System

Abstract

A system is provided.SOLUTION: A system comprising: means for collecting word-of-mouth data; means for pre-processing the collected word-of-mouth data; means for determining malicious word-of-mouth based on the pre-processed word-of-mouth data; means for filtering the malicious word-of-mouth; and means for displaying the filtered high-confidence word-of-mouth to a user.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] In recent years, word-of-mouth reviews have spread across many websites, and users refer to them on a daily basis. However, in reality, malicious reviews such as slanderous comments, factually incorrect comments, and even stealth marketing are rampant, leading users and stores to be misled by unreliable information. This can also lead to a decrease in store sales and the risk of users purchasing disappointing products. For this reason, there is an urgent need to improve the reliability of review sites and provide an information environment that is beneficial to both users and stores. [Means for solving the problem]

[0005] The present invention provides a means for collecting and preprocessing review data. It also includes a means for determining malicious reviews based on the preprocessed review data, and a means for detecting reviews containing negative words, reviews containing patterns determined to be stealth marketing, and reviews with high negative sentiment scores through sentiment analysis, and determining the relevant reviews as malicious. It also includes a means for learning past posting history and account trends to detect and eliminate malicious reviews from specific accounts. Finally, it includes a means for displaying filtered, reliable reviews to users, and performs this series of processes in real time to provide reliable review information.

[0006] "Collection" is the process of retrieving review data from external websites and databases and aggregating it in one place.

[0007] "Preprocessing" is the process of removing unnecessary information from collected word-of-mouth data and cleaning the text to make it easier to analyze.

[0008] "Determination" is the process of identifying whether a review is malicious or not based on pre-processed review data using specific criteria or algorithms.

[0009] "Massive reviews" refer to reviews that contain negative words, patterns that are considered to be stealth marketing, or have a high negative sentiment score based on sentiment analysis.

[0010] "Filtering" is the process of excluding determined malicious reviews and extracting only reliable reviews.

[0011] "Reliable reviews" are reviews that are judged to be free of malicious content and based on actual consumer evaluations and experiences.

[0012] "User" refers to an internet user who uses word-of-mouth information to check the evaluation of a product or service.

[0013] "Real-time" means that the entire process from collecting review data to pre-processing, judging, filtering, and displaying it is carried out instantly and without delay.

[0014] "Account trends" refers to the characteristics and patterns of reviews posted by a particular user in the past.

[0015] "Sentiment analysis" is a technology that automatically analyzes positive, negative, and neutral emotions from the content of review text.

[0016] "Stealth marketing" is a marketing technique in which a third party poses as a third party and excessively recommends a product or service to consumers. [Brief explanation of the drawings]

[0017] [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

[0018] 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.

[0019] First, the terms used in the following description will be explained.

[0020] 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).

[0021] 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.

[0022] 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.

[0023] 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.

[0024] 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."

[0025] [First embodiment]

[0026] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0027] 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.

[0028] 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).

[0029] 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.

[0030] 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.

[0031] 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.

[0032] 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.

[0033] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0034] 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.

[0035] 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.

[0036] 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.

[0037] 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."

[0038] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. The system is configured around a server and operates in the following steps.

[0039] Collecting word-of-mouth data

[0040] The server collects real-time review data via websites and APIs. The review data includes metadata such as text, author information, and posting date and time.

[0041] Data Preprocessing

[0042] The server pre-processes the collected review data, which includes basic text cleaning such as trimming whitespace, removing HTML tags and special characters, and converting everything to lowercase, ensuring the data is formatted consistently.

[0043] Detecting malicious reviews

[0044] The server analyzes the pre-processed review data and detects malicious reviews. This detection is done using several techniques:

[0045] 1. Negative word check: The server scans the review content using a predefined list of negative words (e.g., "worst," "disappointed," "bad," "defective," etc.) to determine whether these words are included.

[0046] 2. Stealth marketing detection: The server checks whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) appear multiple times in the reviews to determine whether they are stealth marketing.

[0047] 3. Sentiment analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it determines the review to be malicious.

[0048] Review filtering

[0049] The server filters out reviews that are deemed malicious and extracts the rest as reliable reviews, allowing users to view only these filtered reviews.

[0050] Learn about past posting history and account trends

[0051] The server learns users' past posting history and account trends, and has the ability to detect patterns of malicious reviews from specific accounts. This allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0052] Real-time processing

[0053] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always view the latest information.

[0054] Specific examples

[0055] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious in the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times. On the other hand, user B's review saying, "I had a bad experience, the product was defective." is identified as malicious through negative word checking and sentiment analysis.

[0056] After filtering, only User C's review, "The product works as expected and I am satisfied," remains as a reliable review and is finally viewable by the user.

[0057] In this way, the system of the present invention collects word-of-mouth data and performs a series of processes from pre-processing, to detecting malicious word-of-mouth, filtering, and displaying, to provide highly reliable word-of-mouth information.

[0058] The processing flow will be explained below.

[0059] Step 1:

[0060] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0061] Step 2:

[0062] The server preprocesses the collected review data, specifically performing basic text cleaning such as removing whitespace and HTML tags, converting everything to lowercase, etc. This step standardizes the data into a consistent format that makes it easier to analyze.

[0063] Step 3:

[0064] The server identifies malicious reviews based on the pre-processed review data. First, it scans the content of the review to see if it contains any of the negative words listed (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0065] Step 4:

[0066] The server performs stealth marketing detection, identifying over-promotional and fake reviews by checking if certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in reviews. This technique identifies potentially inauthentic reviews.

[0067] Step 5:

[0068] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the text to identify positive, negative, and neutral sentiment, and classifies the review as malicious if the sentiment is high. This allows reviews with a high percentage of negative content to be detected.

[0069] Step 6:

[0070] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users view them, and provides only reliable information.

[0071] Step 7:

[0072] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows the server to proactively deal with specific accounts and prevent malicious reviews before they occur.

[0073] Step 8:

[0074] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0075] Step 9:

[0076] The device displays filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0077] Example 1

[0078] 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."

[0079] Word-of-mouth information is an important factor in determining the reputation of products and services on the Internet, but it is often mixed with false information caused by malicious word-of-mouth and stealth marketing. This makes it difficult for users to obtain reliable information. The object of the present invention is to solve these problems and enable users to quickly obtain reliable word-of-mouth information.

[0080] 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.

[0081] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for learning the user's past posting history and account trends to detect patterns of malicious word-of-mouth, and means for executing the collection, preprocessing, determination, and filtering processes in real time, thereby enabling the user to always obtain the latest reliable word-of-mouth information.

[0082] "Word-of-mouth data" is text data that includes user ratings and impressions of products and services.

[0083] "Real-time" means that data processing and information provision is immediate and without delay.

[0084] "Preprocessing" is a series of text cleaning tasks performed to standardize the data format and make it easier to analyze.

[0085] "Malicious reviews" refer to ratings and opinions that contain false or exaggerated content and may mislead other users.

[0086] "Filtering" is the process of sorting data based on specific conditions and eliminating unnecessary data.

[0087] "Reliable reviews" refer to text data that contains accurate and objective evaluations and opinions.

[0088] "Stealth marketing" is a marketing technique that makes products appear as word-of-mouth reviews from ordinary users, when in fact they are for promotional purposes.

[0089] "Collection" is the process of obtaining data from websites and APIs on the internet.

[0090] "Past posting history" is a record of all posts a user has made to date.

[0091] "Account trends" is information that indicates the posting patterns and reliability of a particular user account.

[0092] "Negative words" refer to words that have negative connotations in user ratings and impressions.

[0093] This invention is a system that analyzes word-of-mouth data, filters out malicious word-of-mouth, and displays only reliable word-of-mouth. A specific implementation method of this system will be described below.

[0094] Hardware and Software Use

[0095] The server is the center of the system and works in conjunction with multiple pieces of software. The server mainly uses the following software:

[0096] Operating systems such as Ubuntu

[0097] Apache Kafka: A streaming platform for real-time data processing

[0098] Apache HTTP Server: Web server

[0099] Python: Use various libraries for data processing, sentiment analysis, and filtering

[0100] Requests: A library for web scraping

[0101] BeautifulSoup: Parsing HTML data

[0102] pandas: Data Shaping and Preprocessing

[0103] re (regular expressions): used for text cleaning

[0104] TextBlob: Used for sentiment analysis

[0105] scikit-learn (sklearn): Used to train and predict machine learning models

[0106] Data collection

[0107] The server collects review data in real time via websites and APIs. For example, the server retrieves data from web pages using Python's Requests library and parses it with BeautifulSoup to extract review information, as shown below. Review data can also be retrieved in JSON format using APIs.

[0108] Data Preprocessing

[0109] The server preprocesses the collected review data by using the pandas library to format the data, the regular expressions library (re) to remove HTML tags and special characters, and lowercasing all text to maintain a consistent data format.

[0110] Detecting malicious reviews

[0111] The server performs text analysis to detect malicious reviews, mainly using the following methods:

[0112] 1. Check for negative words:

[0113] The server scans the reviews using a specified list of negative words to see if they contain any of the relevant words.

[0114] 2. Stealth Marketing Detection:

[0115] Check if certain marketing phrases appear multiple times to determine if they are stealth marketing.

[0116] 3. Sentiment analysis:

[0117] We use the TextBlob library to calculate a sentiment score for reviews, and consider reviews with a high negative score to be malicious.

[0118] Review filtering

[0119] The server filters out the bad reviews and extracts only the reliable reviews. To do this, it uses a pandas DataFrame and filters based on a specific column (e.g., a bad flag).

[0120] Learn about past posting history and account trends

[0121] The server learns users' past posting history and account trends, uses a logistic regression model from scikit-learn to learn patterns from the data, and applies the model to new posts to detect malicious accounts.

[0122] Real-time processing

[0123] The server processes data in real time using Apache Kafka, allowing data collection, analysis, filtering, and display to users to be performed consistently in real time.

[0124] Specific examples

[0125] For example, if user A posts a review saying, "This product is amazing, I highly recommend it!", the server will pre-clean the review and then evaluate it using stealth marketing detection techniques. Because phrases like "amazing" and "highly recommend" are used multiple times, this is determined to be a malicious review. On the other hand, if user B posts, "I had a bad experience, the product was defective," this will be determined to be a malicious review through negative word checking and sentiment analysis. Finally, only user C's review, "The product works as expected and I am satisfied," is displayed as a reliable review.

[0126] Prompt Sentence Examples

[0127] You can use the following prompt for your generative AI model: "Please filter the following review sentence to detect malicious reviews: 'This product is amazing, highly recommend it!'"

[0128] As described above, by implementing the present invention, users can quickly obtain highly reliable word-of-mouth information.

[0129] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0130] Step 1: Collect review data

[0131] The server collects review data from websites and APIs. To do this, it uses Python's Requests library to obtain the HTML source and parses it with BeautifulSoup. It also obtains data in JSON format via a RESTful API. The input is the webpage URL or API endpoint, and the output is the collected raw data. Specifically, it obtains the HTML source and extracts review information using BeautifulSoup. When obtaining data from an API, it parses the JSON-formatted data.

[0132] Step 2: Preprocessing the data

[0133] The server preprocesses the collected review data. Specifically, it uses pandas to format the data and regular expressions (re) to remove HTML tags and special characters. It also lowercases the entire text to ensure a consistent data format. The input is the raw data collected in step 1, and the output is the cleaned data. Specifically, it loads the data into a pandas DataFrame, removes whitespace with the str.strip() method, and removes HTML tags with re.sub().

[0134] Step 3: Detecting malicious reviews

[0135] The server analyzes the pre-processed data to detect malicious reviews using the following techniques:

[0136] 1. Negative Word Check: Scan the reviews using a specified list of negative words.

[0137] 2. Stealth marketing detection: Check if a particular marketing phrase appears multiple times.

[0138] 3. Sentiment Analysis: We use the TextBlob library to calculate sentiment scores and consider messages malicious if they have a high negative score.

[0139] The input is preprocessed data, and the output is data flagged as malicious. Specifically, it checks words in the text against a negative word list and flags any matching words. It uses TextBlob to calculate a sentiment score, and if it exceeds a threshold, it is deemed malicious.

[0140] Step 4: Filter reviews

[0141] The server filters data determined to be malicious reviews and extracts only reliable reviews. The input is data flagged as malicious, and the output is reliable data. Specifically, it uses the filtering function of pandas DataFrame to extract only reviews that are not flagged as malicious.

[0142] Step 5: Study past posting history and account trends

[0143] The server learns the user's past posting history and account trends. It uses a machine learning model (such as scikit-learn's logistic regression model) to learn malicious patterns based on past data. The input is the user's past posting history data, and the output is the trained model. Specifically, it collects past data, performs feature engineering, and then trains a logistic regression model.

[0144] Step 6: Real-time processing

[0145] The server processes data in real time using Apache Kafka. All processing (collection, preprocessing, judgment, filtering) is performed instantly, always providing the latest and most reliable word-of-mouth information. The input is word-of-mouth data flowing in real time, and the output is highly reliable word-of-mouth information updated in real time. Specifically, it uses Apache Kafka to continuously receive streaming data and executes each step in real time.

[0146] Through these steps, the system can provide users with reliable word-of-mouth information.

[0147] (Application example 1)

[0148] 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."

[0149] Much of the online word-of-mouth information is unreliable. In particular, the mixture of malicious word-of-mouth and stealth marketing makes it difficult to obtain reliable information. In this situation, it is a difficult task for users to quickly obtain accurate and reliable word-of-mouth information.

[0150] 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.

[0151] In this invention, the server includes means for collecting review data, means for preprocessing the collected review data, means for determining malicious reviews based on the preprocessed review data, means for filtering malicious reviews, means for displaying the filtered, reliable reviews to the user, means for learning the user's past posting history and account trends to detect malicious reviews from specific accounts, and means for processing the review data in real time, thereby enabling the user to always view the latest, reliable review information.

[0152] "Word-of-mouth data" is text information in which Internet users write their opinions and evaluations of a product or service.

[0153] "Preprocessing" refers to the process of cleaning the collected review data, such as trimming white space, removing HTML tags and special characters, and converting the entire data to lowercase, in order to unify it into a consistent format.

[0154] "Malicious reviews" are reviews that are likely to mislead other users, such as those that include excessive use of negative words or patterns that are considered to be stealth marketing.

[0155] "Filtering means" refers to the process of removing malicious reviews from the preprocessed review data and extracting only reliable reviews.

[0156] "Reliable reviews" are reviews that have not been judged to be malicious and contain information that is useful to other users and does not cause misunderstandings.

[0157] "Past posting history" is a record of all review data posted by a particular user in the past.

[0158] "Account trends" refers to consistent patterns or behaviors of a particular user based on their past posting history.

[0159] "Real-time processing means" refers to the process of instantly pre-processing, analyzing, and filtering collected review data to provide users with the most up-to-date information.

[0160] This invention is a system that analyzes word-of-mouth data, filters out malicious reviews, and displays only reliable reviews. This system is mainly composed of a server that collects word-of-mouth data via communication devices.

[0161] The server collects review data in real time via websites and APIs. The review data includes metadata such as text, author information, and posting date and time. The collected data is preprocessed by trimming whitespace, removing HTML tags and special characters, and converting all data to lowercase. This unifies the data into a consistent format.

[0162] The server then analyzes the pre-processed review data to detect malicious reviews. This detection is performed using the following techniques:

[0163] 1. Negative Word Check: Scan reviews using a predefined list of negative words (e.g., "terrible," "disappointing," "bad," "poor," etc.) to determine if they contain these words.

[0164] 2. Stealth Marketing Detection: Checks whether certain marketing phrases (e.g., "Buy Now," "Amazing," "Highly Recommended," etc.) are included multiple times in reviews to determine whether they are stealth marketing.

[0165] 3. Sentiment analysis: Using a sentiment analysis tool (e.g., TextBlob), we calculate the sentiment score of the review, and if the sentiment is highly negative, we classify the review as malicious.

[0166] Furthermore, the server is able to learn users' past posting history and account trends, and detect malicious reviews from specific accounts, which allows posts from malicious accounts to be removed before they occur.

[0167] Reviews that are determined to be malicious after these steps are filtered out, and the rest are provided to users as reliable reviews. Real-time processing ensures that users can always refer to the latest, most reliable review information.

[0168] For example, if a user posts a review saying, "This product is amazing, I highly recommend it!", the phrases "amazing" and "highly recommend" appear multiple times, which will be identified as malicious by the stealth marketing detection process. On the other hand, a review saying, "I had a bad experience, the product was defective" will be identified as malicious by negative word checks and sentiment analysis tools. After filtering, only reliable reviews are ultimately displayed to users.

[0169] An example prompt is:

[0170] Generate a Python script to filter out malicious reviews from a list. The script should preprocess the review text, check for predefined negative words, detect stealth marketing phrases, and perform sentiment analysis. Reviews with negative sentiment or containing any of the predefined negative words or stealth marketing phrases should be filtered out.

[0171] Main Hardware:

[0172] Server: AWS EC2, etc.

[0173] Smartphones: iPhone, Android devices

[0174] Main software:

[0175] Programming language: Python

[0176] Sentiment Analysis Tool: TextBlob

[0177] Text processing library: re library

[0178] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0179] Step 1:

[0180] The server collects review data from websites and APIs via communication devices. The input is review data obtained through various interfaces, and the output is raw, unprocessed data.

[0181] Step 2:

[0182] The server preprocesses the collected review data, which includes trimming whitespace, removing HTML tags and special characters, and lowercasing the entire data. The input is raw review data, and the output is cleaned text data.

[0183] Step 3:

[0184] The server determines whether a review is malicious based on the pre-processed review data. This step includes the following specific operations:

[0185] Negative Word Check: The input is the cleaned text data, which is checked against a predefined list of negative words to see if it contains any words that are deemed malicious. The output is a list of text that is determined to be malicious.

[0186] Stealth Marketing Detection: The input is similarly cleaned text data, and it checks for multiple occurrences of specific marketing phrases. The output is a list of suspicious marketing text.

[0187] Sentiment analysis: The input is similarly formatted text data, and a sentiment analysis tool (such as TextBlob) is used to calculate a sentiment score. The output is a list of texts with high negative sentiment scores.

[0188] Step 4:

[0189] The server filters reviews that are determined to be malicious. The input is a list of texts determined to be malicious in step 3, and the output is reliable review data with malicious reviews removed.

[0190] Step 5:

[0191] The server learns the user's past posting history and account trends. In this step, it analyzes previously collected data and learns the patterns of reviews posted by specific accounts. The input is the past posting history and account information, and the output is the trained model.

[0192] Step 6:

[0193] The server detects malicious reviews from specific accounts in real time. Using the model trained in Step 5, the input is newly posted review data, and the output is the result of determining whether the account is malicious.

[0194] Step 7:

[0195] The server displays the filtered and reliable reviews to the user. The input is the filtered review data, and the output is the final review information that the user can view.

[0196] Through these processing steps, users can always view reliable and up-to-date word-of-mouth information.

[0197] 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.

[0198] This invention combines a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews with an emotion engine that recognizes user emotions. The system is centered around a server and operates in the following steps:

[0199] Collecting word-of-mouth data

[0200] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0201] Data Preprocessing

[0202] The server pre-processes the collected review data, which includes text cleaning, such as removing white space, HTML tags, and special characters, to ensure the data is in a consistent format and easier to analyze.

[0203] Detecting malicious reviews

[0204] The server determines whether a review is malicious based on the pre-processed review data. It uses the following method:

[0205] 1. Negative word check: The server scans the review content based on a predefined list of negative words to check whether the word is included. This method detects reviews that contain slander or negative content.

[0206] 2. Stealth Marketing Detection: The server checks whether a particular marketing phrase is included multiple times in a review to determine whether it is stealth marketing. This method identifies potentially inauthentic reviews.

[0207] 3. Sentiment Analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it will classify the review as malicious, thereby detecting reviews with a lot of negative content.

[0208] Review filtering

[0209] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0210] Recognizing user emotions with an emotion engine

[0211] The server uses an emotion engine to analyze the user's reactions when viewing reviews. This analysis calculates the user's emotion score, and based on this, recommended reviews are displayed preferentially.

[0212] Learning user emotion history

[0213] The server learns the user's emotional history and filters reviews that match the user's preferences, allowing the user to see reliable reviews that are optimized for their individual preferences.

[0214] Real-time processing

[0215] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0216] Specific examples

[0217] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious by the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times.

[0218] On the other hand, User B's review, "I had a bad experience, the product was defective.", is determined to be a malicious review through negative word checking and sentiment analysis.

[0219] Furthermore, User C's review, "The product works as expected and I am satisfied," remains a highly reliable review and is given a high positive score by the user's sentiment engine, so it is displayed preferentially to other users.

[0220] In this way, the system of the present invention performs a series of processes in real time, from collecting word-of-mouth data, preprocessing, detecting malicious word-of-mouth, filtering, recognizing user sentiment, to displaying it, thereby providing highly reliable word-of-mouth information.

[0221] The processing flow will be explained below.

[0222] Step 1:

[0223] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This step ensures that the latest review information is always added to the database.

[0224] Step 2:

[0225] The server preprocesses the collected review data, specifically by cleaning the text, removing whitespace and HTML tags, escaping special characters, and converting all text to lowercase. This preprocessing unifies the data into a consistent format, facilitating the next analysis step.

[0226] Step 3:

[0227] The server identifies malicious reviews based on the preprocessed review data. In this step, it first scans the review content using a negative word list to check for the presence of words included in the list (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0228] Step 4:

[0229] The server performs stealth marketing detection by checking whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in the reviews. This step identifies excessive promotional and fake reviews and detects reviews that are considered stealth marketing.

[0230] Step 5:

[0231] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the entire text and identifies positive, negative, or neutral sentiment. If the sentiment score is significantly negative, it determines the review to be malicious. This method detects reviews with a high percentage of negative content.

[0232] Step 6:

[0233] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0234] Step 7:

[0235] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0236] Step 8:

[0237] The server uses an emotion engine to analyze the user's reactions when viewing reviews, calculates the user's emotion score, and prioritizes the display of recommended reviews based on the user's emotional state.

[0238] Step 9:

[0239] The server learns the user's emotional history and filters reviews that match the user's preferences. This step allows users to see reliable reviews that are optimized for their individual preferences.

[0240] Step 10:

[0241] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0242] Step 11:

[0243] The terminal displays the filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0244] Example 2

[0245] 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."

[0246] In today's internet environment, consumers use many review sites and social media to check product and service reviews. However, these reviews often contain malicious information, such as slander, false information, and stealth marketing, making it difficult for consumers to obtain reliable information. Furthermore, reviews are uniformly displayed without considering user sentiment, which can lead to lower user satisfaction. To address these issues, a system is needed that can effectively filter malicious reviews, display only reliable reviews, and recognize user sentiment to provide optimal reviews.

[0247] 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.

[0248] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for analyzing the user's emotions, and means for preferentially displaying word-of-mouth based on the analyzed user's emotion score. This allows the user to not only access reliable word-of-mouth information, but also obtain word-of-mouth information that best suits their emotions.

[0249] "Word-of-mouth data" refers to ratings and comments written by consumers and users about products and services.

[0250] "Means of collection" refers to technologies and tools for automatically obtaining data from websites, APIs, etc. on the Internet.

[0251] "Preprocessing methods" refers to technologies and algorithms that perform processes such as removing white space and special characters, and standardizing data to improve the quality of collected data.

[0252] "Means of determination" refers to the technology or algorithms that use pre-processed data to determine whether it is malicious content.

[0253] "Filtering methods" refers to technologies and algorithms that filter out determined to be malicious data and select only reliable data.

[0254] "Display means" refers to the interface or technology used to visually present the filtered, reliable data to the user.

[0255] "Means of sentiment analysis" refers to technologies and tools for analyzing users' reactions and emotions when viewing data.

[0256] The "emotion score" refers to an index that quantifies the user's emotional response and quantitatively evaluates their emotional state.

[0257] "Preferential display method" refers to technology or algorithms that prioritize displaying certain data to users over other data based on emotional scores.

[0258] "System" refers to the entire technical device that combines and operates the above means.

[0259] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. It also incorporates an emotion engine that recognizes user emotions. The system is centered around a server and operates using the following hardware and software:

[0260] Hardware and Software

[0261] The server is equipped with a high-performance processor and has a large amount of memory and storage. To process HTTP requests from users, the server uses web server software such as Nginx or Apache. The database used is MySQL or PostgreSQL.

[0262] The collection method uses the Python library BeautifulSoup and Requests, which allows data to be collected in real time from review sites and APIs.

[0263] For preprocessing, we use Python regular expressions (re module) and BeautifulSoup to clean the data, removing unnecessary whitespace, HTML tags, and special characters.

[0264] The method uses TextBlob and NLTK libraries to check for negative words and perform sentiment analysis, and scikit-learn to detect stealth marketing.

[0265] The filtering method involves using SQL queries to manipulate the database and remove malicious reviews.

[0266] To display reviews, we use JavaScript frameworks (React and Vue.js) to show users reliable reviews.

[0267] Sentiment analysis involves using tools such as Google Analytics and Mixpanel to collect and analyze user responses and calculate an emotional score.

[0268] The priority display method implements an algorithm to prioritize the display of specific reviews based on their sentiment scores.

[0269] Specific examples

[0270] For example, if User A posts a review saying, "This product is amazing, I highly recommend it!", this review goes through the following process:

[0271] 1. Through collection methods, the server collects reviews from websites and APIs.

[0272] 2. Preprocessing allows the server to remove whitespace, HTML tags, and special characters to ensure the data is in a consistent format.

[0273] 3. Using the determination method, the server detects the phrases "amazing" and "highly recommend" and determines that the review is malicious, possibly representing stealth marketing.

[0274] On the other hand, if User B posts a review saying, "I had a bad experience, the product was defective," this review will go through the following process:

[0275] 1. Data is collected through collection methods.

[0276] 2. The data is cleaned using preprocessing methods.

[0277] 3. Based on the judgment method, the review is determined to be malicious because it contains the negative words "bad" and "defective."

[0278] User C's review, "The product works as expected and I am satisfied," goes through the following process:

[0279] 1. Data is collected through collection methods.

[0280] 2. The data is cleaned using preprocessing methods.

[0281] 3. The evaluation method determines that the review is reliable because it does not contain any negative words.

[0282] 4. Display method This review will be displayed to users.

[0283] Prompt Sentence Examples

[0284] Here are some examples of prompts for generative AI models:

[0285] 1. What text pattern should I set to detect a review that says "This product is amazing, highly recommend it!" as a stealth marketing pattern?

[0286] 2. I would like to detect reviews that say, "I had a bad experience, the product was defective." using a negative word check. Can you please give me a sample negative word list to use?

[0287] In this way, the system can consistently collect and preprocess review data, detect malicious reviews, display reliable reviews, and recognize user emotions.

[0288] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0289] Step 1:

[0290] Collecting word-of-mouth data

[0291] The server collects real-time review data from websites and APIs, including metadata such as text content, author information, and posting date and time.

[0292] Specific behavior:

[0293] Input: URL of the target website or API endpoint

[0294] Data processing: Use the Requests module to get the HTML data of the target web page, parse the HTML with BeautifulSoup, and extract the necessary data.

[0295] Output: The text content, author information, posting date and time, and other metadata are stored on the server in a structured format (e.g., JSON).

[0296] Step 2:

[0297] Data Preprocessing

[0298] The server preprocesses the collected review data, specifically removing whitespace, HTML tags, and special characters.

[0299] Specific behavior:

[0300] Input: Collected raw data (unprocessed word-of-mouth data)

[0301] Data processing: Using Python regular expressions (re module) and BeautifulSoup, we removed unnecessary whitespace, HTML tags, and special characters from the text.

[0302] Output: Cleaned, consistent review data

[0303] Step 3:

[0304] Detecting malicious reviews

[0305] The server detects malicious reviews based on the preprocessed review data using the following method.

[0306] 1. Check for negative words:

[0307] Input: Preprocessed review data

[0308] Data processing: Using TextBlob and NLTK libraries, we matched the negative word list with the review text.

[0309] Output: Reviews flagged as malicious

[0310] 2. Stealth Marketing Detection:

[0311] Input: Preprocessed review data

[0312] Data processing: Using scikit-learn, we count the number of times a specific phrase appears in the review text, and if it exceeds a threshold, we determine it to be stealth marketing.

[0313] Output: Reviews flagged as malicious

[0314] 3. Sentiment analysis:

[0315] Input: Preprocessed review data

[0316] Data processing: Calculates sentiment scores using Google Cloud Natural Language API and TextBlob, and judges messages with high negative sentiment to be malicious.

[0317] Output: Review data with sentiment scores, review data flagged as malicious

[0318] Step 4:

[0319] Review filtering

[0320] The server filters reviews that are determined to be malicious and extracts only reliable reviews.

[0321] Specific behavior:

[0322] Input: Judged review data

[0323] Data processing: Search the database using SQL queries and filter out records that are determined to be malicious.

[0324] Output: Reliable review data

[0325] Step 5:

[0326] Recognizing user emotions with an emotion engine

[0327] The server uses an emotion engine to analyze users' reactions when viewing reviews and calculates an emotion score, which allows the server to prioritize and display reviews based on the user's emotional response.

[0328] Specific behavior:

[0329] Input: Real-time user browsing data (clicks, scrolls, dwell time, etc.)

[0330] Data processing: Analyze user behavior data using tools such as Google Analytics and Mixpanel

[0331] Output: User behavior data with sentiment scores

[0332] Step 6:

[0333] Learning user emotion history

[0334] The server learns the user's emotional history and filters reviews that match the user's preferences.

[0335] Specific behavior:

[0336] Input: User's past emotion score data

[0337] Data processing: Using TensorFlow and PyTorch, we train emotion history using machine learning models.

[0338] Output: User-specific recommendation data based on the trained model

[0339] Step 7:

[0340] Real-time processing

[0341] The server performs all processing in real time, providing the most up-to-date and reliable word-of-mouth information.

[0342] Specific behavior:

[0343] Input: Real-time collected review data

[0344] Data processing: Streaming data using Apache Kafka and performing real-time data analytics using Apache Flink

[0345] Output: Providing up-to-date and reliable review data

[0346] This allows users to always have access to the latest and most reliable word-of-mouth information, as well as to obtain the most appropriate word-of-mouth information that matches their own feelings.

[0347] (Application example 2)

[0348] 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."

[0349] In recent years, online word-of-mouth information has played an important role in consumer purchasing decisions. However, malicious word-of-mouth and stealth marketing can lead to consumers receiving incorrect information. Furthermore, it is difficult to display optimal word-of-mouth information based on user sentiment. Therefore, there is a need for a system that can filter malicious word-of-mouth in real time and display word-of-mouth information based on user sentiment, so that consumers can make appropriate purchasing decisions based on reliable information.

[0350] 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 word-of-mouth data, means for pre-processing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the pre-processed word-of-mouth data, means for filtering malicious word-of-mouth, and means for customizing and displaying the filtered reliable word-of-mouth based on the user's emotion score. This allows users to make purchasing decisions based on reliable word-of-mouth information, and not only is it possible to protect users from malicious word-of-mouth, but it also enables users to view appropriate word-of-mouth information according to their emotions.

[0351] "Review data" refers to information, including ratings and comments, posted by users about products and services on the Internet or other platforms.

[0352] "Collection means" refers to devices or software used to obtain word-of-mouth data via websites or APIs on the Internet.

[0353] "Preprocessing means" refers to devices or software that remove noise and unnecessary information (e.g., white space, HTML tags, special characters) from collected word-of-mouth data and standardize it into a consistent format.

[0354] The "determination means" refers to a device or software that analyzes the pre-processed review data and identifies malicious reviews by detecting negative words and stealth marketing patterns.

[0355] The "filtering means" refers to a device or software that excludes malicious reviews identified by the determination means and extracts only reliable reviews.

[0356] The "customized display means" refers to a device or software that optimizes and displays filtered, reliable reviews based on the user's sentiment score.

[0357] The "user sentiment score" is a numerical value calculated by analyzing the user's emotions and mood, and is information used to customize the order and content of reviews.

[0358] This invention is a system for providing reliable review information that users can use with confidence. The system executes a series of steps in real time to collect, preprocess, judge, filter, and customize the display of review data. The main hardware used is a server, and the software includes an API for collecting review data, a data preprocessing tool, a sentiment analysis tool, and a machine learning model.

[0359] Server roles and processes:

[0360] The server collects review data in real time via websites and APIs. The collected data is stored in a format that includes metadata such as text, author information, and posting date and time. The data is then preprocessed to remove HTML tags and special characters, reduce white space, and so on, so that the data is unified into a consistent format.

[0361] The server analyzes the pre-processed review data to detect patterns of negative words and hidden marketing tactics. To this end, it uses sentiment analysis tools (e.g., TextBlob) to calculate a sentiment score for each review, and also incorporates logic to detect stealth marketing patterns. Reviews that are determined to be malicious are then filtered out, and only reliable reviews are retained.

[0362] Terminal roles and operations:

[0363] When a user browses reviews on their smartphone or other device, the server uses an emotion engine to calculate the user's emotion score, which allows the server to prioritize and display the most relevant reviews based on the user's emotions, thereby encouraging better purchasing behavior.

[0364] Specifically, the server uses an emotion engine to learn the user's emotional history and displays highly reliable written reviews tailored to the user's emotion score. This allows reviews to be updated in real time based on user behavior, ensuring that highly reliable information is continually provided. This system design allows users to make purchasing decisions with confidence based on authentic reviews.

[0365] Examples:

[0366] Let's say User A is browsing products on an online shopping site. If a review is posted saying, "This product is amazing, highly recommend it!", the server will detect the frequent occurrence of the phrases "amazing" and "highly recommend" and determine that this review is malicious. On the other hand, a review saying, "I had a bad experience, the product was defective" will be determined to be negative and malicious through sentiment analysis. However, a review saying, "The product works as expected and I am satisfied" will be determined to be reliable, and will be displayed preferentially if it is positive and has a high user sentiment score. This series of processes allows users to always view the most up-to-date and reliable review information.

[0367] Example prompt sentence:

[0368] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[0369] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0370] Step 1:

[0371] The server collects review data in real time via websites and APIs. The input data is review data including metadata such as text, author information, and posting date and time. The server stores this data in a database and converts it into a format that can be processed by the analytics engine.

[0372] Step 2:

[0373] The server preprocesses the collected review data. Specifically, it removes HTML tags and special characters and reduces white space. The input is the review data collected in step 1, and the output is cleaned, consistently formatted review data. It uses data preprocessing tools (e.g., regular expression libraries).

[0374] Step 3:

[0375] The server identifies malicious reviews based on the preprocessed review data. Specifically, it detects patterns of negative words and marketing techniques. The input is the preprocessed review data, and the output is a list of reviews that are determined to be malicious. A sentiment analysis tool (e.g., TextBlob) is used to calculate a sentiment score and match it with a list of negative words and patterns of marketing techniques.

[0376] Step 4:

[0377] The server filters out malicious reviews. Specifically, it removes malicious reviews determined in step 3 from the review data and extracts only reliable reviews. The input is a list of reviews determined to be malicious and the preprocessed review data, and the output is the filtered, reliable review data.

[0378] Step 5:

[0379] The device displays filtered, reliable reviews to the user. The server uses an emotion engine to calculate the user's emotion score when the user browses reviews. The input data is the user's browsing history and reaction data. The output is the user's emotion score. Based on this emotion score, the review data is prioritized and displayed on the device.

[0380] Step 6:

[0381] The server uses a learning model to analyze the user's sentiment history and optimize future displays. The input is the user's past sentiment score and corresponding review data, and the output is future display suggestions based on the user's sentiment pattern. A generative AI model is used to make predictions and improve user satisfaction.

[0382] Step 7:

[0383] Users can view optimized and reliable reviews and make purchasing decisions. User reactions are fed back to the system and reflected in the next display. Prompts are used to solicit user feedback and improve the accuracy of the system.

[0384] Examples of specific prompts include:

[0385] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[0386] 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.

[0387] 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.

[0388] 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.

[0389] [Second embodiment]

[0390] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0391] 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.

[0392] 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).

[0393] 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.

[0394] 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.

[0395] 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).

[0396] 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.

[0397] 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.

[0398] 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.

[0399] 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.

[0400] 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.

[0401] 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."

[0402] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. The system is configured around a server and operates in the following steps.

[0403] Collecting word-of-mouth data

[0404] The server collects real-time review data via websites and APIs. The review data includes metadata such as text, author information, and posting date and time.

[0405] Data Preprocessing

[0406] The server pre-processes the collected review data, which includes basic text cleaning such as trimming whitespace, removing HTML tags and special characters, and converting everything to lowercase, ensuring the data is formatted consistently.

[0407] Detecting malicious reviews

[0408] The server analyzes the pre-processed review data and detects malicious reviews. This detection is done using several techniques:

[0409] 1. Negative word check: The server scans the review content using a predefined list of negative words (e.g., "worst," "disappointed," "bad," "defective," etc.) to determine whether these words are included.

[0410] 2. Stealth marketing detection: The server checks whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) appear multiple times in the reviews to determine whether they are stealth marketing.

[0411] 3. Sentiment analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it determines the review to be malicious.

[0412] Review filtering

[0413] The server filters out reviews that are deemed malicious and extracts the rest as reliable reviews, allowing users to view only these filtered reviews.

[0414] Learn about past posting history and account trends

[0415] The server learns users' past posting history and account trends, and has the ability to detect patterns of malicious reviews from specific accounts. This allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0416] Real-time processing

[0417] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always view the latest information.

[0418] Specific examples

[0419] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious in the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times. On the other hand, user B's review saying, "I had a bad experience, the product was defective." is identified as malicious through negative word checking and sentiment analysis.

[0420] After filtering, only User C's review, "The product works as expected and I am satisfied," remains as a reliable review and is finally viewable by the user.

[0421] In this way, the system of the present invention collects word-of-mouth data and performs a series of processes from pre-processing, to detecting malicious word-of-mouth, filtering, and displaying, to provide highly reliable word-of-mouth information.

[0422] The processing flow will be explained below.

[0423] Step 1:

[0424] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0425] Step 2:

[0426] The server preprocesses the collected review data, specifically performing basic text cleaning such as removing whitespace and HTML tags, converting everything to lowercase, etc. This step standardizes the data into a consistent format that makes it easier to analyze.

[0427] Step 3:

[0428] The server identifies malicious reviews based on the pre-processed review data. First, it scans the content of the review to see if it contains any of the negative words listed (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0429] Step 4:

[0430] The server performs stealth marketing detection, identifying over-promotional and fake reviews by checking if certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in reviews. This technique identifies potentially inauthentic reviews.

[0431] Step 5:

[0432] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the text to identify positive, negative, and neutral sentiment, and classifies the review as malicious if the sentiment is high. This allows reviews with a high percentage of negative content to be detected.

[0433] Step 6:

[0434] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users view them, and provides only reliable information.

[0435] Step 7:

[0436] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows the server to proactively deal with specific accounts and prevent malicious reviews before they occur.

[0437] Step 8:

[0438] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0439] Step 9:

[0440] The device displays filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0441] Example 1

[0442] 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."

[0443] Word-of-mouth information is an important factor in determining the reputation of products and services on the Internet, but it is often mixed with false information caused by malicious word-of-mouth and stealth marketing. This makes it difficult for users to obtain reliable information. The object of the present invention is to solve these problems and enable users to quickly obtain reliable word-of-mouth information.

[0444] 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.

[0445] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for learning the user's past posting history and account trends to detect patterns of malicious word-of-mouth, and means for executing the collection, preprocessing, determination, and filtering processes in real time, thereby enabling the user to always obtain the latest reliable word-of-mouth information.

[0446] "Word-of-mouth data" is text data that includes user ratings and impressions of products and services.

[0447] "Real-time" means that data processing and information provision is immediate and without delay.

[0448] "Preprocessing" is a series of text cleaning tasks performed to standardize the data format and make it easier to analyze.

[0449] "Malicious reviews" refer to ratings and opinions that contain false or exaggerated content and may mislead other users.

[0450] "Filtering" is the process of sorting data based on specific conditions and eliminating unnecessary data.

[0451] "Reliable reviews" refer to text data that contains accurate and objective evaluations and opinions.

[0452] "Stealth marketing" is a marketing technique that makes products appear as word-of-mouth reviews from ordinary users, when in fact they are for promotional purposes.

[0453] "Collection" is the process of obtaining data from websites and APIs on the internet.

[0454] "Past posting history" is a record of all posts a user has made to date.

[0455] "Account trends" is information that indicates the posting patterns and reliability of a particular user account.

[0456] "Negative words" refer to words that have negative connotations in user ratings and impressions.

[0457] This invention is a system that analyzes word-of-mouth data, filters out malicious word-of-mouth, and displays only reliable word-of-mouth. A specific implementation method of this system will be described below.

[0458] Hardware and Software Use

[0459] The server is the center of the system and works in conjunction with multiple pieces of software. The server mainly uses the following software:

[0460] Operating systems such as Ubuntu

[0461] Apache Kafka: A streaming platform for real-time data processing

[0462] Apache HTTP Server: Web server

[0463] Python: Use various libraries for data processing, sentiment analysis, and filtering

[0464] Requests: A library for web scraping

[0465] BeautifulSoup: Parsing HTML data

[0466] pandas: Data Shaping and Preprocessing

[0467] re (regular expressions): used for text cleaning

[0468] TextBlob: Used for sentiment analysis

[0469] scikit-learn (sklearn): Used to train and predict machine learning models

[0470] Data collection

[0471] The server collects review data in real time via websites and APIs. For example, the server retrieves data from web pages using Python's Requests library and parses it with BeautifulSoup to extract review information, as shown below. Review data can also be retrieved in JSON format using APIs.

[0472] Data Preprocessing

[0473] The server preprocesses the collected review data by using the pandas library to format the data, the regular expressions library (re) to remove HTML tags and special characters, and lowercasing all text to maintain a consistent data format.

[0474] Detecting malicious reviews

[0475] The server performs text analysis to detect malicious reviews, mainly using the following methods:

[0476] 1. Check for negative words:

[0477] The server scans the reviews using a specified list of negative words to see if they contain any of the relevant words.

[0478] 2. Stealth Marketing Detection:

[0479] Check if certain marketing phrases appear multiple times to determine if they are stealth marketing.

[0480] 3. Sentiment analysis:

[0481] We use the TextBlob library to calculate a sentiment score for reviews, and consider reviews with a high negative score to be malicious.

[0482] Review filtering

[0483] The server filters out the bad reviews and extracts only the reliable reviews. To do this, it uses a pandas DataFrame and filters based on a specific column (e.g., a bad flag).

[0484] Learn about past posting history and account trends

[0485] The server learns users' past posting history and account trends, uses a logistic regression model from scikit-learn to learn patterns from the data, and applies the model to new posts to detect malicious accounts.

[0486] Real-time processing

[0487] The server processes data in real time using Apache Kafka, allowing data collection, analysis, filtering, and display to users to be performed consistently in real time.

[0488] Specific examples

[0489] For example, if user A posts a review saying, "This product is amazing, I highly recommend it!", the server will pre-clean the review and then evaluate it using stealth marketing detection techniques. Because phrases like "amazing" and "highly recommend" are used multiple times, this is determined to be a malicious review. On the other hand, if user B posts, "I had a bad experience, the product was defective," this will be determined to be a malicious review through negative word checking and sentiment analysis. Finally, only user C's review, "The product works as expected and I am satisfied," is displayed as a reliable review.

[0490] Prompt Sentence Examples

[0491] You can use the following prompt for your generative AI model: "Please filter the following review sentence to detect malicious reviews: 'This product is amazing, highly recommend it!'"

[0492] As described above, by implementing the present invention, users can quickly obtain highly reliable word-of-mouth information.

[0493] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0494] Step 1: Collect review data

[0495] The server collects review data from websites and APIs. To do this, it uses Python's Requests library to obtain the HTML source and parses it with BeautifulSoup. It also obtains data in JSON format via a RESTful API. The input is the webpage URL or API endpoint, and the output is the collected raw data. Specifically, it obtains the HTML source and extracts review information using BeautifulSoup. When obtaining data from an API, it parses the JSON-formatted data.

[0496] Step 2: Preprocessing the data

[0497] The server preprocesses the collected review data. Specifically, it uses pandas to format the data and regular expressions (re) to remove HTML tags and special characters. It also lowercases the entire text to ensure a consistent data format. The input is the raw data collected in step 1, and the output is the cleaned data. Specifically, it loads the data into a pandas DataFrame, removes whitespace with the str.strip() method, and removes HTML tags with re.sub().

[0498] Step 3: Detecting malicious reviews

[0499] The server analyzes the pre-processed data to detect malicious reviews using the following techniques:

[0500] 1. Negative Word Check: Scan the reviews using a specified list of negative words.

[0501] 2. Stealth marketing detection: Check if a particular marketing phrase appears multiple times.

[0502] 3. Sentiment Analysis: We use the TextBlob library to calculate sentiment scores and consider messages malicious if they have a high negative score.

[0503] The input is preprocessed data, and the output is data flagged as malicious. Specifically, it checks words in the text against a negative word list and flags any matching words. It uses TextBlob to calculate a sentiment score, and if it exceeds a threshold, it is deemed malicious.

[0504] Step 4: Filter reviews

[0505] The server filters data determined to be malicious reviews and extracts only reliable reviews. The input is data flagged as malicious, and the output is reliable data. Specifically, it uses the filtering function of pandas DataFrame to extract only reviews that are not flagged as malicious.

[0506] Step 5: Study past posting history and account trends

[0507] The server learns the user's past posting history and account trends. It uses a machine learning model (such as scikit-learn's logistic regression model) to learn malicious patterns based on past data. The input is the user's past posting history data, and the output is the trained model. Specifically, it collects past data, performs feature engineering, and then trains a logistic regression model.

[0508] Step 6: Real-time processing

[0509] The server processes data in real time using Apache Kafka. All processing (collection, preprocessing, judgment, filtering) is performed instantly, always providing the latest and most reliable word-of-mouth information. The input is word-of-mouth data flowing in real time, and the output is highly reliable word-of-mouth information updated in real time. Specifically, it uses Apache Kafka to continuously receive streaming data and executes each step in real time.

[0510] Through these steps, the system can provide users with reliable word-of-mouth information.

[0511] (Application example 1)

[0512] 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."

[0513] Much of the online word-of-mouth information is unreliable. In particular, the mixture of malicious word-of-mouth and stealth marketing makes it difficult to obtain reliable information. In this situation, it is a difficult task for users to quickly obtain accurate and reliable word-of-mouth information.

[0514] 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.

[0515] In this invention, the server includes means for collecting review data, means for preprocessing the collected review data, means for determining malicious reviews based on the preprocessed review data, means for filtering malicious reviews, means for displaying the filtered, reliable reviews to the user, means for learning the user's past posting history and account trends to detect malicious reviews from specific accounts, and means for processing the review data in real time, thereby enabling the user to always view the latest, reliable review information.

[0516] "Word-of-mouth data" is text information in which Internet users write their opinions and evaluations of a product or service.

[0517] "Preprocessing" refers to the process of cleaning the collected review data, such as trimming white space, removing HTML tags and special characters, and converting the entire data to lowercase, in order to unify it into a consistent format.

[0518] "Malicious reviews" are reviews that are likely to mislead other users, such as those that include excessive use of negative words or patterns that are considered to be stealth marketing.

[0519] "Filtering means" refers to the process of removing malicious reviews from the preprocessed review data and extracting only reliable reviews.

[0520] "Reliable reviews" are reviews that have not been judged to be malicious and contain information that is useful to other users and does not cause misunderstandings.

[0521] "Past posting history" is a record of all review data posted by a particular user in the past.

[0522] "Account trends" refers to consistent patterns or behaviors of a particular user based on their past posting history.

[0523] "Real-time processing means" refers to the process of instantly pre-processing, analyzing, and filtering collected review data to provide users with the most up-to-date information.

[0524] This invention is a system that analyzes word-of-mouth data, filters out malicious reviews, and displays only reliable reviews. This system is mainly composed of a server that collects word-of-mouth data via communication devices.

[0525] The server collects review data in real time via websites and APIs. The review data includes metadata such as text, author information, and posting date and time. The collected data is preprocessed by trimming whitespace, removing HTML tags and special characters, and converting all data to lowercase. This unifies the data into a consistent format.

[0526] The server then analyzes the pre-processed review data to detect malicious reviews. This detection is performed using the following techniques:

[0527] 1. Negative Word Check: Scan reviews using a predefined list of negative words (e.g., "terrible," "disappointing," "bad," "poor," etc.) to determine if they contain these words.

[0528] 2. Stealth Marketing Detection: Checks whether certain marketing phrases (e.g., "Buy Now," "Amazing," "Highly Recommended," etc.) are included multiple times in reviews to determine whether they are stealth marketing.

[0529] 3. Sentiment analysis: Using a sentiment analysis tool (e.g., TextBlob), we calculate the sentiment score of the review, and if the sentiment is highly negative, we classify the review as malicious.

[0530] Furthermore, the server is able to learn users' past posting history and account trends, and detect malicious reviews from specific accounts, which allows posts from malicious accounts to be removed before they occur.

[0531] Reviews that are determined to be malicious after these steps are filtered out, and the rest are provided to users as reliable reviews. Real-time processing ensures that users can always refer to the latest, most reliable review information.

[0532] For example, if a user posts a review saying, "This product is amazing, I highly recommend it!", the phrases "amazing" and "highly recommend" appear multiple times, which will be identified as malicious by the stealth marketing detection process. On the other hand, a review saying, "I had a bad experience, the product was defective" will be identified as malicious by negative word checks and sentiment analysis tools. After filtering, only reliable reviews are ultimately displayed to users.

[0533] An example prompt is:

[0534] Generate a Python script to filter out malicious reviews from a list. The script should preprocess the review text, check for predefined negative words, detect stealth marketing phrases, and perform sentiment analysis. Reviews with negative sentiment or containing any of the predefined negative words or stealth marketing phrases should be filtered out.

[0535] Main Hardware:

[0536] Server: AWS EC2, etc.

[0537] Smartphones: iPhone, Android devices

[0538] Main software:

[0539] Programming language: Python

[0540] Sentiment Analysis Tool: TextBlob

[0541] Text processing library: re library

[0542] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0543] Step 1:

[0544] The server collects review data from websites and APIs via communication devices. The input is review data obtained through various interfaces, and the output is raw, unprocessed data.

[0545] Step 2:

[0546] The server preprocesses the collected review data, which includes trimming whitespace, removing HTML tags and special characters, and lowercasing the entire data. The input is raw review data, and the output is cleaned text data.

[0547] Step 3:

[0548] The server determines whether a review is malicious based on the pre-processed review data. This step includes the following specific operations:

[0549] Negative Word Check: The input is the cleaned text data, which is checked against a predefined list of negative words to see if it contains any words that are deemed malicious. The output is a list of text that is determined to be malicious.

[0550] Stealth Marketing Detection: The input is similarly cleaned text data, and it checks for multiple occurrences of specific marketing phrases. The output is a list of suspicious marketing text.

[0551] Sentiment analysis: The input is similarly formatted text data, and a sentiment analysis tool (such as TextBlob) is used to calculate a sentiment score. The output is a list of texts with high negative sentiment scores.

[0552] Step 4:

[0553] The server filters reviews that are determined to be malicious. The input is a list of texts determined to be malicious in step 3, and the output is reliable review data with malicious reviews removed.

[0554] Step 5:

[0555] The server learns the user's past posting history and account trends. In this step, it analyzes previously collected data and learns the patterns of reviews posted by specific accounts. The input is the past posting history and account information, and the output is the trained model.

[0556] Step 6:

[0557] The server detects malicious reviews from specific accounts in real time. Using the model trained in Step 5, the input is newly posted review data, and the output is the result of determining whether the account is malicious.

[0558] Step 7:

[0559] The server displays the filtered and reliable reviews to the user. The input is the filtered review data, and the output is the final review information that the user can view.

[0560] Through these processing steps, users can always view reliable and up-to-date word-of-mouth information.

[0561] 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.

[0562] This invention combines a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews with an emotion engine that recognizes user emotions. The system is centered around a server and operates in the following steps:

[0563] Collecting word-of-mouth data

[0564] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0565] Data Preprocessing

[0566] The server pre-processes the collected review data, which includes text cleaning, such as removing white space, HTML tags, and special characters, to ensure the data is in a consistent format and easier to analyze.

[0567] Detecting malicious reviews

[0568] The server determines whether a review is malicious based on the pre-processed review data. It uses the following method:

[0569] 1. Negative word check: The server scans the review content based on a predefined list of negative words to check whether the word is included. This method detects reviews that contain slander or negative content.

[0570] 2. Stealth Marketing Detection: The server checks whether a particular marketing phrase is included multiple times in a review to determine whether it is stealth marketing. This method identifies potentially inauthentic reviews.

[0571] 3. Sentiment Analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it will classify the review as malicious, thereby detecting reviews with a lot of negative content.

[0572] Review filtering

[0573] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0574] Recognizing user emotions with an emotion engine

[0575] The server uses an emotion engine to analyze the user's reactions when viewing reviews. This analysis calculates the user's emotion score, and based on this, recommended reviews are displayed preferentially.

[0576] Learning user emotion history

[0577] The server learns the user's emotional history and filters reviews that match the user's preferences, allowing the user to see reliable reviews that are optimized for their individual preferences.

[0578] Real-time processing

[0579] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0580] Specific examples

[0581] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious by the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times.

[0582] On the other hand, User B's review, "I had a bad experience, the product was defective.", is determined to be a malicious review through negative word checking and sentiment analysis.

[0583] Furthermore, User C's review, "The product works as expected and I am satisfied," remains a highly reliable review and is given a high positive score by the user's sentiment engine, so it is displayed preferentially to other users.

[0584] In this way, the system of the present invention performs a series of processes in real time, from collecting word-of-mouth data, preprocessing, detecting malicious word-of-mouth, filtering, recognizing user sentiment, to displaying it, thereby providing highly reliable word-of-mouth information.

[0585] The processing flow will be explained below.

[0586] Step 1:

[0587] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This step ensures that the latest review information is always added to the database.

[0588] Step 2:

[0589] The server preprocesses the collected review data, specifically by cleaning the text, removing whitespace and HTML tags, escaping special characters, and converting all text to lowercase. This preprocessing unifies the data into a consistent format, facilitating the next analysis step.

[0590] Step 3:

[0591] The server identifies malicious reviews based on the preprocessed review data. In this step, it first scans the review content using a negative word list to check for the presence of words included in the list (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0592] Step 4:

[0593] The server performs stealth marketing detection by checking whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in the reviews. This step identifies excessive promotional and fake reviews and detects reviews that are considered stealth marketing.

[0594] Step 5:

[0595] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the entire text and identifies positive, negative, or neutral sentiment. If the sentiment score is significantly negative, it determines the review to be malicious. This method detects reviews with a high percentage of negative content.

[0596] Step 6:

[0597] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0598] Step 7:

[0599] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0600] Step 8:

[0601] The server uses an emotion engine to analyze the user's reactions when viewing reviews, calculates the user's emotion score, and prioritizes the display of recommended reviews based on the user's emotional state.

[0602] Step 9:

[0603] The server learns the user's emotional history and filters reviews that match the user's preferences. This step allows users to see reliable reviews that are optimized for their individual preferences.

[0604] Step 10:

[0605] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0606] Step 11:

[0607] The terminal displays the filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0608] Example 2

[0609] 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."

[0610] In today's internet environment, consumers use many review sites and social media to check product and service reviews. However, these reviews often contain malicious information, such as slander, false information, and stealth marketing, making it difficult for consumers to obtain reliable information. Furthermore, reviews are uniformly displayed without considering user sentiment, which can lead to lower user satisfaction. To address these issues, a system is needed that can effectively filter malicious reviews, display only reliable reviews, and recognize user sentiment to provide optimal reviews.

[0611] 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.

[0612] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for analyzing the user's emotions, and means for preferentially displaying word-of-mouth based on the analyzed user's emotion score. This allows the user to not only access reliable word-of-mouth information, but also obtain word-of-mouth information that best suits their emotions.

[0613] "Word-of-mouth data" refers to ratings and comments written by consumers and users about products and services.

[0614] "Means of collection" refers to technologies and tools for automatically obtaining data from websites, APIs, etc. on the Internet.

[0615] "Preprocessing methods" refers to technologies and algorithms that perform processes such as removing white space and special characters, and standardizing data to improve the quality of collected data.

[0616] "Means of determination" refers to the technology or algorithms that use pre-processed data to determine whether it is malicious content.

[0617] "Filtering methods" refers to technologies and algorithms that filter out determined to be malicious data and select only reliable data.

[0618] "Display means" refers to the interface or technology used to visually present the filtered, reliable data to the user.

[0619] "Means of sentiment analysis" refers to technologies and tools for analyzing users' reactions and emotions when viewing data.

[0620] The "emotion score" refers to an index that quantifies the user's emotional response and quantitatively evaluates their emotional state.

[0621] "Preferential display method" refers to technology or algorithms that prioritize displaying certain data to users over other data based on emotional scores.

[0622] "System" refers to the entire technical device that combines and operates the above means.

[0623] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. It also incorporates an emotion engine that recognizes user emotions. The system is centered around a server and operates using the following hardware and software:

[0624] Hardware and Software

[0625] The server is equipped with a high-performance processor and has a large amount of memory and storage. To process HTTP requests from users, the server uses web server software such as Nginx or Apache. The database used is MySQL or PostgreSQL.

[0626] The collection method uses the Python library BeautifulSoup and Requests, which allows data to be collected in real time from review sites and APIs.

[0627] For preprocessing, we use Python regular expressions (re module) and BeautifulSoup to clean the data, removing unnecessary whitespace, HTML tags, and special characters.

[0628] The method uses TextBlob and NLTK libraries to check for negative words and perform sentiment analysis, and scikit-learn to detect stealth marketing.

[0629] The filtering method involves using SQL queries to manipulate the database and remove malicious reviews.

[0630] To display reviews, we use JavaScript frameworks (React and Vue.js) to show users reliable reviews.

[0631] Sentiment analysis involves using tools such as Google Analytics and Mixpanel to collect and analyze user responses and calculate an emotional score.

[0632] The priority display method implements an algorithm to prioritize the display of specific reviews based on their sentiment scores.

[0633] Specific examples

[0634] For example, if User A posts a review saying, "This product is amazing, I highly recommend it!", this review goes through the following process:

[0635] 1. Through collection methods, the server collects reviews from websites and APIs.

[0636] 2. Preprocessing allows the server to remove whitespace, HTML tags, and special characters to ensure the data is in a consistent format.

[0637] 3. Using the determination method, the server detects the phrases "amazing" and "highly recommend" and determines that the review is malicious, possibly representing stealth marketing.

[0638] On the other hand, if User B posts a review saying, "I had a bad experience, the product was defective," this review will go through the following process:

[0639] 1. Data is collected through collection methods.

[0640] 2. The data is cleaned using preprocessing methods.

[0641] 3. Based on the judgment method, the review is determined to be malicious because it contains the negative words "bad" and "defective."

[0642] User C's review, "The product works as expected and I am satisfied," goes through the following process:

[0643] 1. Data is collected through collection methods.

[0644] 2. The data is cleaned using preprocessing methods.

[0645] 3. The evaluation method determines that the review is reliable because it does not contain any negative words.

[0646] 4. Display method This review will be displayed to users.

[0647] Prompt Sentence Examples

[0648] Here are some examples of prompts for generative AI models:

[0649] 1. What text pattern should I set to detect a review that says "This product is amazing, highly recommend it!" as a stealth marketing pattern?

[0650] 2. I would like to detect reviews that say, "I had a bad experience, the product was defective." using a negative word check. Can you please give me a sample negative word list to use?

[0651] In this way, the system can consistently collect and preprocess review data, detect malicious reviews, display reliable reviews, and recognize user emotions.

[0652] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0653] Step 1:

[0654] Collecting word-of-mouth data

[0655] The server collects real-time review data from websites and APIs, including metadata such as text content, author information, and posting date and time.

[0656] Specific behavior:

[0657] Input: URL of the target website or API endpoint

[0658] Data processing: Use the Requests module to get the HTML data of the target web page, parse the HTML with BeautifulSoup, and extract the necessary data.

[0659] Output: The text content, author information, posting date and time, and other metadata are stored on the server in a structured format (e.g., JSON).

[0660] Step 2:

[0661] Data Preprocessing

[0662] The server preprocesses the collected review data, specifically removing whitespace, HTML tags, and special characters.

[0663] Specific behavior:

[0664] Input: Collected raw data (unprocessed word-of-mouth data)

[0665] Data processing: Using Python regular expressions (re module) and BeautifulSoup, we removed unnecessary whitespace, HTML tags, and special characters from the text.

[0666] Output: Cleaned, consistent review data

[0667] Step 3:

[0668] Detecting malicious reviews

[0669] The server detects malicious reviews based on the preprocessed review data using the following method.

[0670] 1. Check for negative words:

[0671] Input: Preprocessed review data

[0672] Data processing: Using TextBlob and NLTK libraries, we matched the negative word list with the review text.

[0673] Output: Reviews flagged as malicious

[0674] 2. Stealth Marketing Detection:

[0675] Input: Preprocessed review data

[0676] Data processing: Using scikit-learn, we count the number of times a specific phrase appears in the review text, and if it exceeds a threshold, we determine it to be stealth marketing.

[0677] Output: Reviews flagged as malicious

[0678] 3. Sentiment analysis:

[0679] Input: Preprocessed review data

[0680] Data processing: Calculates sentiment scores using Google Cloud Natural Language API and TextBlob, and judges messages with high negative sentiment to be malicious.

[0681] Output: Review data with sentiment scores, review data flagged as malicious

[0682] Step 4:

[0683] Review filtering

[0684] The server filters reviews that are determined to be malicious and extracts only reliable reviews.

[0685] Specific behavior:

[0686] Input: Judged review data

[0687] Data processing: Search the database using SQL queries and filter out records that are determined to be malicious.

[0688] Output: Reliable review data

[0689] Step 5:

[0690] Recognizing user emotions with an emotion engine

[0691] The server uses an emotion engine to analyze users' reactions when viewing reviews and calculates an emotion score, which allows the server to prioritize and display reviews based on the user's emotional response.

[0692] Specific behavior:

[0693] Input: Real-time user browsing data (clicks, scrolls, dwell time, etc.)

[0694] Data processing: Analyze user behavior data using tools such as Google Analytics and Mixpanel

[0695] Output: User behavior data with sentiment scores

[0696] Step 6:

[0697] Learning user emotion history

[0698] The server learns the user's emotional history and filters reviews that match the user's preferences.

[0699] Specific behavior:

[0700] Input: User's past emotion score data

[0701] Data processing: Using TensorFlow and PyTorch, we train emotion history using machine learning models.

[0702] Output: User-specific recommendation data based on the trained model

[0703] Step 7:

[0704] Real-time processing

[0705] The server performs all processing in real time, providing the most up-to-date and reliable word-of-mouth information.

[0706] Specific behavior:

[0707] Input: Real-time collected review data

[0708] Data processing: Streaming data using Apache Kafka and performing real-time data analytics using Apache Flink

[0709] Output: Providing up-to-date and reliable review data

[0710] This allows users to always have access to the latest and most reliable word-of-mouth information, as well as to obtain the most appropriate word-of-mouth information that matches their own feelings.

[0711] (Application example 2)

[0712] 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."

[0713] In recent years, online word-of-mouth information has played an important role in consumer purchasing decisions. However, malicious word-of-mouth and stealth marketing can lead to consumers receiving incorrect information. Furthermore, it is difficult to display optimal word-of-mouth information based on user sentiment. Therefore, there is a need for a system that can filter malicious word-of-mouth in real time and display word-of-mouth information based on user sentiment, so that consumers can make appropriate purchasing decisions based on reliable information.

[0714] 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 word-of-mouth data, means for pre-processing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the pre-processed word-of-mouth data, means for filtering malicious word-of-mouth, and means for customizing and displaying the filtered reliable word-of-mouth based on the user's emotion score. This allows users to make purchasing decisions based on reliable word-of-mouth information, and not only is it possible to protect users from malicious word-of-mouth, but it also enables users to view appropriate word-of-mouth information according to their emotions.

[0715] "Review data" refers to information, including ratings and comments, posted by users about products and services on the Internet or other platforms.

[0716] "Collection means" refers to devices or software used to obtain word-of-mouth data via websites or APIs on the Internet.

[0717] "Preprocessing means" refers to devices or software that remove noise and unnecessary information (e.g., white space, HTML tags, special characters) from collected word-of-mouth data and standardize it into a consistent format.

[0718] The "determination means" refers to a device or software that analyzes the pre-processed review data and identifies malicious reviews by detecting negative words and stealth marketing patterns.

[0719] The "filtering means" refers to a device or software that excludes malicious reviews identified by the determination means and extracts only reliable reviews.

[0720] The "customized display means" refers to a device or software that optimizes and displays filtered, reliable reviews based on the user's sentiment score.

[0721] The "user sentiment score" is a numerical value calculated by analyzing the user's emotions and mood, and is information used to customize the order and content of reviews.

[0722] This invention is a system for providing reliable review information that users can use with confidence. The system executes a series of steps in real time to collect, preprocess, judge, filter, and customize the display of review data. The main hardware used is a server, and the software includes an API for collecting review data, a data preprocessing tool, a sentiment analysis tool, and a machine learning model.

[0723] Server roles and processes:

[0724] The server collects review data in real time via websites and APIs. The collected data is stored in a format that includes metadata such as text, author information, and posting date and time. The data is then preprocessed to remove HTML tags and special characters, reduce white space, and so on, so that the data is unified into a consistent format.

[0725] The server analyzes the pre-processed review data to detect patterns of negative words and hidden marketing tactics. To this end, it uses sentiment analysis tools (e.g., TextBlob) to calculate a sentiment score for each review, and also incorporates logic to detect stealth marketing patterns. Reviews that are determined to be malicious are then filtered out, and only reliable reviews are retained.

[0726] Terminal roles and operations:

[0727] When a user browses reviews on their smartphone or other device, the server uses an emotion engine to calculate the user's emotion score, which allows the server to prioritize and display the most relevant reviews based on the user's emotions, thereby encouraging better purchasing behavior.

[0728] Specifically, the server uses an emotion engine to learn the user's emotional history and displays highly reliable written reviews tailored to the user's emotion score. This allows reviews to be updated in real time based on user behavior, ensuring that highly reliable information is continually provided. This system design allows users to make purchasing decisions with confidence based on authentic reviews.

[0729] Examples:

[0730] Let's say User A is browsing products on an online shopping site. If a review is posted saying, "This product is amazing, highly recommend it!", the server will detect the frequent occurrence of the phrases "amazing" and "highly recommend" and determine that this review is malicious. On the other hand, a review saying, "I had a bad experience, the product was defective" will be determined to be negative and malicious through sentiment analysis. However, a review saying, "The product works as expected and I am satisfied" will be determined to be reliable, and will be displayed preferentially if it is positive and has a high user sentiment score. This series of processes allows users to always view the most up-to-date and reliable review information.

[0731] Example prompt sentence:

[0732] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[0733] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0734] Step 1:

[0735] The server collects review data in real time via websites and APIs. The input data is review data including metadata such as text, author information, and posting date and time. The server stores this data in a database and converts it into a format that can be processed by the analytics engine.

[0736] Step 2:

[0737] The server preprocesses the collected review data. Specifically, it removes HTML tags and special characters and reduces white space. The input is the review data collected in step 1, and the output is cleaned, consistently formatted review data. It uses data preprocessing tools (e.g., regular expression libraries).

[0738] Step 3:

[0739] The server identifies malicious reviews based on the preprocessed review data. Specifically, it detects patterns of negative words and marketing techniques. The input is the preprocessed review data, and the output is a list of reviews that are determined to be malicious. A sentiment analysis tool (e.g., TextBlob) is used to calculate a sentiment score and match it with a list of negative words and patterns of marketing techniques.

[0740] Step 4:

[0741] The server filters out malicious reviews. Specifically, it removes malicious reviews determined in step 3 from the review data and extracts only reliable reviews. The input is a list of reviews determined to be malicious and the preprocessed review data, and the output is the filtered, reliable review data.

[0742] Step 5:

[0743] The device displays filtered, reliable reviews to the user. The server uses an emotion engine to calculate the user's emotion score when the user browses reviews. The input data is the user's browsing history and reaction data. The output is the user's emotion score. Based on this emotion score, the review data is prioritized and displayed on the device.

[0744] Step 6:

[0745] The server uses a learning model to analyze the user's sentiment history and optimize future displays. The input is the user's past sentiment score and corresponding review data, and the output is future display suggestions based on the user's sentiment pattern. A generative AI model is used to make predictions and improve user satisfaction.

[0746] Step 7:

[0747] Users can view optimized and reliable reviews and make purchasing decisions. User reactions are fed back to the system and reflected in the next display. Prompts are used to solicit user feedback and improve the accuracy of the system.

[0748] Examples of specific prompts include:

[0749] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[0750] 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.

[0751] 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.

[0752] 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.

[0753] [Third embodiment]

[0754] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0755] 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.

[0756] 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).

[0757] 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.

[0758] 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.

[0759] 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).

[0760] 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.

[0761] 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.

[0762] 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.

[0763] 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.

[0764] 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.

[0765] 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."

[0766] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. The system is configured around a server and operates in the following steps.

[0767] Collecting word-of-mouth data

[0768] The server collects real-time review data via websites and APIs. The review data includes metadata such as text, author information, and posting date and time.

[0769] Data Preprocessing

[0770] The server pre-processes the collected review data, which includes basic text cleaning such as trimming whitespace, removing HTML tags and special characters, and converting everything to lowercase, ensuring the data is formatted consistently.

[0771] Detecting malicious reviews

[0772] The server analyzes the pre-processed review data and detects malicious reviews. This detection is done using several techniques:

[0773] 1. Negative word check: The server scans the review content using a predefined list of negative words (e.g., "worst," "disappointed," "bad," "defective," etc.) to determine whether these words are included.

[0774] 2. Stealth marketing detection: The server checks whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) appear multiple times in the reviews to determine whether they are stealth marketing.

[0775] 3. Sentiment analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it determines the review to be malicious.

[0776] Review filtering

[0777] The server filters out reviews that are deemed malicious and extracts the rest as reliable reviews, allowing users to view only these filtered reviews.

[0778] Learn about past posting history and account trends

[0779] The server learns users' past posting history and account trends, and has the ability to detect patterns of malicious reviews from specific accounts. This allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0780] Real-time processing

[0781] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always view the latest information.

[0782] Specific examples

[0783] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious in the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times. On the other hand, user B's review saying, "I had a bad experience, the product was defective." is identified as malicious through negative word checking and sentiment analysis.

[0784] After filtering, only User C's review, "The product works as expected and I am satisfied," remains as a reliable review and is finally viewable by the user.

[0785] In this way, the system of the present invention collects word-of-mouth data and performs a series of processes from pre-processing, to detecting malicious word-of-mouth, filtering, and displaying, to provide highly reliable word-of-mouth information.

[0786] The processing flow will be explained below.

[0787] Step 1:

[0788] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0789] Step 2:

[0790] The server preprocesses the collected review data, specifically performing basic text cleaning such as removing whitespace and HTML tags, converting everything to lowercase, etc. This step standardizes the data into a consistent format that makes it easier to analyze.

[0791] Step 3:

[0792] The server identifies malicious reviews based on the pre-processed review data. First, it scans the content of the review to see if it contains any of the negative words listed (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0793] Step 4:

[0794] The server performs stealth marketing detection, identifying over-promotional and fake reviews by checking if certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in reviews. This technique identifies potentially inauthentic reviews.

[0795] Step 5:

[0796] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the text to identify positive, negative, and neutral sentiment, and classifies the review as malicious if the sentiment is high. This allows reviews with a high percentage of negative content to be detected.

[0797] Step 6:

[0798] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users view them, and provides only reliable information.

[0799] Step 7:

[0800] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows the server to proactively deal with specific accounts and prevent malicious reviews before they occur.

[0801] Step 8:

[0802] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0803] Step 9:

[0804] The device displays filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0805] Example 1

[0806] 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."

[0807] Word-of-mouth information is an important factor in determining the reputation of products and services on the Internet, but it is often mixed with false information caused by malicious word-of-mouth and stealth marketing. This makes it difficult for users to obtain reliable information. The object of the present invention is to solve these problems and enable users to quickly obtain reliable word-of-mouth information.

[0808] 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.

[0809] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for learning the user's past posting history and account trends to detect patterns of malicious word-of-mouth, and means for executing the collection, preprocessing, determination, and filtering processes in real time, thereby enabling the user to always obtain the latest reliable word-of-mouth information.

[0810] "Word-of-mouth data" is text data that includes user ratings and impressions of products and services.

[0811] "Real-time" means that data processing and information provision is immediate and without delay.

[0812] "Preprocessing" is a series of text cleaning tasks performed to standardize the data format and make it easier to analyze.

[0813] "Malicious reviews" refer to ratings and opinions that contain false or exaggerated content and may mislead other users.

[0814] "Filtering" is the process of sorting data based on specific conditions and eliminating unnecessary data.

[0815] "Reliable reviews" refer to text data that contains accurate and objective evaluations and opinions.

[0816] "Stealth marketing" is a marketing technique that makes products appear as word-of-mouth reviews from ordinary users, when in fact they are for promotional purposes.

[0817] "Collection" is the process of obtaining data from websites and APIs on the internet.

[0818] "Past posting history" is a record of all posts a user has made to date.

[0819] "Account trends" is information that indicates the posting patterns and reliability of a particular user account.

[0820] "Negative words" refer to words that have negative connotations in user ratings and impressions.

[0821] This invention is a system that analyzes word-of-mouth data, filters out malicious word-of-mouth, and displays only reliable word-of-mouth. A specific implementation method of this system will be described below.

[0822] Hardware and Software Use

[0823] The server is the center of the system and works in conjunction with multiple pieces of software. The server mainly uses the following software:

[0824] Operating systems such as Ubuntu

[0825] Apache Kafka: A streaming platform for real-time data processing

[0826] Apache HTTP Server: Web server

[0827] Python: Use various libraries for data processing, sentiment analysis, and filtering

[0828] Requests: A library for web scraping

[0829] BeautifulSoup: Parsing HTML data

[0830] pandas: Data Shaping and Preprocessing

[0831] re (regular expressions): used for text cleaning

[0832] TextBlob: Used for sentiment analysis

[0833] scikit-learn (sklearn): Used to train and predict machine learning models

[0834] Data collection

[0835] The server collects review data in real time via websites and APIs. For example, the server retrieves data from web pages using Python's Requests library and parses it with BeautifulSoup to extract review information, as shown below. Review data can also be retrieved in JSON format using APIs.

[0836] Data Preprocessing

[0837] The server preprocesses the collected review data by using the pandas library to format the data, the regular expressions library (re) to remove HTML tags and special characters, and lowercasing all text to maintain a consistent data format.

[0838] Detecting malicious reviews

[0839] The server performs text analysis to detect malicious reviews, mainly using the following methods:

[0840] 1. Check for negative words:

[0841] The server scans the reviews using a specified list of negative words to see if they contain any of the relevant words.

[0842] 2. Stealth Marketing Detection:

[0843] Check if certain marketing phrases appear multiple times to determine if they are stealth marketing.

[0844] 3. Sentiment analysis:

[0845] We use the TextBlob library to calculate a sentiment score for reviews, and consider reviews with a high negative score to be malicious.

[0846] Review filtering

[0847] The server filters out the bad reviews and extracts only the reliable reviews. To do this, it uses a pandas DataFrame and filters based on a specific column (e.g., a bad flag).

[0848] Learn about past posting history and account trends

[0849] The server learns users' past posting history and account trends, uses a logistic regression model from scikit-learn to learn patterns from the data, and applies the model to new posts to detect malicious accounts.

[0850] Real-time processing

[0851] The server processes data in real time using Apache Kafka, allowing data collection, analysis, filtering, and display to users to be performed consistently in real time.

[0852] Specific examples

[0853] For example, if user A posts a review saying, "This product is amazing, I highly recommend it!", the server will pre-clean the review and then evaluate it using stealth marketing detection techniques. Because phrases like "amazing" and "highly recommend" are used multiple times, this is determined to be a malicious review. On the other hand, if user B posts, "I had a bad experience, the product was defective," this will be determined to be a malicious review through negative word checking and sentiment analysis. Finally, only user C's review, "The product works as expected and I am satisfied," is displayed as a reliable review.

[0854] Prompt Sentence Examples

[0855] You can use the following prompt for your generative AI model: "Please filter the following review sentence to detect malicious reviews: 'This product is amazing, highly recommend it!'"

[0856] As described above, by implementing the present invention, users can quickly obtain highly reliable word-of-mouth information.

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

[0858] Step 1: Collect review data

[0859] The server collects review data from websites and APIs. To do this, it uses Python's Requests library to obtain the HTML source and parses it with BeautifulSoup. It also obtains data in JSON format via a RESTful API. The input is the webpage URL or API endpoint, and the output is the collected raw data. Specifically, it obtains the HTML source and extracts review information using BeautifulSoup. When obtaining data from an API, it parses the JSON-formatted data.

[0860] Step 2: Preprocessing the data

[0861] The server preprocesses the collected review data. Specifically, it uses pandas to format the data and regular expressions (re) to remove HTML tags and special characters. It also lowercases the entire text to ensure a consistent data format. The input is the raw data collected in step 1, and the output is the cleaned data. Specifically, it loads the data into a pandas DataFrame, removes whitespace with the str.strip() method, and removes HTML tags with re.sub().

[0862] Step 3: Detecting malicious reviews

[0863] The server analyzes the pre-processed data to detect malicious reviews using the following techniques:

[0864] 1. Negative Word Check: Scan the reviews using a specified list of negative words.

[0865] 2. Stealth marketing detection: Check if a particular marketing phrase appears multiple times.

[0866] 3. Sentiment Analysis: We use the TextBlob library to calculate sentiment scores and consider messages malicious if they have a high negative score.

[0867] The input is preprocessed data, and the output is data flagged as malicious. Specifically, it checks words in the text against a negative word list and flags any matching words. It uses TextBlob to calculate a sentiment score, and if it exceeds a threshold, it is deemed malicious.

[0868] Step 4: Filter reviews

[0869] The server filters data determined to be malicious reviews and extracts only reliable reviews. The input is data flagged as malicious, and the output is reliable data. Specifically, it uses the filtering function of pandas DataFrame to extract only reviews that are not flagged as malicious.

[0870] Step 5: Study past posting history and account trends

[0871] The server learns the user's past posting history and account trends. It uses a machine learning model (such as scikit-learn's logistic regression model) to learn malicious patterns based on past data. The input is the user's past posting history data, and the output is the trained model. Specifically, it collects past data, performs feature engineering, and then trains a logistic regression model.

[0872] Step 6: Real-time processing

[0873] The server processes data in real time using Apache Kafka. All processing (collection, preprocessing, judgment, filtering) is performed instantly, always providing the latest and most reliable word-of-mouth information. The input is word-of-mouth data flowing in real time, and the output is highly reliable word-of-mouth information updated in real time. Specifically, it uses Apache Kafka to continuously receive streaming data and executes each step in real time.

[0874] Through these steps, the system can provide users with reliable word-of-mouth information.

[0875] (Application example 1)

[0876] 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."

[0877] Much of the online word-of-mouth information is unreliable. In particular, the mixture of malicious word-of-mouth and stealth marketing makes it difficult to obtain reliable information. In this situation, it is a difficult task for users to quickly obtain accurate and reliable word-of-mouth information.

[0878] 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.

[0879] In this invention, the server includes means for collecting review data, means for preprocessing the collected review data, means for determining malicious reviews based on the preprocessed review data, means for filtering malicious reviews, means for displaying the filtered, reliable reviews to the user, means for learning the user's past posting history and account trends to detect malicious reviews from specific accounts, and means for processing the review data in real time, thereby enabling the user to always view the latest, reliable review information.

[0880] "Word-of-mouth data" is text information in which Internet users write their opinions and evaluations of a product or service.

[0881] "Preprocessing" refers to the process of cleaning the collected review data, such as trimming white space, removing HTML tags and special characters, and converting the entire data to lowercase, in order to unify it into a consistent format.

[0882] "Malicious reviews" are reviews that are likely to mislead other users, such as those that include excessive use of negative words or patterns that are considered to be stealth marketing.

[0883] "Filtering means" refers to the process of removing malicious reviews from the preprocessed review data and extracting only reliable reviews.

[0884] "Reliable reviews" are reviews that have not been judged to be malicious and contain information that is useful to other users and does not cause misunderstandings.

[0885] "Past posting history" is a record of all review data posted by a particular user in the past.

[0886] "Account trends" refers to consistent patterns or behaviors of a particular user based on their past posting history.

[0887] "Real-time processing means" refers to the process of instantly pre-processing, analyzing, and filtering collected review data to provide users with the most up-to-date information.

[0888] This invention is a system that analyzes word-of-mouth data, filters out malicious reviews, and displays only reliable reviews. This system is mainly composed of a server that collects word-of-mouth data via communication devices.

[0889] The server collects review data in real time via websites and APIs. The review data includes metadata such as text, author information, and posting date and time. The collected data is preprocessed by trimming whitespace, removing HTML tags and special characters, and converting all data to lowercase. This unifies the data into a consistent format.

[0890] The server then analyzes the pre-processed review data to detect malicious reviews. This detection is performed using the following techniques:

[0891] 1. Negative Word Check: Scan reviews using a predefined list of negative words (e.g., "terrible," "disappointing," "bad," "poor," etc.) to determine if they contain these words.

[0892] 2. Stealth Marketing Detection: Checks whether certain marketing phrases (e.g., "Buy Now," "Amazing," "Highly Recommended," etc.) are included multiple times in reviews to determine whether they are stealth marketing.

[0893] 3. Sentiment analysis: Using a sentiment analysis tool (e.g., TextBlob), we calculate the sentiment score of the review, and if the sentiment is highly negative, we classify the review as malicious.

[0894] Furthermore, the server is able to learn users' past posting history and account trends, and detect malicious reviews from specific accounts, which allows posts from malicious accounts to be removed before they occur.

[0895] Reviews that are determined to be malicious after these steps are filtered out, and the rest are provided to users as reliable reviews. Real-time processing ensures that users can always refer to the latest, most reliable review information.

[0896] For example, if a user posts a review saying, "This product is amazing, I highly recommend it!", the phrases "amazing" and "highly recommend" appear multiple times, which will be identified as malicious by the stealth marketing detection process. On the other hand, a review saying, "I had a bad experience, the product was defective" will be identified as malicious by negative word checks and sentiment analysis tools. After filtering, only reliable reviews are ultimately displayed to users.

[0897] An example prompt is:

[0898] Generate a Python script to filter out malicious reviews from a list. The script should preprocess the review text, check for predefined negative words, detect stealth marketing phrases, and perform sentiment analysis. Reviews with negative sentiment or containing any of the predefined negative words or stealth marketing phrases should be filtered out.

[0899] Main Hardware:

[0900] Server: AWS EC2, etc.

[0901] Smartphones: iPhone, Android devices

[0902] Main software:

[0903] Programming language: Python

[0904] Sentiment Analysis Tool: TextBlob

[0905] Text processing library: re library

[0906] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0907] Step 1:

[0908] The server collects review data from websites and APIs via communication devices. The input is review data obtained through various interfaces, and the output is raw, unprocessed data.

[0909] Step 2:

[0910] The server preprocesses the collected review data, which includes trimming whitespace, removing HTML tags and special characters, and lowercasing the entire data. The input is raw review data, and the output is cleaned text data.

[0911] Step 3:

[0912] The server determines whether a review is malicious based on the pre-processed review data. This step includes the following specific operations:

[0913] Negative Word Check: The input is the cleaned text data, which is checked against a predefined list of negative words to see if it contains any words that are deemed malicious. The output is a list of text that is determined to be malicious.

[0914] Stealth Marketing Detection: The input is similarly cleaned text data, and it checks for multiple occurrences of specific marketing phrases. The output is a list of suspicious marketing text.

[0915] Sentiment analysis: The input is similarly formatted text data, and a sentiment analysis tool (such as TextBlob) is used to calculate a sentiment score. The output is a list of texts with high negative sentiment scores.

[0916] Step 4:

[0917] The server filters reviews that are determined to be malicious. The input is a list of texts determined to be malicious in step 3, and the output is reliable review data with malicious reviews removed.

[0918] Step 5:

[0919] The server learns the user's past posting history and account trends. In this step, it analyzes previously collected data and learns the patterns of reviews posted by specific accounts. The input is the past posting history and account information, and the output is the trained model.

[0920] Step 6:

[0921] The server detects malicious reviews from specific accounts in real time. Using the model trained in Step 5, the input is newly posted review data, and the output is the result of determining whether the account is malicious.

[0922] Step 7:

[0923] The server displays the filtered and reliable reviews to the user. The input is the filtered review data, and the output is the final review information that the user can view.

[0924] Through these processing steps, users can always view reliable and up-to-date word-of-mouth information.

[0925] 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.

[0926] This invention combines a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews with an emotion engine that recognizes user emotions. The system is centered around a server and operates in the following steps:

[0927] Collecting word-of-mouth data

[0928] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[0929] Data Preprocessing

[0930] The server pre-processes the collected review data, which includes text cleaning, such as removing white space, HTML tags, and special characters, to ensure the data is in a consistent format and easier to analyze.

[0931] Detecting malicious reviews

[0932] The server determines whether a review is malicious based on the pre-processed review data. It uses the following method:

[0933] 1. Negative word check: The server scans the review content based on a predefined list of negative words to check whether the word is included. This method detects reviews that contain slander or negative content.

[0934] 2. Stealth Marketing Detection: The server checks whether a particular marketing phrase is included multiple times in a review to determine whether it is stealth marketing. This method identifies potentially inauthentic reviews.

[0935] 3. Sentiment Analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it will classify the review as malicious, thereby detecting reviews with a lot of negative content.

[0936] Review filtering

[0937] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0938] Recognizing user emotions with an emotion engine

[0939] The server uses an emotion engine to analyze the user's reactions when viewing reviews. This analysis calculates the user's emotion score, and based on this, recommended reviews are displayed preferentially.

[0940] Learning user emotion history

[0941] The server learns the user's emotional history and filters reviews that match the user's preferences, allowing the user to see reliable reviews that are optimized for their individual preferences.

[0942] Real-time processing

[0943] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0944] Specific examples

[0945] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious by the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times.

[0946] On the other hand, User B's review, "I had a bad experience, the product was defective.", is determined to be a malicious review through negative word checking and sentiment analysis.

[0947] Furthermore, User C's review, "The product works as expected and I am satisfied," remains a highly reliable review and is given a high positive score by the user's sentiment engine, so it is displayed preferentially to other users.

[0948] In this way, the system of the present invention performs a series of processes in real time, from collecting word-of-mouth data, preprocessing, detecting malicious word-of-mouth, filtering, recognizing user sentiment, to displaying it, thereby providing highly reliable word-of-mouth information.

[0949] The processing flow will be explained below.

[0950] Step 1:

[0951] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This step ensures that the latest review information is always added to the database.

[0952] Step 2:

[0953] The server preprocesses the collected review data, specifically by cleaning the text, removing whitespace and HTML tags, escaping special characters, and converting all text to lowercase. This preprocessing unifies the data into a consistent format, facilitating the next analysis step.

[0954] Step 3:

[0955] The server identifies malicious reviews based on the preprocessed review data. In this step, it first scans the review content using a negative word list to check for the presence of words included in the list (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[0956] Step 4:

[0957] The server performs stealth marketing detection by checking whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in the reviews. This step identifies excessive promotional and fake reviews and detects reviews that are considered stealth marketing.

[0958] Step 5:

[0959] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the entire text and identifies positive, negative, or neutral sentiment. If the sentiment score is significantly negative, it determines the review to be malicious. This method detects reviews with a high percentage of negative content.

[0960] Step 6:

[0961] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[0962] Step 7:

[0963] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[0964] Step 8:

[0965] The server uses an emotion engine to analyze the user's reactions when viewing reviews, calculates the user's emotion score, and prioritizes the display of recommended reviews based on the user's emotional state.

[0966] Step 9:

[0967] The server learns the user's emotional history and filters reviews that match the user's preferences. This step allows users to see reliable reviews that are optimized for their individual preferences.

[0968] Step 10:

[0969] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[0970] Step 11:

[0971] The terminal displays the filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[0972] Example 2

[0973] 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."

[0974] In today's internet environment, consumers use many review sites and social media to check product and service reviews. However, these reviews often contain malicious information, such as slander, false information, and stealth marketing, making it difficult for consumers to obtain reliable information. Furthermore, reviews are uniformly displayed without considering user sentiment, which can lead to lower user satisfaction. To address these issues, a system is needed that can effectively filter malicious reviews, display only reliable reviews, and recognize user sentiment to provide optimal reviews.

[0975] 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.

[0976] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for analyzing the user's emotions, and means for preferentially displaying word-of-mouth based on the analyzed user's emotion score. This allows the user to not only access reliable word-of-mouth information, but also obtain word-of-mouth information that best suits their emotions.

[0977] "Word-of-mouth data" refers to ratings and comments written by consumers and users about products and services.

[0978] "Means of collection" refers to technologies and tools for automatically obtaining data from websites, APIs, etc. on the Internet.

[0979] "Preprocessing methods" refers to technologies and algorithms that perform processes such as removing white space and special characters, and standardizing data to improve the quality of collected data.

[0980] "Means of determination" refers to the technology or algorithms that use pre-processed data to determine whether it is malicious content.

[0981] "Filtering methods" refers to technologies and algorithms that filter out determined to be malicious data and select only reliable data.

[0982] "Display means" refers to the interface or technology used to visually present the filtered, reliable data to the user.

[0983] "Means of sentiment analysis" refers to technologies and tools for analyzing users' reactions and emotions when viewing data.

[0984] The "emotion score" refers to an index that quantifies the user's emotional response and quantitatively evaluates their emotional state.

[0985] "Preferential display method" refers to technology or algorithms that prioritize displaying certain data to users over other data based on emotional scores.

[0986] "System" refers to the entire technical device that combines and operates the above means.

[0987] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. It also incorporates an emotion engine that recognizes user emotions. The system is centered around a server and operates using the following hardware and software:

[0988] Hardware and Software

[0989] The server is equipped with a high-performance processor and has a large amount of memory and storage. To process HTTP requests from users, the server uses web server software such as Nginx or Apache. The database used is MySQL or PostgreSQL.

[0990] The collection method uses the Python library BeautifulSoup and Requests, which allows data to be collected in real time from review sites and APIs.

[0991] For preprocessing, we use Python regular expressions (re module) and BeautifulSoup to clean the data, removing unnecessary whitespace, HTML tags, and special characters.

[0992] The method uses TextBlob and NLTK libraries to check for negative words and perform sentiment analysis, and scikit-learn to detect stealth marketing.

[0993] The filtering method involves using SQL queries to manipulate the database and remove malicious reviews.

[0994] To display reviews, we use JavaScript frameworks (React and Vue.js) to show users reliable reviews.

[0995] Sentiment analysis involves using tools such as Google Analytics and Mixpanel to collect and analyze user responses and calculate an emotional score.

[0996] The priority display method implements an algorithm to prioritize the display of specific reviews based on their sentiment scores.

[0997] Specific examples

[0998] For example, if User A posts a review saying, "This product is amazing, I highly recommend it!", this review goes through the following process:

[0999] 1. Through collection methods, the server collects reviews from websites and APIs.

[1000] 2. Preprocessing allows the server to remove whitespace, HTML tags, and special characters to ensure the data is in a consistent format.

[1001] 3. Using the determination method, the server detects the phrases "amazing" and "highly recommend" and determines that the review is malicious, possibly representing stealth marketing.

[1002] On the other hand, if User B posts a review saying, "I had a bad experience, the product was defective," this review will go through the following process:

[1003] 1. Data is collected through collection methods.

[1004] 2. The data is cleaned using preprocessing methods.

[1005] 3. Based on the judgment method, the review is determined to be malicious because it contains the negative words "bad" and "defective."

[1006] User C's review, "The product works as expected and I am satisfied," goes through the following process:

[1007] 1. Data is collected through collection methods.

[1008] 2. The data is cleaned using preprocessing methods.

[1009] 3. The evaluation method determines that the review is reliable because it does not contain any negative words.

[1010] 4. Display method This review will be displayed to users.

[1011] Prompt Sentence Examples

[1012] Here are some examples of prompts for generative AI models:

[1013] 1. What text pattern should I set to detect a review that says "This product is amazing, highly recommend it!" as a stealth marketing pattern?

[1014] 2. I would like to detect reviews that say, "I had a bad experience, the product was defective." using a negative word check. Can you please give me a sample negative word list to use?

[1015] In this way, the system can consistently collect and preprocess review data, detect malicious reviews, display reliable reviews, and recognize user emotions.

[1016] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1017] Step 1:

[1018] Collecting word-of-mouth data

[1019] The server collects real-time review data from websites and APIs, including metadata such as text content, author information, and posting date and time.

[1020] Specific behavior:

[1021] Input: URL of the target website or API endpoint

[1022] Data processing: Use the Requests module to get the HTML data of the target web page, parse the HTML with BeautifulSoup, and extract the necessary data.

[1023] Output: The text content, author information, posting date and time, and other metadata are stored on the server in a structured format (e.g., JSON).

[1024] Step 2:

[1025] Data Preprocessing

[1026] The server preprocesses the collected review data, specifically removing whitespace, HTML tags, and special characters.

[1027] Specific behavior:

[1028] Input: Collected raw data (unprocessed word-of-mouth data)

[1029] Data processing: Using Python regular expressions (re module) and BeautifulSoup, we removed unnecessary whitespace, HTML tags, and special characters from the text.

[1030] Output: Cleaned, consistent review data

[1031] Step 3:

[1032] Detecting malicious reviews

[1033] The server detects malicious reviews based on the preprocessed review data using the following method.

[1034] 1. Check for negative words:

[1035] Input: Preprocessed review data

[1036] Data processing: Using TextBlob and NLTK libraries, we matched the negative word list with the review text.

[1037] Output: Reviews flagged as malicious

[1038] 2. Stealth Marketing Detection:

[1039] Input: Preprocessed review data

[1040] Data processing: Using scikit-learn, we count the number of times a specific phrase appears in the review text, and if it exceeds a threshold, we determine it to be stealth marketing.

[1041] Output: Reviews flagged as malicious

[1042] 3. Sentiment analysis:

[1043] Input: Preprocessed review data

[1044] Data processing: Calculates sentiment scores using Google Cloud Natural Language API and TextBlob, and judges messages with high negative sentiment to be malicious.

[1045] Output: Review data with sentiment scores, review data flagged as malicious

[1046] Step 4:

[1047] Review filtering

[1048] The server filters reviews that are determined to be malicious and extracts only reliable reviews.

[1049] Specific behavior:

[1050] Input: Judged review data

[1051] Data processing: Search the database using SQL queries and filter out records that are determined to be malicious.

[1052] Output: Reliable review data

[1053] Step 5:

[1054] Recognizing user emotions with an emotion engine

[1055] The server uses an emotion engine to analyze users' reactions when viewing reviews and calculates an emotion score, which allows the server to prioritize and display reviews based on the user's emotional response.

[1056] Specific behavior:

[1057] Input: Real-time user browsing data (clicks, scrolls, dwell time, etc.)

[1058] Data processing: Analyze user behavior data using tools such as Google Analytics and Mixpanel

[1059] Output: User behavior data with sentiment scores

[1060] Step 6:

[1061] Learning user emotion history

[1062] The server learns the user's emotional history and filters reviews that match the user's preferences.

[1063] Specific behavior:

[1064] Input: User's past emotion score data

[1065] Data processing: Using TensorFlow and PyTorch, we train emotion history using machine learning models.

[1066] Output: User-specific recommendation data based on the trained model

[1067] Step 7:

[1068] Real-time processing

[1069] The server performs all processing in real time, providing the most up-to-date and reliable word-of-mouth information.

[1070] Specific behavior:

[1071] Input: Real-time collected review data

[1072] Data processing: Streaming data using Apache Kafka and performing real-time data analytics using Apache Flink

[1073] Output: Providing up-to-date and reliable review data

[1074] This allows users to always have access to the latest and most reliable word-of-mouth information, as well as to obtain the most appropriate word-of-mouth information that matches their own feelings.

[1075] (Application example 2)

[1076] 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."

[1077] In recent years, online word-of-mouth information has played an important role in consumer purchasing decisions. However, malicious word-of-mouth and stealth marketing can lead to consumers receiving incorrect information. Furthermore, it is difficult to display optimal word-of-mouth information based on user sentiment. Therefore, there is a need for a system that can filter malicious word-of-mouth in real time and display word-of-mouth information based on user sentiment, so that consumers can make appropriate purchasing decisions based on reliable information.

[1078] 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 word-of-mouth data, means for pre-processing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the pre-processed word-of-mouth data, means for filtering malicious word-of-mouth, and means for customizing and displaying the filtered reliable word-of-mouth based on the user's emotion score. This allows users to make purchasing decisions based on reliable word-of-mouth information, and not only is it possible to protect users from malicious word-of-mouth, but it also enables users to view appropriate word-of-mouth information according to their emotions.

[1079] "Review data" refers to information, including ratings and comments, posted by users about products and services on the Internet or other platforms.

[1080] "Collection means" refers to devices or software used to obtain word-of-mouth data via websites or APIs on the Internet.

[1081] "Preprocessing means" refers to devices or software that remove noise and unnecessary information (e.g., white space, HTML tags, special characters) from collected word-of-mouth data and standardize it into a consistent format.

[1082] The "determination means" refers to a device or software that analyzes the pre-processed review data and identifies malicious reviews by detecting negative words and stealth marketing patterns.

[1083] The "filtering means" refers to a device or software that excludes malicious reviews identified by the determination means and extracts only reliable reviews.

[1084] The "customized display means" refers to a device or software that optimizes and displays filtered, reliable reviews based on the user's sentiment score.

[1085] The "user sentiment score" is a numerical value calculated by analyzing the user's emotions and mood, and is information used to customize the order and content of reviews.

[1086] This invention is a system for providing reliable review information that users can use with confidence. The system executes a series of steps in real time to collect, preprocess, judge, filter, and customize the display of review data. The main hardware used is a server, and the software includes an API for collecting review data, a data preprocessing tool, a sentiment analysis tool, and a machine learning model.

[1087] Server roles and processes:

[1088] The server collects review data in real time via websites and APIs. The collected data is stored in a format that includes metadata such as text, author information, and posting date and time. The data is then preprocessed to remove HTML tags and special characters, reduce white space, and so on, so that the data is unified into a consistent format.

[1089] The server analyzes the pre-processed review data to detect patterns of negative words and hidden marketing tactics. To this end, it uses sentiment analysis tools (e.g., TextBlob) to calculate a sentiment score for each review, and also incorporates logic to detect stealth marketing patterns. Reviews that are determined to be malicious are then filtered out, and only reliable reviews are retained.

[1090] Terminal roles and operations:

[1091] When a user browses reviews on their smartphone or other device, the server uses an emotion engine to calculate the user's emotion score, which allows the server to prioritize and display the most relevant reviews based on the user's emotions, thereby encouraging better purchasing behavior.

[1092] Specifically, the server uses an emotion engine to learn the user's emotional history and displays highly reliable written reviews tailored to the user's emotion score. This allows reviews to be updated in real time based on user behavior, ensuring that highly reliable information is continually provided. This system design allows users to make purchasing decisions with confidence based on authentic reviews.

[1093] Examples:

[1094] Let's say User A is browsing products on an online shopping site. If a review is posted saying, "This product is amazing, highly recommend it!", the server will detect the frequent occurrence of the phrases "amazing" and "highly recommend" and determine that this review is malicious. On the other hand, a review saying, "I had a bad experience, the product was defective" will be determined to be negative and malicious through sentiment analysis. However, a review saying, "The product works as expected and I am satisfied" will be determined to be reliable, and will be displayed preferentially if it is positive and has a high user sentiment score. This series of processes allows users to always view the most up-to-date and reliable review information.

[1095] Example prompt sentence:

[1096] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[1097] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1098] Step 1:

[1099] The server collects review data in real time via websites and APIs. The input data is review data including metadata such as text, author information, and posting date and time. The server stores this data in a database and converts it into a format that can be processed by the analytics engine.

[1100] Step 2:

[1101] The server preprocesses the collected review data. Specifically, it removes HTML tags and special characters and reduces white space. The input is the review data collected in step 1, and the output is cleaned, consistently formatted review data. It uses data preprocessing tools (e.g., regular expression libraries).

[1102] Step 3:

[1103] The server identifies malicious reviews based on the preprocessed review data. Specifically, it detects patterns of negative words and marketing techniques. The input is the preprocessed review data, and the output is a list of reviews that are determined to be malicious. A sentiment analysis tool (e.g., TextBlob) is used to calculate a sentiment score and match it with a list of negative words and patterns of marketing techniques.

[1104] Step 4:

[1105] The server filters out malicious reviews. Specifically, it removes malicious reviews determined in step 3 from the review data and extracts only reliable reviews. The input is a list of reviews determined to be malicious and the preprocessed review data, and the output is the filtered, reliable review data.

[1106] Step 5:

[1107] The device displays filtered, reliable reviews to the user. The server uses an emotion engine to calculate the user's emotion score when the user browses reviews. The input data is the user's browsing history and reaction data. The output is the user's emotion score. Based on this emotion score, the review data is prioritized and displayed on the device.

[1108] Step 6:

[1109] The server uses a learning model to analyze the user's sentiment history and optimize future displays. The input is the user's past sentiment score and corresponding review data, and the output is future display suggestions based on the user's sentiment pattern. A generative AI model is used to make predictions and improve user satisfaction.

[1110] Step 7:

[1111] Users can view optimized and reliable reviews and make purchasing decisions. User reactions are fed back to the system and reflected in the next display. Prompts are used to solicit user feedback and improve the accuracy of the system.

[1112] Examples of specific prompts include:

[1113] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[1114] 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.

[1115] 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.

[1116] 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.

[1117] [Fourth embodiment]

[1118] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1119] 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.

[1120] 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).

[1121] 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.

[1122] 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.

[1123] 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).

[1124] 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.

[1125] 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.

[1126] 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.

[1127] 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.

[1128] 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.

[1129] 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.

[1130] 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."

[1131] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. The system is configured around a server and operates in the following steps.

[1132] Collecting word-of-mouth data

[1133] The server collects real-time review data via websites and APIs. The review data includes metadata such as text, author information, and posting date and time.

[1134] Data Preprocessing

[1135] The server pre-processes the collected review data, which includes basic text cleaning such as trimming whitespace, removing HTML tags and special characters, and converting everything to lowercase, ensuring the data is formatted consistently.

[1136] Detecting malicious reviews

[1137] The server analyzes the pre-processed review data and detects malicious reviews. This detection is done using several techniques:

[1138] 1. Negative word check: The server scans the review content using a predefined list of negative words (e.g., "worst," "disappointed," "bad," "defective," etc.) to determine whether these words are included.

[1139] 2. Stealth marketing detection: The server checks whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) appear multiple times in the reviews to determine whether they are stealth marketing.

[1140] 3. Sentiment analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it determines the review to be malicious.

[1141] Review filtering

[1142] The server filters out reviews that are deemed malicious and extracts the rest as reliable reviews, allowing users to view only these filtered reviews.

[1143] Learn about past posting history and account trends

[1144] The server learns users' past posting history and account trends, and has the ability to detect patterns of malicious reviews from specific accounts. This allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[1145] Real-time processing

[1146] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always view the latest information.

[1147] Specific examples

[1148] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious in the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times. On the other hand, user B's review saying, "I had a bad experience, the product was defective." is identified as malicious through negative word checking and sentiment analysis.

[1149] After filtering, only User C's review, "The product works as expected and I am satisfied," remains as a reliable review and is finally viewable by the user.

[1150] In this way, the system of the present invention collects word-of-mouth data and performs a series of processes from pre-processing, to detecting malicious word-of-mouth, filtering, and displaying, to provide highly reliable word-of-mouth information.

[1151] The processing flow will be explained below.

[1152] Step 1:

[1153] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[1154] Step 2:

[1155] The server preprocesses the collected review data, specifically performing basic text cleaning such as removing whitespace and HTML tags, converting everything to lowercase, etc. This step standardizes the data into a consistent format that makes it easier to analyze.

[1156] Step 3:

[1157] The server identifies malicious reviews based on the pre-processed review data. First, it scans the content of the review to see if it contains any of the negative words listed (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[1158] Step 4:

[1159] The server performs stealth marketing detection, identifying over-promotional and fake reviews by checking if certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in reviews. This technique identifies potentially inauthentic reviews.

[1160] Step 5:

[1161] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the text to identify positive, negative, and neutral sentiment, and classifies the review as malicious if the sentiment is high. This allows reviews with a high percentage of negative content to be detected.

[1162] Step 6:

[1163] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users view them, and provides only reliable information.

[1164] Step 7:

[1165] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows the server to proactively deal with specific accounts and prevent malicious reviews before they occur.

[1166] Step 8:

[1167] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[1168] Step 9:

[1169] The device displays filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[1170] Example 1

[1171] 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."

[1172] Word-of-mouth information is an important factor in determining the reputation of products and services on the Internet, but it is often mixed with false information caused by malicious word-of-mouth and stealth marketing. This makes it difficult for users to obtain reliable information. The object of the present invention is to solve these problems and enable users to quickly obtain reliable word-of-mouth information.

[1173] 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.

[1174] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for learning the user's past posting history and account trends to detect patterns of malicious word-of-mouth, and means for executing the collection, preprocessing, determination, and filtering processes in real time, thereby enabling the user to always obtain the latest reliable word-of-mouth information.

[1175] "Word-of-mouth data" is text data that includes user ratings and impressions of products and services.

[1176] "Real-time" means that data processing and information provision is immediate and without delay.

[1177] "Preprocessing" is a series of text cleaning tasks performed to standardize the data format and make it easier to analyze.

[1178] "Malicious reviews" refer to ratings and opinions that contain false or exaggerated content and may mislead other users.

[1179] "Filtering" is the process of sorting data based on specific conditions and eliminating unnecessary data.

[1180] "Reliable reviews" refer to text data that contains accurate and objective evaluations and opinions.

[1181] "Stealth marketing" is a marketing technique that makes products appear as word-of-mouth reviews from ordinary users, when in fact they are for promotional purposes.

[1182] "Collection" is the process of obtaining data from websites and APIs on the internet.

[1183] "Past posting history" is a record of all posts a user has made to date.

[1184] "Account trends" is information that indicates the posting patterns and reliability of a particular user account.

[1185] "Negative words" refer to words that have negative connotations in user ratings and impressions.

[1186] This invention is a system that analyzes word-of-mouth data, filters out malicious word-of-mouth, and displays only reliable word-of-mouth. A specific implementation method of this system will be described below.

[1187] Hardware and Software Use

[1188] The server is the center of the system and works in conjunction with multiple pieces of software. The server mainly uses the following software:

[1189] Operating systems such as Ubuntu

[1190] Apache Kafka: A streaming platform for real-time data processing

[1191] Apache HTTP Server: Web server

[1192] Python: Use various libraries for data processing, sentiment analysis, and filtering

[1193] Requests: A library for web scraping

[1194] BeautifulSoup: Parsing HTML data

[1195] pandas: Data Shaping and Preprocessing

[1196] re (regular expressions): used for text cleaning

[1197] TextBlob: Used for sentiment analysis

[1198] scikit-learn (sklearn): Used to train and predict machine learning models

[1199] Data collection

[1200] The server collects review data in real time via websites and APIs. For example, the server retrieves data from web pages using Python's Requests library and parses it with BeautifulSoup to extract review information, as shown below. Review data can also be retrieved in JSON format using APIs.

[1201] Data Preprocessing

[1202] The server preprocesses the collected review data by using the pandas library to format the data, the regular expressions library (re) to remove HTML tags and special characters, and lowercasing all text to maintain a consistent data format.

[1203] Detecting malicious reviews

[1204] The server performs text analysis to detect malicious reviews, mainly using the following methods:

[1205] 1. Check for negative words:

[1206] The server scans the reviews using a specified list of negative words to see if they contain any of the relevant words.

[1207] 2. Stealth Marketing Detection:

[1208] Check if certain marketing phrases appear multiple times to determine if they are stealth marketing.

[1209] 3. Sentiment analysis:

[1210] We use the TextBlob library to calculate a sentiment score for reviews, and consider reviews with a high negative score to be malicious.

[1211] Review filtering

[1212] The server filters out the bad reviews and extracts only the reliable reviews. To do this, it uses a pandas DataFrame and filters based on a specific column (e.g., a bad flag).

[1213] Learn about past posting history and account trends

[1214] The server learns users' past posting history and account trends, uses a logistic regression model from scikit-learn to learn patterns from the data, and applies the model to new posts to detect malicious accounts.

[1215] Real-time processing

[1216] The server processes data in real time using Apache Kafka, allowing data collection, analysis, filtering, and display to users to be performed consistently in real time.

[1217] Specific examples

[1218] For example, if user A posts a review saying, "This product is amazing, I highly recommend it!", the server will pre-clean the review and then evaluate it using stealth marketing detection techniques. Because phrases like "amazing" and "highly recommend" are used multiple times, this is determined to be a malicious review. On the other hand, if user B posts, "I had a bad experience, the product was defective," this will be determined to be a malicious review through negative word checking and sentiment analysis. Finally, only user C's review, "The product works as expected and I am satisfied," is displayed as a reliable review.

[1219] Prompt Sentence Examples

[1220] You can use the following prompt for your generative AI model: "Please filter the following review sentence to detect malicious reviews: 'This product is amazing, highly recommend it!'"

[1221] As described above, by implementing the present invention, users can quickly obtain highly reliable word-of-mouth information.

[1222] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1223] Step 1: Collect review data

[1224] The server collects review data from websites and APIs. To do this, it uses Python's Requests library to obtain the HTML source and parses it with BeautifulSoup. It also obtains data in JSON format via a RESTful API. The input is the webpage URL or API endpoint, and the output is the collected raw data. Specifically, it obtains the HTML source and extracts review information using BeautifulSoup. When obtaining data from an API, it parses the JSON-formatted data.

[1225] Step 2: Preprocessing the data

[1226] The server preprocesses the collected review data. Specifically, it uses pandas to format the data and regular expressions (re) to remove HTML tags and special characters. It also lowercases the entire text to ensure a consistent data format. The input is the raw data collected in step 1, and the output is the cleaned data. Specifically, it loads the data into a pandas DataFrame, removes whitespace with the str.strip() method, and removes HTML tags with re.sub().

[1227] Step 3: Detecting malicious reviews

[1228] The server analyzes the pre-processed data to detect malicious reviews using the following techniques:

[1229] 1. Negative Word Check: Scan the reviews using a specified list of negative words.

[1230] 2. Stealth marketing detection: Check if a particular marketing phrase appears multiple times.

[1231] 3. Sentiment Analysis: We use the TextBlob library to calculate sentiment scores and consider messages malicious if they have a high negative score.

[1232] The input is preprocessed data, and the output is data flagged as malicious. Specifically, it checks words in the text against a negative word list and flags any matching words. It uses TextBlob to calculate a sentiment score, and if it exceeds a threshold, it is deemed malicious.

[1233] Step 4: Filter reviews

[1234] The server filters data determined to be malicious reviews and extracts only reliable reviews. The input is data flagged as malicious, and the output is reliable data. Specifically, it uses the filtering function of pandas DataFrame to extract only reviews that are not flagged as malicious.

[1235] Step 5: Study past posting history and account trends

[1236] The server learns the user's past posting history and account trends. It uses a machine learning model (such as scikit-learn's logistic regression model) to learn malicious patterns based on past data. The input is the user's past posting history data, and the output is the trained model. Specifically, it collects past data, performs feature engineering, and then trains a logistic regression model.

[1237] Step 6: Real-time processing

[1238] The server processes data in real time using Apache Kafka. All processing (collection, preprocessing, judgment, filtering) is performed instantly, always providing the latest and most reliable word-of-mouth information. The input is word-of-mouth data flowing in real time, and the output is highly reliable word-of-mouth information updated in real time. Specifically, it uses Apache Kafka to continuously receive streaming data and executes each step in real time.

[1239] Through these steps, the system can provide users with reliable word-of-mouth information.

[1240] (Application example 1)

[1241] 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."

[1242] Much of the online word-of-mouth information is unreliable. In particular, the mixture of malicious word-of-mouth and stealth marketing makes it difficult to obtain reliable information. In this situation, it is a difficult task for users to quickly obtain accurate and reliable word-of-mouth information.

[1243] 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.

[1244] In this invention, the server includes means for collecting review data, means for preprocessing the collected review data, means for determining malicious reviews based on the preprocessed review data, means for filtering malicious reviews, means for displaying the filtered, reliable reviews to the user, means for learning the user's past posting history and account trends to detect malicious reviews from specific accounts, and means for processing the review data in real time, thereby enabling the user to always view the latest, reliable review information.

[1245] "Word-of-mouth data" is text information in which Internet users write their opinions and evaluations of a product or service.

[1246] "Preprocessing" refers to the process of cleaning the collected review data, such as trimming white space, removing HTML tags and special characters, and converting the entire data to lowercase, in order to unify it into a consistent format.

[1247] "Malicious reviews" are reviews that are likely to mislead other users, such as those that include excessive use of negative words or patterns that are considered to be stealth marketing.

[1248] "Filtering means" refers to the process of removing malicious reviews from the preprocessed review data and extracting only reliable reviews.

[1249] "Reliable reviews" are reviews that have not been judged to be malicious and contain information that is useful to other users and does not cause misunderstandings.

[1250] "Past posting history" is a record of all review data posted by a particular user in the past.

[1251] "Account trends" refers to consistent patterns or behaviors of a particular user based on their past posting history.

[1252] "Real-time processing means" refers to the process of instantly pre-processing, analyzing, and filtering collected review data to provide users with the most up-to-date information.

[1253] This invention is a system that analyzes word-of-mouth data, filters out malicious reviews, and displays only reliable reviews. This system is mainly composed of a server that collects word-of-mouth data via communication devices.

[1254] The server collects review data in real time via websites and APIs. The review data includes metadata such as text, author information, and posting date and time. The collected data is preprocessed by trimming whitespace, removing HTML tags and special characters, and converting all data to lowercase. This unifies the data into a consistent format.

[1255] The server then analyzes the pre-processed review data to detect malicious reviews. This detection is performed using the following techniques:

[1256] 1. Negative Word Check: Scan reviews using a predefined list of negative words (e.g., "terrible," "disappointing," "bad," "poor," etc.) to determine if they contain these words.

[1257] 2. Stealth Marketing Detection: Checks whether certain marketing phrases (e.g., "Buy Now," "Amazing," "Highly Recommended," etc.) are included multiple times in reviews to determine whether they are stealth marketing.

[1258] 3. Sentiment analysis: Using a sentiment analysis tool (e.g., TextBlob), we calculate the sentiment score of the review, and if the sentiment is highly negative, we classify the review as malicious.

[1259] Furthermore, the server is able to learn users' past posting history and account trends, and detect malicious reviews from specific accounts, which allows posts from malicious accounts to be removed before they occur.

[1260] Reviews that are determined to be malicious after these steps are filtered out, and the rest are provided to users as reliable reviews. Real-time processing ensures that users can always refer to the latest, most reliable review information.

[1261] For example, if a user posts a review saying, "This product is amazing, I highly recommend it!", the phrases "amazing" and "highly recommend" appear multiple times, which will be identified as malicious by the stealth marketing detection process. On the other hand, a review saying, "I had a bad experience, the product was defective" will be identified as malicious by negative word checks and sentiment analysis tools. After filtering, only reliable reviews are ultimately displayed to users.

[1262] An example prompt is:

[1263] Generate a Python script to filter out malicious reviews from a list. The script should preprocess the review text, check for predefined negative words, detect stealth marketing phrases, and perform sentiment analysis. Reviews with negative sentiment or containing any of the predefined negative words or stealth marketing phrases should be filtered out.

[1264] Main Hardware:

[1265] Server: AWS EC2, etc.

[1266] Smartphones: iPhone, Android devices

[1267] Main software:

[1268] Programming language: Python

[1269] Sentiment Analysis Tool: TextBlob

[1270] Text processing library: re library

[1271] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1272] Step 1:

[1273] The server collects review data from websites and APIs via communication devices. The input is review data obtained through various interfaces, and the output is raw, unprocessed data.

[1274] Step 2:

[1275] The server preprocesses the collected review data, which includes trimming whitespace, removing HTML tags and special characters, and lowercasing the entire data. The input is raw review data, and the output is cleaned text data.

[1276] Step 3:

[1277] The server determines whether a review is malicious based on the pre-processed review data. This step includes the following specific operations:

[1278] Negative Word Check: The input is the cleaned text data, which is checked against a predefined list of negative words to see if it contains any words that are deemed malicious. The output is a list of text that is determined to be malicious.

[1279] Stealth Marketing Detection: The input is similarly cleaned text data, and it checks for multiple occurrences of specific marketing phrases. The output is a list of suspicious marketing text.

[1280] Sentiment analysis: The input is similarly formatted text data, and a sentiment analysis tool (such as TextBlob) is used to calculate a sentiment score. The output is a list of texts with high negative sentiment scores.

[1281] Step 4:

[1282] The server filters reviews that are determined to be malicious. The input is a list of texts determined to be malicious in step 3, and the output is reliable review data with malicious reviews removed.

[1283] Step 5:

[1284] The server learns the user's past posting history and account trends. In this step, it analyzes previously collected data and learns the patterns of reviews posted by specific accounts. The input is the past posting history and account information, and the output is the trained model.

[1285] Step 6:

[1286] The server detects malicious reviews from specific accounts in real time. Using the model trained in Step 5, the input is newly posted review data, and the output is the result of determining whether the account is malicious.

[1287] Step 7:

[1288] The server displays the filtered and reliable reviews to the user. The input is the filtered review data, and the output is the final review information that the user can view.

[1289] Through these processing steps, users can always view reliable and up-to-date word-of-mouth information.

[1290] 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.

[1291] This invention combines a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews with an emotion engine that recognizes user emotions. The system is centered around a server and operates in the following steps:

[1292] Collecting word-of-mouth data

[1293] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This ensures that the latest review information is always added to the database.

[1294] Data Preprocessing

[1295] The server pre-processes the collected review data, which includes text cleaning, such as removing white space, HTML tags, and special characters, to ensure the data is in a consistent format and easier to analyze.

[1296] Detecting malicious reviews

[1297] The server determines whether a review is malicious based on the pre-processed review data. It uses the following method:

[1298] 1. Negative word check: The server scans the review content based on a predefined list of negative words to check whether the word is included. This method detects reviews that contain slander or negative content.

[1299] 2. Stealth Marketing Detection: The server checks whether a particular marketing phrase is included multiple times in a review to determine whether it is stealth marketing. This method identifies potentially inauthentic reviews.

[1300] 3. Sentiment Analysis: The server uses a sentiment analysis tool to calculate the sentiment score of the review, and if the sentiment is highly negative, it will classify the review as malicious, thereby detecting reviews with a lot of negative content.

[1301] Review filtering

[1302] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[1303] Recognizing user emotions with an emotion engine

[1304] The server uses an emotion engine to analyze the user's reactions when viewing reviews. This analysis calculates the user's emotion score, and based on this, recommended reviews are displayed preferentially.

[1305] Learning user emotion history

[1306] The server learns the user's emotional history and filters reviews that match the user's preferences, allowing the user to see reliable reviews that are optimized for their individual preferences.

[1307] Real-time processing

[1308] The server performs these processes in real time to provide the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[1309] Specific examples

[1310] For example, suppose user A posts a review saying, "This product is amazing, highly recommend it!" After preprocessing, this review is identified as malicious by the stealth marketing detection process because the phrases "amazing" and "highly recommend" appear multiple times.

[1311] On the other hand, User B's review, "I had a bad experience, the product was defective.", is determined to be a malicious review through negative word checking and sentiment analysis.

[1312] Furthermore, User C's review, "The product works as expected and I am satisfied," remains a highly reliable review and is given a high positive score by the user's sentiment engine, so it is displayed preferentially to other users.

[1313] In this way, the system of the present invention performs a series of processes in real time, from collecting word-of-mouth data, preprocessing, detecting malicious word-of-mouth, filtering, recognizing user sentiment, to displaying it, thereby providing highly reliable word-of-mouth information.

[1314] The processing flow will be explained below.

[1315] Step 1:

[1316] The server collects review data in real time via websites and APIs. The collected data includes metadata such as text, author information, and posting date and time. This step ensures that the latest review information is always added to the database.

[1317] Step 2:

[1318] The server preprocesses the collected review data, specifically by cleaning the text, removing whitespace and HTML tags, escaping special characters, and converting all text to lowercase. This preprocessing unifies the data into a consistent format, facilitating the next analysis step.

[1319] Step 3:

[1320] The server identifies malicious reviews based on the preprocessed review data. In this step, it first scans the review content using a negative word list to check for the presence of words included in the list (e.g., "worst," "disappointed," "bad," "defective," etc.). This method detects reviews that contain slander or negative content.

[1321] Step 4:

[1322] The server performs stealth marketing detection by checking whether certain marketing phrases (e.g., "Buy this now," "amazing," "highly recommend," etc.) are included multiple times in the reviews. This step identifies excessive promotional and fake reviews and detects reviews that are considered stealth marketing.

[1323] Step 5:

[1324] The server uses a sentiment analysis tool to calculate a sentiment score for each review. It analyzes the entire text and identifies positive, negative, or neutral sentiment. If the sentiment score is significantly negative, it determines the review to be malicious. This method detects reviews with a high percentage of negative content.

[1325] Step 6:

[1326] The server filters out the determined malicious reviews and extracts only reliable reviews. This filtering process eliminates malicious reviews when users browse, and provides only reliable information.

[1327] Step 7:

[1328] The server learns the user's past posting history and account trends, and detects patterns of malicious reviews from specific accounts. This step allows it to identify posts from malicious accounts and eliminate malicious reviews before they occur.

[1329] Step 8:

[1330] The server uses an emotion engine to analyze the user's reactions when viewing reviews, calculates the user's emotion score, and prioritizes the display of recommended reviews based on the user's emotional state.

[1331] Step 9:

[1332] The server learns the user's emotional history and filters reviews that match the user's preferences. This step allows users to see reliable reviews that are optimized for their individual preferences.

[1333] Step 10:

[1334] The server performs these processes in real time, providing the most up-to-date and reliable word-of-mouth information, allowing users to always refer to the latest word-of-mouth information.

[1335] Step 11:

[1336] The terminal displays the filtered reviews to the user, allowing the user to evaluate products and services based on reliable review information.

[1337] Example 2

[1338] 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."

[1339] In today's internet environment, consumers use many review sites and social media to check product and service reviews. However, these reviews often contain malicious information, such as slander, false information, and stealth marketing, making it difficult for consumers to obtain reliable information. Furthermore, reviews are uniformly displayed without considering user sentiment, which can lead to lower user satisfaction. To address these issues, a system is needed that can effectively filter malicious reviews, display only reliable reviews, and recognize user sentiment to provide optimal reviews.

[1340] 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.

[1341] In this invention, the server includes means for collecting word-of-mouth data, means for preprocessing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the preprocessed word-of-mouth data, means for filtering malicious word-of-mouth, means for displaying the filtered reliable word-of-mouth to the user, means for analyzing the user's emotions, and means for preferentially displaying word-of-mouth based on the analyzed user's emotion score. This allows the user to not only access reliable word-of-mouth information, but also obtain word-of-mouth information that best suits their emotions.

[1342] "Word-of-mouth data" refers to ratings and comments written by consumers and users about products and services.

[1343] "Means of collection" refers to technologies and tools for automatically obtaining data from websites, APIs, etc. on the Internet.

[1344] "Preprocessing methods" refers to technologies and algorithms that perform processes such as removing white space and special characters, and standardizing data to improve the quality of collected data.

[1345] "Means of determination" refers to the technology or algorithms that use pre-processed data to determine whether it is malicious content.

[1346] "Filtering methods" refers to technologies and algorithms that filter out determined to be malicious data and select only reliable data.

[1347] "Display means" refers to the interface or technology used to visually present the filtered, reliable data to the user.

[1348] "Means of sentiment analysis" refers to technologies and tools for analyzing users' reactions and emotions when viewing data.

[1349] The "emotion score" refers to an index that quantifies the user's emotional response and quantitatively evaluates their emotional state.

[1350] "Preferential display method" refers to technology or algorithms that prioritize displaying certain data to users over other data based on emotional scores.

[1351] "System" refers to the entire technical device that combines and operates the above means.

[1352] This invention is a system that analyzes review data, filters out malicious reviews, and displays only reliable reviews. It also incorporates an emotion engine that recognizes user emotions. The system is centered around a server and operates using the following hardware and software:

[1353] Hardware and Software

[1354] The server is equipped with a high-performance processor and has a large amount of memory and storage. To process HTTP requests from users, the server uses web server software such as Nginx or Apache. The database used is MySQL or PostgreSQL.

[1355] The collection method uses the Python library BeautifulSoup and Requests, which allows data to be collected in real time from review sites and APIs.

[1356] For preprocessing, we use Python regular expressions (re module) and BeautifulSoup to clean the data, removing unnecessary whitespace, HTML tags, and special characters.

[1357] The method uses TextBlob and NLTK libraries to check for negative words and perform sentiment analysis, and scikit-learn to detect stealth marketing.

[1358] The filtering method involves using SQL queries to manipulate the database and remove malicious reviews.

[1359] To display reviews, we use JavaScript frameworks (React and Vue.js) to show users reliable reviews.

[1360] Sentiment analysis involves using tools such as Google Analytics and Mixpanel to collect and analyze user responses and calculate an emotional score.

[1361] The priority display method implements an algorithm to prioritize the display of specific reviews based on their sentiment scores.

[1362] Specific examples

[1363] For example, if User A posts a review saying, "This product is amazing, I highly recommend it!", this review goes through the following process:

[1364] 1. Through collection methods, the server collects reviews from websites and APIs.

[1365] 2. Preprocessing allows the server to remove whitespace, HTML tags, and special characters to ensure the data is in a consistent format.

[1366] 3. Using the determination method, the server detects the phrases "amazing" and "highly recommend" and determines that the review is malicious, possibly representing stealth marketing.

[1367] On the other hand, if User B posts a review saying, "I had a bad experience, the product was defective," this review will go through the following process:

[1368] 1. Data is collected through collection methods.

[1369] 2. The data is cleaned using preprocessing methods.

[1370] 3. Based on the judgment method, the review is determined to be malicious because it contains the negative words "bad" and "defective."

[1371] User C's review, "The product works as expected and I am satisfied," goes through the following process:

[1372] 1. Data is collected through collection methods.

[1373] 2. The data is cleaned using preprocessing methods.

[1374] 3. The evaluation method determines that the review is reliable because it does not contain any negative words.

[1375] 4. Display method This review will be displayed to users.

[1376] Prompt Sentence Examples

[1377] Here are some examples of prompts for generative AI models:

[1378] 1. What text pattern should I set to detect a review that says "This product is amazing, highly recommend it!" as a stealth marketing pattern?

[1379] 2. I would like to detect reviews that say, "I had a bad experience, the product was defective." using a negative word check. Can you please give me a sample negative word list to use?

[1380] In this way, the system can consistently collect and preprocess review data, detect malicious reviews, display reliable reviews, and recognize user emotions.

[1381] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1382] Step 1:

[1383] Collecting word-of-mouth data

[1384] The server collects real-time review data from websites and APIs, including metadata such as text content, author information, and posting date and time.

[1385] Specific behavior:

[1386] Input: URL of the target website or API endpoint

[1387] Data processing: Use the Requests module to get the HTML data of the target web page, parse the HTML with BeautifulSoup, and extract the necessary data.

[1388] Output: The text content, author information, posting date and time, and other metadata are stored on the server in a structured format (e.g., JSON).

[1389] Step 2:

[1390] Data Preprocessing

[1391] The server preprocesses the collected review data, specifically removing whitespace, HTML tags, and special characters.

[1392] Specific behavior:

[1393] Input: Collected raw data (unprocessed word-of-mouth data)

[1394] Data processing: Using Python regular expressions (re module) and BeautifulSoup, we removed unnecessary whitespace, HTML tags, and special characters from the text.

[1395] Output: Cleaned, consistent review data

[1396] Step 3:

[1397] Detecting malicious reviews

[1398] The server detects malicious reviews based on the preprocessed review data using the following method.

[1399] 1. Check for negative words:

[1400] Input: Preprocessed review data

[1401] Data processing: Using TextBlob and NLTK libraries, we matched the negative word list with the review text.

[1402] Output: Reviews flagged as malicious

[1403] 2. Stealth Marketing Detection:

[1404] Input: Preprocessed review data

[1405] Data processing: Using scikit-learn, we count the number of times a specific phrase appears in the review text, and if it exceeds a threshold, we determine it to be stealth marketing.

[1406] Output: Reviews flagged as malicious

[1407] 3. Sentiment analysis:

[1408] Input: Preprocessed review data

[1409] Data processing: Calculates sentiment scores using Google Cloud Natural Language API and TextBlob, and judges messages with high negative sentiment to be malicious.

[1410] Output: Review data with sentiment scores, review data flagged as malicious

[1411] Step 4:

[1412] Review filtering

[1413] The server filters reviews that are determined to be malicious and extracts only reliable reviews.

[1414] Specific behavior:

[1415] Input: Judged review data

[1416] Data processing: Search the database using SQL queries and filter out records that are determined to be malicious.

[1417] Output: Reliable review data

[1418] Step 5:

[1419] Recognizing user emotions with an emotion engine

[1420] The server uses an emotion engine to analyze users' reactions when viewing reviews and calculates an emotion score, which allows the server to prioritize and display reviews based on the user's emotional response.

[1421] Specific behavior:

[1422] Input: Real-time user browsing data (clicks, scrolls, dwell time, etc.)

[1423] Data processing: Analyze user behavior data using tools such as Google Analytics and Mixpanel

[1424] Output: User behavior data with sentiment scores

[1425] Step 6:

[1426] Learning user emotion history

[1427] The server learns the user's emotional history and filters reviews that match the user's preferences.

[1428] Specific behavior:

[1429] Input: User's past emotion score data

[1430] Data processing: Using TensorFlow and PyTorch, we train emotion history using machine learning models.

[1431] Output: User-specific recommendation data based on the trained model

[1432] Step 7:

[1433] Real-time processing

[1434] The server performs all processing in real time, providing the most up-to-date and reliable word-of-mouth information.

[1435] Specific behavior:

[1436] Input: Real-time collected review data

[1437] Data processing: Streaming data using Apache Kafka and performing real-time data analytics using Apache Flink

[1438] Output: Providing up-to-date and reliable review data

[1439] This allows users to always have access to the latest and most reliable word-of-mouth information, as well as to obtain the most appropriate word-of-mouth information that matches their own feelings.

[1440] (Application example 2)

[1441] 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."

[1442] In recent years, online word-of-mouth information has played an important role in consumer purchasing decisions. However, malicious word-of-mouth and stealth marketing can lead to consumers receiving incorrect information. Furthermore, it is difficult to display optimal word-of-mouth information based on user sentiment. Therefore, there is a need for a system that can filter malicious word-of-mouth in real time and display word-of-mouth information based on user sentiment, so that consumers can make appropriate purchasing decisions based on reliable information.

[1443] 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 word-of-mouth data, means for pre-processing the collected word-of-mouth data, means for determining malicious word-of-mouth based on the pre-processed word-of-mouth data, means for filtering malicious word-of-mouth, and means for customizing and displaying the filtered reliable word-of-mouth based on the user's emotion score. This allows users to make purchasing decisions based on reliable word-of-mouth information, and not only is it possible to protect users from malicious word-of-mouth, but it also enables users to view appropriate word-of-mouth information according to their emotions.

[1444] "Review data" refers to information, including ratings and comments, posted by users about products and services on the Internet or other platforms.

[1445] "Collection means" refers to devices or software used to obtain word-of-mouth data via websites or APIs on the Internet.

[1446] "Preprocessing means" refers to devices or software that remove noise and unnecessary information (e.g., white space, HTML tags, special characters) from collected word-of-mouth data and standardize it into a consistent format.

[1447] The "determination means" refers to a device or software that analyzes the pre-processed review data and identifies malicious reviews by detecting negative words and stealth marketing patterns.

[1448] The "filtering means" refers to a device or software that excludes malicious reviews identified by the determination means and extracts only reliable reviews.

[1449] The "customized display means" refers to a device or software that optimizes and displays filtered, reliable reviews based on the user's sentiment score.

[1450] The "user sentiment score" is a numerical value calculated by analyzing the user's emotions and mood, and is information used to customize the order and content of reviews.

[1451] This invention is a system for providing reliable review information that users can use with confidence. The system executes a series of steps in real time to collect, preprocess, judge, filter, and customize the display of review data. The main hardware used is a server, and the software includes an API for collecting review data, a data preprocessing tool, a sentiment analysis tool, and a machine learning model.

[1452] Server roles and processes:

[1453] The server collects review data in real time via websites and APIs. The collected data is stored in a format that includes metadata such as text, author information, and posting date and time. The data is then preprocessed to remove HTML tags and special characters, reduce white space, and so on, so that the data is unified into a consistent format.

[1454] The server analyzes the pre-processed review data to detect patterns of negative words and hidden marketing tactics. To this end, it uses sentiment analysis tools (e.g., TextBlob) to calculate a sentiment score for each review, and also incorporates logic to detect stealth marketing patterns. Reviews that are determined to be malicious are then filtered out, and only reliable reviews are retained.

[1455] Terminal roles and operations:

[1456] When a user browses reviews on their smartphone or other device, the server uses an emotion engine to calculate the user's emotion score, which allows the server to prioritize and display the most relevant reviews based on the user's emotions, thereby encouraging better purchasing behavior.

[1457] Specifically, the server uses an emotion engine to learn the user's emotional history and displays highly reliable written reviews tailored to the user's emotion score. This allows reviews to be updated in real time based on user behavior, ensuring that highly reliable information is continually provided. This system design allows users to make purchasing decisions with confidence based on authentic reviews.

[1458] Examples:

[1459] Let's say User A is browsing products on an online shopping site. If a review is posted saying, "This product is amazing, highly recommend it!", the server will detect the frequent occurrence of the phrases "amazing" and "highly recommend" and determine that this review is malicious. On the other hand, a review saying, "I had a bad experience, the product was defective" will be determined to be negative and malicious through sentiment analysis. However, a review saying, "The product works as expected and I am satisfied" will be determined to be reliable, and will be displayed preferentially if it is positive and has a high user sentiment score. This series of processes allows users to always view the most up-to-date and reliable review information.

[1460] Example prompt sentence:

[1461] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[1462] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1463] Step 1:

[1464] The server collects review data in real time via websites and APIs. The input data is review data including metadata such as text, author information, and posting date and time. The server stores this data in a database and converts it into a format that can be processed by the analytics engine.

[1465] Step 2:

[1466] The server preprocesses the collected review data. Specifically, it removes HTML tags and special characters and reduces white space. The input is the review data collected in step 1, and the output is cleaned, consistently formatted review data. It uses data preprocessing tools (e.g., regular expression libraries).

[1467] Step 3:

[1468] The server identifies malicious reviews based on the preprocessed review data. Specifically, it detects patterns of negative words and marketing techniques. The input is the preprocessed review data, and the output is a list of reviews that are determined to be malicious. A sentiment analysis tool (e.g., TextBlob) is used to calculate a sentiment score and match it with a list of negative words and patterns of marketing techniques.

[1469] Step 4:

[1470] The server filters out malicious reviews. Specifically, it removes malicious reviews determined in step 3 from the review data and extracts only reliable reviews. The input is a list of reviews determined to be malicious and the preprocessed review data, and the output is the filtered, reliable review data.

[1471] Step 5:

[1472] The device displays filtered, reliable reviews to the user. The server uses an emotion engine to calculate the user's emotion score when the user browses reviews. The input data is the user's browsing history and reaction data. The output is the user's emotion score. Based on this emotion score, the review data is prioritized and displayed on the device.

[1473] Step 6:

[1474] The server uses a learning model to analyze the user's sentiment history and optimize future displays. The input is the user's past sentiment score and corresponding review data, and the output is future display suggestions based on the user's sentiment pattern. A generative AI model is used to make predictions and improve user satisfaction.

[1475] Step 7:

[1476] Users can view optimized and reliable reviews and make purchasing decisions. User reactions are fed back to the system and reflected in the next display. Prompts are used to solicit user feedback and improve the accuracy of the system.

[1477] Examples of specific prompts include:

[1478] "Write an honest review of this product. Detail the positives and negatives. Also, tell us in what situations you used it."

[1479] 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.

[1480] 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.

[1481] 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.

[1482] 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.

[1483] 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.

[1484] 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.

[1485] 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).

[1486] 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.

[1487] 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."

[1488] 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.

[1489] 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).

[1490] 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.

[1491] 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.

[1492] 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.

[1493] 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.

[1494] 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.

[1495] 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.

[1496] 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.

[1497] 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.

[1498] 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.

[1499] 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.

[1500] The following is further disclosed regarding the above embodiment.

[1501] (Claim 1)

[1502] A means of collecting word-of-mouth data;

[1503] A means for preprocessing the collected review data;

[1504] A means for determining malicious reviews based on the preprocessed review data;

[1505] How to filter out bad reviews,

[1506] A means to display filtered and reliable reviews to users, and

[1507] A system including:

[1508] (Claim 2)

[1509] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

[1510] (Claim 3)

[1511] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews that include patterns that are determined to be stealth marketing, and determining that the reviews are malicious reviews.

[1512] (Claim 4)

[1513] 2. The system of claim 1, wherein the means for determining malicious reviews includes means for detecting reviews with high negative sentiment scores using sentiment analysis and determining the reviews as malicious reviews.

[1514] (Claim 5)

[1515] The system of claim 1, further comprising means for learning past posting history and account trends and detecting and eliminating malicious reviews from specific accounts.

[1516] (Claim 6)

[1517] The system according to claim 1, further comprising means for performing a series of processes from collecting review data to filtering and displaying reliable reviews in real time.

[1518] "Example 1"

[1519] (Claim 1)

[1520] A means of collecting word-of-mouth data;

[1521] A means for preprocessing the collected review data;

[1522] A means for determining malicious reviews based on the preprocessed review data;

[1523] How to filter out bad reviews,

[1524] A means to display filtered and reliable reviews to users, and

[1525] A method to learn users' past posting history and account trends to detect patterns of malicious reviews,

[1526] A means for performing collection, pre-processing, judgment, and filtering processes in real time;

[1527] A system including:

[1528] (Claim 2)

[1529] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

[1530] (Claim 3)

[1531] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews that include patterns that are determined to be stealth marketing, and determining that the reviews are malicious reviews.

[1532] "Application Example 1"

[1533] (Claim 1)

[1534] A means of collecting word-of-mouth data;

[1535] A means for preprocessing the collected review data;

[1536] A means for determining malicious reviews based on the preprocessed review data;

[1537] How to filter out bad reviews,

[1538] A means to display filtered and reliable reviews to users, and

[1539] A means to learn users' past posting history and account trends and detect malicious reviews from specific accounts,

[1540] A means of processing review data in real time;

[1541] A system including:

[1542] (Claim 2)

[1543] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

[1544] (Claim 3)

[1545] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews that include patterns that are determined to be stealth marketing, and determining that the reviews are malicious reviews.

[1546] "Example 2: Combining Emotion Engines"

[1547] (Claim 1)

[1548] A means of collecting word-of-mouth data;

[1549] A means for preprocessing the collected review data;

[1550] A means for determining malicious reviews based on the preprocessed review data;

[1551] How to filter out bad reviews,

[1552] A means to display filtered and reliable reviews to users, and

[1553] means for analyzing user emotions;

[1554] A means for preferentially displaying reviews based on the analyzed user sentiment score;

[1555] A system including:

[1556] (Claim 2)

[1557] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

[1558] (Claim 3)

[1559] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews that include patterns that are determined to be stealth marketing, and determining that the reviews are malicious reviews.

[1560] "Application example 2 when combining emotion engines"

[1561] (Claim 1)

[1562] A means of collecting word-of-mouth data;

[1563] A means for preprocessing the collected review data;

[1564] A means for determining malicious reviews based on the preprocessed review data;

[1565] How to filter out bad reviews,

[1566] A means to customize and display filtered, reliable reviews based on user sentiment scores;

[1567] A system including:

[1568] (Claim 2)

[1569] 2. The system according to claim 1, wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

[1570] (Claim 3)

[1571] 2. The system of claim 1, wherein the means for determining malicious reviews includes means for detecting reviews that include patterns that are determined to be masked marketing techniques and determining that the reviews are malicious. [Explanation of symbols]

[1572] 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 of collecting word-of-mouth data; A means for preprocessing the collected review data; A means for determining malicious reviews based on the preprocessed review data; How to filter out bad reviews, A means to display filtered and reliable reviews to users, and A system including:

2. The system according to claim 1 , wherein the means for determining malicious reviews includes means for detecting reviews containing negative words and determining the reviews as malicious reviews.

3. 2. The system according to claim 1, wherein the means for determining malicious word-of-mouth includes means for detecting word-of-mouth containing a pattern determined to be stealth marketing, and determining that the word-of-mouth is malicious.

4. The system of claim 1 , wherein the means for determining malicious reviews includes means for detecting reviews with high negative sentiment scores using sentiment analysis and determining the reviews as malicious reviews.

5. The system of claim 1 , further comprising means for learning past posting history and account trends, and detecting and eliminating malicious reviews from specific accounts.

6. The system according to claim 1 , further comprising means for carrying out a series of processes from collecting word-of-mouth data to filtering and displaying reliable word-of-mouth reviews in real time.

Images