[1834] Step 4:

[1835] Detecting slang

[1836] The server uses a trained generative AI model to detect slang in newly acquired text data.

[1837] Input: New text data

[1838] Output: Detected slang words and their context information

[1839] Specific behavior:

[1840] The server uses a generative AI model to analyze new text data and detect slang.

[1841] The server also analyzes the context of the detected code words to determine if they relate to illegal activity.

[1842] Step 5:

[1843] Emotion analysis

[1844] The server uses an emotion engine to recognize emotions from the user's text data.

[1845] Input: User's text data

[1846] Output: Emotion analysis results (e.g. anger, stress, etc.)

[1847] Specific behavior:

[1848] The server uses the Azure Text Analytics API and other tools to analyze the emotional tone and fluctuations of the posts.

[1849] The server evaluates the user's emotional state based on the acquired emotional data.

[1850] Step 6:

[1851] Risk Assessment

[1852] The server combines the analysis results of the emotion engine with the usage patterns of slang to assess the user's risk level.

[1853] Input: Sentiment analysis results, slang usage patterns

[1854] Output: User risk assessment

[1855] Specific behavior:

[1856] The server calculates the risk level by combining the results of the emotion analysis with the frequency of appearance of slang.

[1857] The server uses this assessment to identify high-risk users.

[1858] Step 7:

[1859] notification

[1860] The server notifies the administrator of information about users who are assessed as high risk.

[1861] Input: High-risk user information

[1862] Output:Notify administrator

[1863] Specific behavior:

[1864] The server uses SMTP to send email notifications to the administrator.

[1865] The server also displays the information on the management screen so that the administrator can respond quickly.

[1866] (Application example 2)

[1867] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1868] There is a need to detect signs of illegal or fraudulent activity on the Internet early and take appropriate action. However, the use of slang or coded language makes detection extremely difficult. Furthermore, when identifying potentially dangerous users, it is necessary not only to detect slang but also to consider the user's emotions and the context of their behavior. Currently, there is a lack of systems that can perform such comprehensive analysis in real time.

[1869] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for collecting text data generated on the Internet, means for training a generative AI model that learns slang from the collected text data, means for detecting slang in new text data using the trained generative AI model, means for analyzing slang usage patterns and contexts and evaluating the risk level of the account being analyzed, means for detecting slang and performing sentiment analysis using the generative AI model, means for comprehensively evaluating the results of the sentiment analysis and the slang usage patterns to calculate the risk level, means for identifying risky users based on the evaluation results, and means for notifying an administrator of information about identified risky users. This makes it possible to comprehensively evaluate potentially risky users by combining slang detection and sentiment analysis and quickly address them.

[1870] "Internet-generated text data" refers to text content posted by users on online platforms such as websites, social media platforms, chat applications, blogs, and message boards.

[1871] "Means of collection" refers to collection mechanisms that use functions for efficiently capturing text data from various sources on the Internet, such as web crawlers and APIs (application programming interfaces).

[1872] A "generative AI model" is an algorithm that has been trained using machine learning and deep learning techniques to perform a specific task (in this case, slang detection and sentiment analysis).

[1873] "Training means" refers to the process of training a generative AI model using large amounts of text data, which includes data preprocessing, model training, evaluation, and tuning.

[1874] "Code word detection" refers to the ability to use a trained generative AI model to identify specific coded words or slang within text.

[1875] "Means for performing emotion analysis" refers to the analytical process for determining a user's emotional state (e.g., joy, anger, sadness, surprise, etc.) from text data and the generative AI model used for that analysis.

[1876] "Means for analyzing context" refers to the ability to perform a detailed analysis of the usage of slang and the surrounding context, and determine what that context indicates.

[1877] "Means for assessing risk" refers to a function that comprehensively considers slang usage patterns and the results of sentiment analysis, and expresses in numerical terms or ratings how potentially dangerous a particular user is.

[1878] "Means for identifying risky users" refers to a function that identifies which users pose a high risk based on the assessed risk level and lists those users.

[1879] "Means of notifying administrators" refers to functions for promptly communicating information about identified risky users to administrators, including email notifications, alert systems, dashboard displays, etc.

[1880] "Means of collection in real time" refers to the ability to instantly capture newly generated text data and keep it in a state where it can be analyzed in near real time.

[1881] "Means for saving as a dictionary" refers to the function of saving the usage patterns and meanings of slang that have been learned in a database or file format, and reusing them for subsequent analysis.

[1882] The present invention provides a system for identifying risky users through the detection of slang and analysis of user sentiment, and for dealing with such users appropriately. The following describes in detail the embodiments of the present invention.

[1883] 1. Overall structure

[1884] The server runs a series of processes to collect, analyze, and evaluate text data generated on the internet. The required hardware includes a web crawler and a high-performance server with APIs to collect the data. It also uses deep learning frameworks such as TensorFlow and PyTorch to run generative AI models for sentiment analysis and slang detection.

[1885] 2. Data Collection Phase

[1886] The server collects text data from social media, bulletin boards, blogs, etc. In this process, data is collected periodically using a web crawler or API, temporarily saved in a buffer, and then stored in a database. A scheduling process is set up to collect new data in real time.

[1887] 3. Secret Language Learning Phase

[1888] The server uses the collected text data to train a generative AI model, which then learns specific slang terms and the context in which they are used. For example, if "crystal" is a slang term for stimulants, the model learns their usage patterns from the dataset.

[1889] 4. The jargon analysis phase

[1890] Using a trained generative AI model, the server detects slang in newly collected text data. The model identifies slang used in specific contexts and determines whether the context is related to illegal or fraudulent activity.

[1891] 5. Analysis phase using emotion engine

[1892] The server uses a sentiment analysis engine to recognize emotions from the user's text data. This engine analyzes the emotional tone and fluctuations of the posts. For example, if the post contains the expression "irritated," it will recognize that the user is in an excited state.

[1893] 6. Risky user identification phase

[1894] The server comprehensively evaluates the results of the emotion analysis and the usage patterns of slang to calculate the user's risk level. The evaluation takes into account factors such as frequency of posts, frequency of slang use, and emotional fluctuations. If slang is used with tension or excitement, the risk level is rated high.

[1895] 7. Notification Phase

[1896] The server identifies users who are assessed as high risk and notifies administrators of this information in the form of email, dashboard display, alerts, etc.

[1897] Specific examples

[1898] For example, suppose there is a social media post like the one below.

[1899] "Bring me some crystal tonight, at the place I told you about. I'm getting annoyed."

[1900] When this post is entered into the system, the server processes it as follows:

[1901] 1. The server collects the posts and stores them in a database.

[1902] 2. A trained generative AI model recognizes that "crystal" is slang for stimulants.

[1903] 3. The server analyzes the context of the post and determines that the combination of the phrases "crystal" and "at that place I mentioned earlier" suggests an illegal transaction.

[1904] 4. The server recognizes from the text "I'm irritated" that the user is in an excited state.

[1905] 5. The server evaluates the user's risk level based on the results of emotion analysis and the pattern of slang usage, and determines that the user is at high risk.

[1906] 6. The server notifies the administrator of the result and prompts them to take appropriate measures.

[1907] Prompt Sentence Examples

[1908] "Analyze the emotions and jargon contained in the given text: 'Bring me some crystal tonight, in the place I mentioned earlier. I'm frustrated.'"

[1909] This system will enable early identification of potentially dangerous users on the Internet and enable rapid response.

[1910] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1911] Step 1:

[1912] The server collects text data generated on the Internet. Specifically, it uses a web crawler or API to obtain text data from social media sites, message boards, blogs, etc. The input is the target URL or API endpoint, and the output is the obtained text data. The obtained text data is temporarily saved in a buffer and then stored in a database.

[1913] Step 2:

[1914] The server uses the collected text data to train a generative AI model that learns slang. Specifically, the collected data is preprocessed (text cleaning, tokenization, etc.) and then input into the generative AI model. The input is the preprocessed text data, and the output is a trained generative AI model. This model learns specific slang and its usage context.

[1915] Step 3:

[1916] The server uses the trained generative AI model to detect slang words in newly collected text data. Specifically, the collected text data is input into the trained model to determine whether it contains slang words. The input is the new text data, and the output is the detected slang words and their location information.

[1917] Step 4:

[1918] The server analyzes the usage patterns and context of the detected slang to evaluate the meaning behind it. Specifically, it performs contextual analysis to determine whether the detected slang is related to illegal or fraudulent activity. The input is the slang and its surrounding context, and the output is the analysis result.

[1919] Step 5:

[1920] The server uses an emotion analysis engine to recognize emotions from the user's text data. Specifically, the text data is input into the emotion analysis engine to determine the user's emotional state (e.g., anger, joy, fear, etc.). The input is the text data, and the output is the determined emotion label and its score.

[1921] Step 6:

[1922] The server comprehensively evaluates the results of the sentiment analysis and the usage patterns of slang to calculate the user's risk level. Specifically, it calculates an overall risk score by taking into account the frequency of slang usage, emotional fluctuations, frequency of posting, etc. The input is the results of the sentiment analysis and slang analysis, and the output is a risk score.

[1923] Step 7:

[1924] The server identifies risky users based on the assessed risk level and notifies the administrator of this information. Specifically, the server communicates the information to the administrator via email or an alert on the dashboard. The input is the risk level score, and the output is a notification message.

[1925] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice 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 voice data.

[1926] 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> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) 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.

[1927] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[1928] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1929] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[1930] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[1931] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[1932] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[1933] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[1934] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[1935] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[1936] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[1937] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[1938] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[1939] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[1940] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[1941] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[1942] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[1943] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[1944] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[1945] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[1946] The following is further disclosed regarding the above embodiment.

[1947] (Claim 1)

[1948] A means for collecting text data generated on the internet;

[1949] A means of training a generative AI model that learns jargon from collected text data; and

[1950] means for detecting slang terms in novel text data using the trained generative AI model;

[1951] A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account;

[1952] A means for identifying risky users based on the evaluation results;

[1953] A system including:

[1954] (Claim 2)

[1955] 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

[1956] (Claim 3)

[1957] 10. The system of claim 1, further comprising means for storing the learned slang usage patterns as a dictionary and for detecting slang using the dictionary.

[1958] "Example 1"

[1959] (Claim 1)

[1960] A means for collecting text data generated on the internet;

[1961] A means for temporarily storing the collected text data in a buffer and storing it in a database;

[1962] A means of training a generative AI model that learns jargon from collected text data; and

[1963] means for detecting slang terms in novel text data using the trained generative AI model;

[1964] A means for analyzing the context of slang usage in newly collected text data;

[1965] A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account;

[1966] A method for identifying risky users based on the evaluation results and notifying the administrator as necessary;

[1967] A system including:

[1968] (Claim 2)

[1969] 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

[1970] (Claim 3)

[1971] 10. The system of claim 1, further comprising means for storing the learned slang usage patterns as a dictionary and for detecting slang using the dictionary.

[1972] "Application Example 1"

[1973] (Claim 1)

[1974] A means for collecting text data generated on the internet;

[1975] A means of training a generative AI model that learns jargon from collected text data; and

[1976] means for detecting slang terms in novel text data using the trained generative AI model;

[1977] A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account;

[1978] A means for identifying risky users based on the evaluation results;

[1979] A means of notifying users of the analyzed information in real time,

[1980] means for collecting and analyzing audio data;

[1981] means for collecting and analyzing image data;

[1982] A system including:

[1983] (Claim 2)

[1984] 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

[1985] (Claim 3)

[1986] 10. The system of claim 1, further comprising means for storing the learned slang usage patterns as a dictionary and for detecting slang using the dictionary.

[1987] "Example 2: Combining Emotion Engines"

[1988] (Claim 1)

[1989] A means for collecting text data generated on the internet;

[1990] A means of training a generative AI model that learns jargon from collected text data; and

[1991] means for detecting slang terms in novel text data using the trained generative AI model;

[1992] A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account;

[1993] A means for identifying risky users based on the evaluation results;

[1994] means for recognizing emotions from user text data using an emotion engine;

[1995] a means for assessing the risk level of a user based on the analysis results of the emotion engine and the usage pattern of slang;

[1996] A means of notifying administrators of high-risk user information;

[1997] A system including:

[1998] (Claim 2)

[1999] 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

[2000] (Claim 3)

[2001] 10. The system of claim 1, further comprising means for storing the learned slang usage patterns as a dictionary and for detecting slang using the dictionary.

[2002] "Application example 2 when combining emotion engines"

[2003] (Claim 1)

[2004] A means for collecting text data generated on the internet;

[2005] A means of training a generative AI model that learns jargon from collected text data; and

[2006] means for detecting slang terms in novel text data using the trained generative AI model;

[2007] A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account;

[2008] A means of using generative AI models to perform sentiment analysis in addition to slang detection;

[2009] A means for comprehensively evaluating the results of emotion analysis and patterns of slang usage to calculate the degree of risk;

[2010] A means for identifying risky users based on the evaluation results;

[2011] a means for notifying an administrator of information about the identified risky user;

[2012] A system including:

[2013] (Claim 2)

[2014] 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

[2015] (Claim 3)

[2016] 10. The system of claim 1, further comprising means for storing the learned slang usage patterns as a dictionary and for detecting slang using the dictionary. [Explanation of symbols]

[2017] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. A means for collecting text data generated on the internet; A means of training a generative AI model that learns jargon from collected text data; and means for detecting slang terms in novel text data using the trained generative AI model; A method for analyzing the usage patterns and context of slang to assess the risk level of the analyzed account; A means for identifying risky users based on the evaluation results; A system including:

2. 10. The system of claim 1, further comprising means for collecting new text data in real time and storing it for analysis.

3. 2. The system of claim 1, further comprising means for storing the learned usage patterns of slang as a dictionary and for detecting slang using the dictionary.

Images