System

Abstract

A system is provided.SOLUTION: A system, comprising: means for collecting review comments; means for performing data pre-processing on the review comments; means for comparing the review comments with a characteristic word list of the review comments to determine product relevance; means for analyzing patterns of the review comments to determine unnatural reviews; means for comparing the review comments to determine comments with the same content as copy reviews; and means for feeding back relevance, unnaturalness, and copy determination results of the reviews.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

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

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

[0004] On online shopping platforms, product reviews are an important source of information for buyers. However, some reviews have problems such as fake reviews, unnatural reviews, and copied and pasted reviews, which reduces the reliability of reviews. These fraudulent reviews can distort the actual evaluation of a product and be misleading to consumers. Therefore, a system that provides accurate and reliable review information is needed. [Means for solving the problem]

[0005] The present invention includes a means for collecting review comments, performing data preprocessing, and then comparing each review comment with a feature word list to determine product relevance. It also includes a means for analyzing the patterns of the review comments to determine unnatural reviews, and a means for comparing review comments to determine whether comments with identical content are copy-and-paste reviews. Furthermore, by providing feedback on the review relevance, unnaturalness, and copy-and-paste determination results, a system is provided that provides accurate and reliable review information. The system of the present invention enables consumers to make purchasing decisions based on more accurate information, contributing to improving the quality of the entire online shopping platform.

[0006] A "review comment" is text information in which a user writes an evaluation or opinion about a product they have purchased.

[0007] "Data preprocessing" is the process of removing unnecessary information from collected text data and preparing it in an analyzable form.

[0008] A "characteristic word list" is a list of predefined words based on specific product attributes or characteristics, and is used to determine product relevance.

[0009] "Product relevance determination" is the process of determining whether a review comment is relevant to a particular product.

[0010] "Pattern analysis" is a method of analyzing the content and patterns of multiple review comments to detect similarities and anomalies.

[0011] "Unnatural reviews" are review comments that have similar content and are posted consecutively within a short period of time, which may be considered unnatural.

[0012] A "copy-and-paste review" is a review comment that is posted by copying and pasting the exact same text as an existing review comment.

[0013] "Feedback means" refers to the system part that includes a function to notify the reviewer and administrator of the results of the assessment of relevance, unnaturalness, and copy-and-paste. [Brief explanation of the drawings]

[0014] [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

[0015] 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.

[0016] First, the terms used in the following description will be explained.

[0017] 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).

[0018] 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.

[0019] 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.

[0020] 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.

[0021] 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."

[0022] [First embodiment]

[0023] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0024] 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.

[0025] 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).

[0026] 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.

[0027] 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.

[0028] 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.

[0029] 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.

[0030] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0031] 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.

[0032] 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.

[0033] 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.

[0034] 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."

[0035] The system of the present invention evaluates the reliability of review comments on shopping platforms and provides appropriate feedback to enable users to obtain more accurate review information. The program processing of this system is explained below in natural language.

[0036] Process Overview

[0037] 1. Collecting review comments

[0038] The server periodically retrieves all review comments from the shopping platform via API, which allows new review comments to be added to the system.

[0039] 2. Data Preprocessing

[0040] The server preprocesses the collected review comment data, which includes removing unnecessary spaces and special characters, splitting review comments, and standardizing the language.

[0041] 3. Product relevance determination

[0042] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is set.

[0043] The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is determined to be a "review with high product relevance."

[0044] Examples:

[0045] Product: Smartphone

[0046] Review comment: "The image quality of this phone is excellent."

[0047] Score: The word list includes "image quality," so the score is high.

[0048] 4. Identifying unnatural reviews

[0049] The server analyzes the patterns of review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time.

[0050] The server flags reviews if it detects any unnatural patterns.

[0051] Examples:

[0052] Review 1: "Great product!"

[0053] Review 2: "Great product!"

[0054] Review 3: "Great product!"

[0055] Verdict: Unnatural review and flagging.

[0056] 5. Judging copy-paste reviews

[0057] Every time a new review comment is posted, the server compares it with all existing review comments to see if there are any that are exactly the same.

[0058] If the server finds an exact match, it flags the review as a "copy-and-paste review."

[0059] Examples:

[0060] Review 1: "This product is amazing!"

[0061] Review 2: "This product is amazing!"

[0062] Verdict: Copy-paste review and flagging.

[0063] 6. Feedback

[0064] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being related to the product, unnatural, or copied and pasted.

[0065] The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the result of the judgment.

[0066] This system allows users to select products based on reliable review information, improving the evaluation quality of the entire online shopping platform.

[0067] The processing flow will be explained below.

[0068] Step 1:

[0069] Collecting review comments

[0070] The server uses the API from the shopping platform to periodically retrieve all review comments. This collection includes review comments for all products.

[0071] Step 2:

[0072] Data Preprocessing

[0073] The server stores the collected review comments as text data. Next, it performs a cleaning process to remove unnecessary spaces, special characters, and HTML tags from the review comments. It then tokenizes the review comments into words and sentences to prepare them for analysis.

[0074] Step 3:

[0075] Product relevance determination

[0076] The server pre-sets a word list based on the characteristics and attributes of each product. When a review comment is posted, the server compares the words in the comment with the word list. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is used. The server scores the frequency with which words from the word list appear in the comment, and if the score is above a certain level, it determines that the review is "highly relevant to the product."

[0077] Step 4:

[0078] Identifying unnatural reviews

[0079] The server analyzes all review comments posted within a certain period of time to check for patterns in content, determining whether reviews with similar content or writing style have been posted in succession within a short period of time. If the server detects an unnatural pattern, it flags the review comment.

[0080] Step 5:

[0081] Judgment on copy-paste reviews

[0082] Each time a new review comment is submitted, the server compares it with all existing review comments. Using sophisticated text comparison algorithms, the server determines whether a review comment with the exact same content already exists. If the server finds a match, it flags the review comment as a "copy-and-paste review."

[0083] Step 6:

[0084] Feedback and Filtering

[0085] The server will allow a review comment to be posted only if it is not flagged as irrelevant, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the posting. The server will also notify the administrator of the fraudulent review and help the administrator review the review.

[0086] Based on these concrete steps, we will be able to provide users with reliable review information and improve the quality of the entire shopping platform.

[0087] Example 1

[0088] 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."

[0089] On online shopping platforms, the reliability of review comments is important for users to use when selecting products. However, fake reviews, copy-paste reviews, and reviews with unnatural patterns are rampant, making it difficult for users to obtain accurate information. Therefore, there is a need for a system that can evaluate the reliability of review comments and provide users with accurate review information.

[0090] 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.

[0091] In this invention, the server includes a means for collecting review comments, a means for performing data preprocessing of the review comments, a means for comparing the review comments with a feature word list to determine product relevance, a means for analyzing patterns of the review comments to determine unnatural reviews, a means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, and a means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, thereby enabling users to select products based on highly reliable review information.

[0092] "Review comments" are opinions and evaluations posted by users about products on online shopping platforms.

[0093] "Data preprocessing" refers to the process of removing unnecessary spaces and special characters from the text data of collected review comments, and performing tokenization and language standardization.

[0094] The "characteristic word list" is a list of keywords based on product attributes that are set to evaluate the reliability of review comments about products.

[0095] "Product relevance" is an index that indicates the degree to which a review comment mentions a specific product and the degree of match with the feature word list.

[0096] "Pattern analysis" is the process of analyzing the posting patterns of review comments and determining whether the same or similar content has been posted consecutively within a certain period of time.

[0097] An "unnatural review" is a review that is deemed to deviate from normal user behavior, such as when similar comments are posted consecutively within a short period of time.

[0098] A "copy-and-paste review" is a comment that is posted by simply copying an existing review comment, and refers to a review whose content is exactly the same.

[0099] "Feedback" is a process that notifies users and administrators of the relevance, unnaturalness, and copy-and-paste judgment results of review comments, and allows only appropriate reviews.

[0100] The present invention provides a system for evaluating the reliability of review comments on an online shopping platform, thereby enabling users to obtain more accurate review information. Specific embodiments are described in detail below.

[0101] Collecting review comments

[0102] The server periodically retrieves all review comments using the shopping platform's API. This periodically saves the latest review comments, including the most recent ones, to the database. The specific software used is the Python requests library.

[0103] Data Preprocessing

[0104] The server preprocesses the collected review comment data, including removing unnecessary whitespace and special characters, tokenizing the text data, and unifying the language. This process uses the Python nltk library and re module.

[0105] Product relevance determination

[0106] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a feature word list such as "image quality," "battery," and "operability" is set. The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is deemed to be "highly relevant to the product." The specific software used is the Python scikit-learn library.

[0107] Specific examples

[0108] Product: Smartphone

[0109] Review comment: "The image quality of this phone is excellent."

[0110] Since the word list includes "image quality," it is judged to have a high score.

[0111] Identifying unnatural reviews

[0112] The server analyzes patterns in review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time. If the server detects an unnatural pattern, it flags those reviews. This process uses the Python Pandas library and time series analysis.

[0113] Specific examples

[0114] Review 1: "Great product!"

[0115] Review 2: "Great product!"

[0116] Review 3: "Great product!"

[0117] The server will flag these as unnatural reviews.

[0118] Judgment on copy-paste reviews

[0119] The server compares each new review comment submitted with all existing review comments to see if there are any identical comments. If the server finds a match, it flags the review as a "copy-and-paste review." The software uses the Python difflib library.

[0120] Specific examples

[0121] Review 1: "This product is amazing!"

[0122] Review 2: "This product is amazing!"

[0123] The server determines that Review 2 is a copy-and-paste review and flags it.

[0124] feedback

[0125] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. Inappropriate reviews are filtered out and an error message is displayed to the user. The server also notifies the administrator of the review content and the result. This process uses the Python Flask or Django framework and a mail server (SMTP) for notifications.

[0126] Prompt Sentence Examples

[0127] Please rate the user-submitted review comment "This phone has excellent battery life" and provide feedback based on the following criteria:

[0128] conditions:

[0129] 1. To determine whether a review is highly relevant to the product, it is scored based on a list of characteristic words such as "battery," "duty," and "excellent."

[0130] 2. Check to see if similar reviews have been posted consecutively within a certain period of time.

[0131] 3. Compare with existing review comments to check whether the review is a copy-and-paste one.

[0132] This system will enable users to select products based on reliable review information, and is expected to improve the evaluation quality of the entire online shopping platform.

[0133] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0134] Step 1:

[0135] Collects review comments. The server periodically retrieves all review comments through the shopping platform's API. The input is the API request, and the output is the retrieved review comment data in JSON format. The server sends the API request and stores the received data in a database.

[0136] Step 2:

[0137] Perform data preprocessing. The server performs preprocessing on the collected review comment data. The input is JSON-formatted review comment data, and the output is preprocessed text data. The server removes unnecessary whitespace and special characters, tokenizes the text data, and unifies the language.

[0138] Step 3:

[0139] The server determines product relevance by comparing the preprocessed review comments with a feature word list based on product attributes. The input is the preprocessed text data and feature word list, and the output is scored review comment data. The server scores how often words included in the word list appear in the review comments, and if the score is above a certain level, it determines the review to be "highly relevant to the product."

[0140] Step 4:

[0141] The server determines whether a review is unnatural. It analyzes patterns in review comments posted within a certain period of time. The input is review comment data from a specific period of time, and the output is the determination result and flagged data. The server checks whether reviews with similar content have been posted consecutively within a short period of time, and if an unnatural pattern is detected, it flags those reviews.

[0142] Step 5:

[0143] The server determines whether a review is a copy-and-paste review. Whenever a new review comment is posted, it compares it with all existing review comments. The input is the new review comment and existing review comment data, and the output is the determination result and flagged data. If the server finds a matching comment, it flags the review as a "copy-and-paste review."

[0144] Step 6:

[0145] Feedback is provided. The server allows a review to be posted only if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. The input is the review comment and various judgment results, and the output is a notification to the user and administrator. The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the judgment results.

[0146] (Application example 1)

[0147] 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."

[0148] On conventional shopping platforms, many unreliable review comments are posted, causing users to make incorrect purchasing decisions. Furthermore, there is also the problem of similar reviews and copy-paste reviews being posted in a short period of time, which reduces the reliability of the reviews. Therefore, there is a need for the development of a system that provides users with reliable review information and helps them make more accurate purchasing decisions.

[0149] 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.

[0150] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a list of characteristic words to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, means for evaluating the reliability of the review comments in real time and providing feedback, and means for displaying reviews in order of reliability. This allows users to receive reliable reviews in real time and make more accurate purchasing decisions based on them.

[0151] A "review comment" is a written comment written by a user expressing their evaluation or opinion of a product or service.

[0152] "Means of collection" refers to the methods by which data is obtained from online platforms and entered into the system.

[0153] "Data preprocessing" is the process of removing unnecessary information and formatting acquired data to make it easier to analyze.

[0154] A "characteristic word list" is a list of important keywords related to a specific product or service.

[0155] "Product relevance" is the degree to which a review comment relates to a particular product or service.

[0156] "Means for analyzing patterns" are methods for detecting certain templates or similarities in review comments.

[0157] "Unnatural reviews" are comments that deviate from normal user behavior, such as those posted in large numbers in a short period of time or those that contain repeated, repeated comments.

[0158] A "copy-and-paste review" is a review comment in which the exact same content as another comment is copied and pasted.

[0159] "Feedback methods" are methods for notifying users and administrators of the evaluation results and encouraging them to take appropriate action.

[0160] "Means for assessing trustworthiness in real time" refers to a method for automatically determining the accuracy and trustworthiness of a review the moment it is posted.

[0161] "Displaying in order of reliability" is a method of displaying highly rated review comments at the top, making them easier for users to refer to.

[0162] The system of the present invention evaluates the reliability of review comments on a shopping platform, allowing users to obtain more accurate review information. The system is implemented through the following process.

[0163] Hardware used

[0164] Smartphone: Used as a device for users to post review comments.

[0165] Server: A central computer that collects review comments, performs data preprocessing, analysis, evaluation, and feedback.

[0166] API server: Used as an interface to obtain review comments from the shopping platform.

[0167] Software used

[0168] Python: Used for server-side data processing and analysis.

[0169] Flask: A Python web framework used to build API servers.

[0170] MySQL: Used as a database management system to store collected review comments.

[0171] Natural language processing library: Used to preprocess and parse review comments.

[0172] Data processing and calculation flow

[0173] 1. Collecting review comments

[0174] The server periodically retrieves review comments from the shopping platform via API. This API server is built using Flask, and the retrieved data is stored in a MySQL database.

[0175] 2. Data Preprocessing

[0176] The server preprocesses the retrieved review comment data using a Python natural language processing library, which includes removing unnecessary whitespace and special characters, splitting review comments, and standardizing the language.

[0177] 3. Product relevance determination

[0178] The server compares the review comment with a list of characteristic words based on product attributes. For example, a review about a smartphone would use keywords such as "image quality," "battery," and "operability." The server then scores the review comment, and if it scores above a certain level, it determines that the review comment is "highly relevant to the product."

[0179] 4. Identifying unnatural reviews

[0180] The server analyzes review comments posted within a certain period of time to check for similar reviews posted in succession within a short period of time, and if it detects any unnatural patterns, it flags those reviews.

[0181] 5. Judging copy-paste reviews

[0182] The server compares the newly submitted review comment with all existing review comments, checking for exact matches, and if an exact match is found, flagging it as a "copy-and-paste review."

[0183] 6. Feedback

[0184] The server allows reviews to be posted only if they are not flagged as being related to the product, unnatural, or copied and pasted, and provides real-time feedback to users. It also has a function to display reviews in order of reliability.

[0185] Examples of specific examples and prompts

[0186] As a concrete example, consider the following review comment:

[0187] Product: Smartphone

[0188] Reviewer: "This phone's battery lasts a long time."

[0189] An example of a prompt is:

[0190] "Please tell us if this review is relevant to the product: 'This phone's battery lasts a long time.'"

[0191] "Tell me if this review fits an unnatural pattern: 'This phone's battery lasts a long time.'"

[0192] "Please tell us if this review exactly matches an existing review: 'This phone's battery lasts a long time.'"

[0193] This allows users to choose products based on reliable reviews, improving the quality of ratings across online shopping platforms.

[0194] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0195] Step 1:

[0196] Collect review comments:

[0197] The server periodically collects review comments using the API provided by the shopping platform. The server receives the review comments obtained via the API in JSON format and stores them in a MySQL database. The input is the JSON data obtained from the API, and the output is the review comments stored in the database.

[0198] Step 2:

[0199] Data preprocessing:

[0200] The server preprocesses the collected review comment data. It uses Python's regular expression library to remove unnecessary whitespace and special characters, split the review comments into sentences, and converts data in different formats into a unified format. The input is the raw review comments, and the output is the preprocessed, clean data.

[0201] Step 3:

[0202] Product relevance determination:

[0203] The server compares the preprocessed review comments with a pre-defined feature word list. The server calculates the frequency of occurrence of words included in the feature word list, and determines that a product is highly relevant if the frequency is above a certain score. The input is the preprocessed review comments and the feature word list, and the output is data flagged as highly relevant reviews.

[0204] Step 4:

[0205] Unnatural review verdict:

[0206] The server compares the preprocessed review comments with other review comments posted within a certain period of time and analyzes patterns. If a large number of similar review comments are posted in a short period of time, the server determines that they are unnatural reviews and flags them. The input is the preprocessed review comments and past review comments, and the output is data flagged as unnatural reviews.

[0207] Step 5:

[0208] Copypaste review verdict:

[0209] The server compares the newly posted review comment with all existing review comments based on their content similarity. If the server finds a review comment with exactly the same content, it flags it as a copy-and-paste review. The input is the new review comment and the existing review comments, and the output is the data determined to be a copy-and-paste review.

[0210] Step 6:

[0211] feedback:

[0212] The server provides real-time feedback to reviewers based on the review's relevance, unnaturalness, and copy-and-paste judgment results. Reviews are displayed on the user's device in order of reliability. Reviews with low reliability are rejected, and the user is notified of the reason. The input is the review judgment results, and the output is a feedback message.

[0213] This will enable users to make more accurate purchasing decisions based on reliable review comments.

[0214] 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.

[0215] The system of the present invention evaluates the reliability of review comments on a shopping platform, and further optimizes review information by recognizing user sentiment. The program processing of this system is explained below in natural language.

[0216] Process Overview

[0217] 1. Collecting review comments

[0218] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[0219] 2. Data Preprocessing

[0220] The server cleans the collected review comments, removing unnecessary whitespace, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[0221] 3. Product relevance determination

[0222] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a smartphone review, a word list such as "image quality," "battery," and "operability" would be prepared.

[0223] The server scores the number of times the words in the word list appear in the review comments, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[0224] Examples:

[0225] Product: Smartphone

[0226] Review comment: "The image quality of this smartphone is outstanding."

[0227] Score: The word list includes "image quality," so the score is high.

[0228] 4. Identifying unnatural reviews

[0229] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, checking to see if reviews with similar content or writing style have been posted in succession within a short period of time.

[0230] If the server detects an unnatural pattern, it flags the review comment.

[0231] Examples:

[0232] Review 1: "Great product!"

[0233] Review 2: "Great product!"

[0234] Review 3: "Great product!"

[0235] Verdict: Unnatural review and flagging.

[0236] 5. Judging copy-paste reviews

[0237] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[0238] If the server finds a match, it flags the review comment as a "copy and paste review."

[0239] Examples:

[0240] Review 1: "This product is amazing!"

[0241] Review 2: "This product is amazing!"

[0242] Verdict: Copy-paste review and flagging.

[0243] 6. Emotion Recognition by Emotion Engine

[0244] The server analyzes the review comments through a sentiment engine and categorizes them into positive, negative, or neutral sentiment.

[0245] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[0246] Examples:

[0247] Review comment: "The battery on this phone dies so quickly it's useless."

[0248] Sentiment analysis result: Negative

[0249] Score: Reflected as a low rating.

[0250] 7. Feedback and Filtering

[0251] The server will only allow a review comment to be posted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the post.

[0252] The server notifies the user and the administrator of the feedback along with the emotion score from the emotion engine, while the administrator is notified to further review and take action.

[0253] As a result, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind. The introduction of the emotion engine further improves the quality and reliability of reviews, and product evaluation information is more accurately reflected.

[0254] The processing flow will be explained below.

[0255] Step 1:

[0256] Collecting review comments

[0257] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[0258] Step 2:

[0259] Data Preprocessing

[0260] The server cleans the collected review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences to make them analyzable.

[0261] Step 3:

[0262] Product relevance determination

[0263] The server compares review comments using a word list based on the features and attributes of each product. It compares the words in the review comments with the word list and generates a score. For example, a word list related to smartphones might include "image quality," "battery," and "operability."

[0264] The server scores the frequency with which words included in the word list appear, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[0265] Step 4:

[0266] Identifying unnatural reviews

[0267] The server analyzes all review comments posted within a certain period of time to identify patterns in content, determining whether similar reviews or writing styles have been posted in succession within a short period of time, particularly comparing sentence length, vocabulary, and sentence structure.

[0268] If the server detects an unnatural pattern, it flags the review comment.

[0269] Step 5:

[0270] Judgment on copy-paste reviews

[0271] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[0272] If the server finds a match, it flags the review comment as a "copy and paste review."

[0273] Step 6:

[0274] Emotion recognition by emotion engine

[0275] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral sentiments. Specifically, it uses a natural language processing algorithm to extract and score emotional expressions contained in the sentences.

[0276] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[0277] Step 7:

[0278] Feedback and Filtering

[0279] The server will only allow a review comment to be posted if it is not flagged for product relevance, unnaturalness, or copy-and-paste. If a flag is raised, the server will display an error message to the user explaining why the post was rejected.

[0280] The server filters out inappropriate reviews and notifies users and administrators, including the sentiment score generated by the sentiment engine. Administrators are sent detailed information, such as the review content, the reason for flagging, and the sentiment score, and are offered help in reviewing the review if necessary.

[0281] Based on these specific steps, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind.The integration of the sentiment engine further improves the quality and reliability of reviews, and more accurately reflects product evaluation information.

[0282] Example 2

[0283] 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."

[0284] Conventional shopping platforms lack an appropriate mechanism for determining the authenticity of review comments, which creates the risk of users making purchasing decisions based on inaccurate information. Therefore, there is a need to improve the reliability of review comments and provide a safe and secure environment for users. Furthermore, it is necessary to accurately capture the sentiment of review comments in order to more accurately reflect product evaluation information.

[0285] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for collecting review comments, means for performing data preprocessing of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing the patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for analyzing the sentiment of the review comments, means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results, and means for rejecting the posting of a review comment if it is unnatural or a copy-and-paste. This improves the reliability of the review comments, provides an environment where users can refer to reviews with confidence, and makes it possible to more accurately reflect product evaluation information.

[0286] A "review comment" is a text of an opinion or evaluation written by a user on a shopping platform regarding a product or service.

[0287] "Data preprocessing" is the process of removing unnecessary elements from collected review comments and converting them into a format suitable for analysis.

[0288] A "characteristic word list" is a list of keywords related to a particular product or service that is used to compare with review comments.

[0289] "Product relevance determination" is the process of determining whether a review comment is relevant to a particular product or service.

[0290] "Pattern analysis" is the process of analyzing the content and format of review comments to detect specific patterns or anomalies.

[0291] An "unnatural review" is a review comment that shows an unusual pattern, such as being posted repeatedly in a short period of time or in a different style.

[0292] A "copy-and-paste review" refers to a review comment that has exactly the same content as an existing review comment.

[0293] "Sentiment analysis" is the process of analyzing the content of review comments and classifying them into sentiments such as positive, negative, or neutral.

[0294] "Feedback" refers to the provision of information to notify the user or administrator of the results of analysis or judgment, and to encourage them to take necessary action.

[0295] "Posting refusal" refers to the action of not allowing a user to post in order to prevent unnatural reviews or copy-and-paste reviews from being posted.

[0296] "Server" refers to a computer system that manages and executes a series of processes, including collection of review comments, pre-processing, analysis, judgment, feedback, and rejection of submissions.

[0297] The system of the present invention evaluates the reliability of review comments on a shopping platform, recognizes user sentiment, and optimizes review information. Specific operations and processes for this purpose are described below.

[0298] 1. Collecting review comments

[0299] The system retrieves review comments through the shopping platform's API. The server periodically sends requests to the API endpoint and stores the retrieved review comments in a database. Specifically, MySQL or PostgreSQL can be used as the database.

[0300] For example, the server accesses the API endpoint "https: / / example.com / api / reviews" and stores the retrieved review comments in the "reviews" table.

[0301] 2. Data Preprocessing

[0302] The server cleans the retrieved review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences. This preprocessing can be performed using NLPK, a Python NLP library.

[0303] For example, "This product wonderful !" is text cleaned to "This product is great!" and then tokenized.

[0304] 3. Product relevance determination

[0305] The server generates a word list based on the characteristics and attributes of each product and compares it with review comments. For example, a word list containing keywords such as "image quality," "battery," and "operability" is used for smartphone reviews.

[0306] Specifically, the frequency of keywords in review comments is scored, and if the score is above a certain level, it is determined to be a "review with high product relevance." For example, the server analyzes a comment such as "The image quality of this smartphone is outstanding," and assigns a high score because "image quality" is included in the word list.

[0307] 4. Identifying unnatural reviews

[0308] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, specifically, whether review comments with similar content or format have been posted consecutively within a short period of time.

[0309] If an unnatural pattern is detected, the review comments are flagged. For example, if the server sees multiple reviews with the same content saying "Great product!" posted within a short period of time, it will flag them as unnatural.

[0310] 5. Judging copy-paste reviews

[0311] Each time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms (e.g., Python's difflib library) to determine whether they match.

[0312] If a match is found, the server flags the review comment as a "copy-and-paste review." For example, if a comment already exists that says "This product is great!", the server adds a flag.

[0313] 6. Emotion Recognition by Emotion Engine

[0314] The server feeds the review comments into a sentiment analysis engine and classifies them into positive, negative, or neutral sentiments. Specifically, sentiment analysis libraries such as TextBlob and VADER can be used.

[0315] This allows us to calculate an emotion score and reflect that score in the overall rating of the product. For example, we can use TextBlob to analyze a comment like "This phone's battery runs out quickly, so it's useless" and assign it a negative score.

[0316] 7. Feedback and Filtering

[0317] When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user explaining the reason for the post.

[0318] The emotion engine also notifies users and administrators of the emotion score, aiming to improve the reliability of reviews. For example, the server may notify users that "this review has been flagged as unnatural" and encourage them to resubmit. It may also notify administrators that "a new unnatural review has been detected. Please check it."

[0319] Example prompts for generative AI models

[0320] "Category this review comment as it relates to a specific product and its sentiment. Example: 'The image quality of this phone is outstanding.'"

[0321] "Please judge whether the review comments for this listing are unnatural. Example: 'Great product!', 'Great product!'"

[0322] "Check if this review comment matches an existing comment or is a copypasta review. Example: 'This product is amazing!'"

[0323] The above is an embodiment of the system of the present invention.

[0324] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0325] Step 1:

[0326] Collect review comments:

[0327] Input: Shopping platform API endpoint URL

[0328] Specific operation: The server periodically sends a request to an API endpoint to retrieve review comments. For example, it accesses the API endpoint "https: / / example.com / api / reviews".

[0329] Data processing: Metadata (user ID, product ID, posting date and time, etc.) is added to the acquired review comments.

[0330] Output: A dataset containing review comments and metadata

[0331] Specific operation: The server stores the acquired dataset in a database (e.g., MySQL or PostgreSQL).

[0332] Step 2:

[0333] Data preprocessing:

[0334] Input: Review comments stored in the database

[0335] What happens: The server cleans the review comments, specifically using a Python NLP library (e.g., NLPK) to remove unnecessary whitespace, special characters, and HTML tags.

[0336] Data processing: Review comments are tokenized into words and sentences.

[0337] Output: Clean, tokenized review comments

[0338] Specific actions: For example, "This product wonderful !" to "This product is great!" and then split into individual words.

[0339] Step 3:

[0340] Product relevance determination:

[0341] Input: Clean and tokenized review comments, product feature wordlist

[0342] Specific operation: The server compares the review comments with the feature word list, which includes keywords based on product attributes such as "image quality," "battery," and "operability."

[0343] Data calculation: Score the frequency of occurrence of keywords in review comments.

[0344] Output: Product relevance score

[0345] Specific operation: For example, a comment such as "The image quality of this smartphone is outstanding" will be given a high score because the word list includes "image quality."

[0346] Step 4:

[0347] Unnatural review verdict:

[0348] Input: All review comments posted within a certain period

[0349] Specific operation: The server analyzes the content patterns of review comments, checking whether reviews with the same content or writing style have been posted consecutively.

[0350] Data arithmetic: Analysis to detect unnatural patterns

[0351] Output: Review comments flagged as unnatural

[0352] What happens: For example, if multiple reviews with the same content, such as "Great product!", are posted within a short period of time, we flag them as unusual.

[0353] Step 5:

[0354] Copypaste review verdict:

[0355] Input: New review comment, Existing review comment

[0356] What happens: The server uses advanced text comparison algorithms (e.g., Python's difflib library) to compare the new comment with existing review comments.

[0357] Data calculation: The process of checking the content of comments for consistency

[0358] Output: Review comments flagged as copypasta reviews

[0359] Specific behavior: For example, if a comment saying "This product is amazing!" already exists, add a flag.

[0360] Step 6:

[0361] Emotion Recognition with Emotion Engine:

[0362] Input: Review comment

[0363] Specific operation: The server inputs the review comments into a sentiment analysis engine (e.g., TextBlob or VADER) and classifies the sentiment.

[0364] Data arithmetic: Calculating sentiment scores

[0365] Output: Review comments with sentiment scores

[0366] Specific behavior: For example, a comment such as "This phone's battery runs out quickly, making it useless" is scored as negative.

[0367] Step 7:

[0368] Feedback and Filtering:

[0369] Input: Review comments (flagged for relevance, unnaturalness, and copy-paste), sentiment score

[0370] Specific behavior: When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user.

[0371] Data Calculation: Feedback and Submission Judgment

[0372] Output: Notification to users and administrators

[0373] Specific behavior: For example, the server notifies the user that "this review has been flagged as unnatural" and encourages them to resubmit, while simultaneously notifying the administrator that "a new unnatural review has been detected."

[0374] The above are the specific processing steps of the program of this system.

[0375] (Application example 2)

[0376] 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."

[0377] Conventional review comment evaluation systems often lack the functionality to fully guarantee the reliability of reviews. For example, reviews that are unrelated to the product, unnatural reviews, or copy-and-paste reviews are often mixed in, making it difficult for users to obtain accurate information. Furthermore, there is a demand for more accurate product evaluations by analyzing the sentiment of review comments. However, current systems lack sentiment analysis functionality, making it impossible to improve the quality and reliability of reviews.

[0378] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0379] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for performing sentiment analysis of the review comments, and means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results. This allows users to obtain highly reliable review information and evaluate products more accurately.

[0380] "Review comments" refer to impressions and evaluations written by users about products and services.

[0381] "Means of collection" refers to the process or function of obtaining specific information from the internet or databases.

[0382] "Data preprocessing" refers to the process of analyzing collected data or performing pre-analysis processing to remove unnecessary parts and format the data.

[0383] A "characteristic word list" is a list of important keywords related to a specific product or category.

[0384] "Method for determining product relevance" refers to the method or algorithm used to evaluate how relevant a review comment is to a particular product.

[0385] "Methods for analyzing patterns to identify unnatural reviews" refers to methods for analyzing the content and posting patterns of review comments to detect unnatural content and typical actions.

[0386] A "copy-and-paste review" refers to a review comment with the same content, copied and pasted exactly from another review comment.

[0387] "Sentiment analysis" refers to the technology of analyzing user emotions and intentions from text data and classifying them as positive, negative, or neutral.

[0388] "Feedback" refers to the process of returning analysis and evaluation results to the user or system.

[0389] MODE FOR CARRYING OUT THE INVENTION

[0390] This invention is a system for improving the reliability of review comments on shopping platforms and providing optimal information to users. This system collects review comments, performs data preprocessing, and performs product relevance, unnaturalness, copy-and-paste, and sentiment analysis before providing feedback.

[0391] System Overview

[0392] The server includes the following means:

[0393] 1. How to collect review comments:

[0394] The server periodically retrieves all review comments from the shopping platform via API, and the collected review comments are stored in a database.

[0395] 2. Methods for data preprocessing of review comments:

[0396] The server cleans the collected review comments by removing unnecessary spaces, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[0397] 3. A method for determining product relevance by comparing review comments with a feature word list:

[0398] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is prepared. The server scores the number of times words included in the word list appear in the review comments, and if the score is above a certain level, it is determined to be a "review highly relevant to the product."

[0399] 4. Methods for analyzing review comment patterns and identifying unnatural reviews:

[0400] The server analyzes all review comments posted within a certain period of time to check for patterns in content, checking to see if similar reviews or writing styles have been posted in succession within a short period of time, and if an unnatural pattern is detected, flagging the review comment.

[0401] 5. How to compare review comments and identify comments with identical content as copy-paste reviews:

[0402] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to check whether an identical review comment already exists, and if a match is found, flags the review comment as a "copy-and-paste review."

[0403] 6. Methods for sentiment analysis of review comments:

[0404] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral emotions. Based on the scoring by the emotion engine, the server calculates an emotion score for the review comments and reflects it in the overall rating of the product.

[0405] 7. Feedback on review relevance, unnaturalness, copy-paste determination, and sentiment analysis results:

[0406] When a review comment is attempted to be posted, the server allows it to be posted only if it is not flagged as being related to the product, unnatural, or copied and pasted. If a flag is raised, an error message is displayed to the user explaining the reason for the post. Feedback is also provided to the user and administrator along with the emotion score calculated by the emotion engine, and administrators are notified of the review and prompted to take further action.

[0407] Hardware and software used

[0408] Hardware: Servers, cloud storage devices

[0409] Software: Python, requests, BeautifulSoup, re, nltk, textblob, scikit-learn, emotion engine (TextBlob, etc.)

[0410] Specific examples

[0411] As a concrete example, we will show the procedure for processing a review comment such as "The image quality of this smartphone is outstanding." This comment is determined to be highly relevant to the product because it contains the keyword "image quality," and is approved because it does not match unnatural reviews or copy-paste reviews.

[0412] Prompt Sentence Examples

[0413] Example prompts to input to a generative AI model:

[0414] Let's analyze the new review comment, "The image quality of this phone is outstanding." We'll rate it based on the following criteria:

[0415] 1. Product Relevance

[0416] 2. Unnatural patterns

[0417] 3. Copy and paste

[0418] 4. Sentiment analysis

[0419] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0420] Step 1:

[0421] Collecting review comments

[0422] The server periodically calls the shopping platform's API to retrieve review comments for each product. As input, it uses the API endpoint URL and authentication information. As output, it generates JSON data containing the retrieved review comments. This data is stored in the server's database.

[0423] Step 2:

[0424] Data Preprocessing

[0425] The server cleans the collected review comments. As input, it uses the text data of the review comments stored in the database. Specifically, it removes unnecessary whitespace, special characters, and HTML tags, and tokenizes the review comments into words and sentences. As output, clean text data is generated and passed to the next step.

[0426] Step 3:

[0427] Product relevance determination

[0428] The server evaluates the relevance of review comments using a word list based on the features and attributes of each product. As input, it uses the clean review comments and the feature word list. Specifically, it scores how often words included in the feature word list appear in the review comments. As output, it generates a relevance score for each review comment.

[0429] Step 4:

[0430] Identifying unnatural reviews

[0431] The server analyzes all review comments posted within a certain period of time and checks for patterns. As input, it uses text data from review comments over a certain period of time and the current review comment. Specifically, it applies a text pattern matching algorithm to evaluate the similarity. As output, it generates a flag indicating whether the review is unnatural.

[0432] Step 5:

[0433] Judgment on copy-paste reviews

[0434] Each time a new review comment is posted, the server compares it with all existing review comments. As input, it uses data from the new review comment and the existing review comments. Specifically, it uses advanced text comparison algorithms (e.g., cosine similarity) to check whether there are any identical review comments. As output, it generates a flag indicating whether the review is a copy-and-paste review.

[0435] Step 6:

[0436] sentiment analysis

[0437] The server parses the review comments into a sentiment engine. It uses the clean review comments as input. It sends the text to the sentiment engine (e.g., TextBlob) and gets a sentiment score: positive, negative, or neutral. It generates a sentiment score for each review comment as output.

[0438] Step 7:

[0439] Feedback and Filtering

[0440] When an attempt is made to post a review comment, the server provides feedback based on the judgment results obtained in the previous steps. All judgment results (relevance, unnaturalness, copy-paste judgment results, and sentiment score) are used as input. Specific behavior is to allow posting only if no flags are raised, and to display an error message to the user if a flag is raised. More detailed feedback is notified to the administrator. As output, an error message or approval message is displayed to the user, and the review comment is saved in the database.

[0441] 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.

[0442] 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.

[0443] 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.

[0444] [Second embodiment]

[0445] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0446] 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.

[0447] 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).

[0448] 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.

[0449] 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.

[0450] 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).

[0451] 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.

[0452] 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.

[0453] 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.

[0454] 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.

[0455] 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.

[0456] 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."

[0457] The system of the present invention evaluates the reliability of review comments on shopping platforms and provides appropriate feedback to enable users to obtain more accurate review information. The program processing of this system is explained below in natural language.

[0458] Process Overview

[0459] 1. Collecting review comments

[0460] The server periodically retrieves all review comments from the shopping platform via API, which allows new review comments to be added to the system.

[0461] 2. Data Preprocessing

[0462] The server preprocesses the collected review comment data, which includes removing unnecessary spaces and special characters, splitting review comments, and standardizing the language.

[0463] 3. Product relevance determination

[0464] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is set.

[0465] The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is determined to be a "review with high product relevance."

[0466] Examples:

[0467] Product: Smartphone

[0468] Review comment: "The image quality of this phone is excellent."

[0469] Score: The word list includes "image quality," so the score is high.

[0470] 4. Identifying unnatural reviews

[0471] The server analyzes the patterns of review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time.

[0472] The server flags reviews if it detects any unnatural patterns.

[0473] Examples:

[0474] Review 1: "Great product!"

[0475] Review 2: "Great product!"

[0476] Review 3: "Great product!"

[0477] Verdict: Unnatural review and flagging.

[0478] 5. Judging copy-paste reviews

[0479] Every time a new review comment is posted, the server compares it with all existing review comments to see if there are any that are exactly the same.

[0480] If the server finds an exact match, it flags the review as a "copy-and-paste review."

[0481] Examples:

[0482] Review 1: "This product is amazing!"

[0483] Review 2: "This product is amazing!"

[0484] Verdict: Copy-paste review and flagging.

[0485] 6. Feedback

[0486] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being related to the product, unnatural, or copied and pasted.

[0487] The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the result of the judgment.

[0488] This system allows users to select products based on reliable review information, improving the evaluation quality of the entire online shopping platform.

[0489] The processing flow will be explained below.

[0490] Step 1:

[0491] Collecting review comments

[0492] The server uses the API from the shopping platform to periodically retrieve all review comments. This collection includes review comments for all products.

[0493] Step 2:

[0494] Data Preprocessing

[0495] The server stores the collected review comments as text data. Next, it performs a cleaning process to remove unnecessary spaces, special characters, and HTML tags from the review comments. It then tokenizes the review comments into words and sentences to prepare them for analysis.

[0496] Step 3:

[0497] Product relevance determination

[0498] The server pre-sets a word list based on the characteristics and attributes of each product. When a review comment is posted, the server compares the words in the comment with the word list. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is used. The server scores the frequency with which words from the word list appear in the comment, and if the score is above a certain level, it determines that the review is "highly relevant to the product."

[0499] Step 4:

[0500] Identifying unnatural reviews

[0501] The server analyzes all review comments posted within a certain period of time to check for patterns in content, determining whether reviews with similar content or writing style have been posted in succession within a short period of time. If the server detects an unnatural pattern, it flags the review comment.

[0502] Step 5:

[0503] Judgment on copy-paste reviews

[0504] Each time a new review comment is submitted, the server compares it with all existing review comments. Using sophisticated text comparison algorithms, the server determines whether a review comment with the exact same content already exists. If the server finds a match, it flags the review comment as a "copy-and-paste review."

[0505] Step 6:

[0506] Feedback and Filtering

[0507] The server will allow a review comment to be posted only if it is not flagged as irrelevant, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the posting. The server will also notify the administrator of the fraudulent review and help the administrator review the review.

[0508] Based on these concrete steps, we will be able to provide users with reliable review information and improve the quality of the entire shopping platform.

[0509] Example 1

[0510] 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."

[0511] On online shopping platforms, the reliability of review comments is important for users to use when selecting products. However, fake reviews, copy-paste reviews, and reviews with unnatural patterns are rampant, making it difficult for users to obtain accurate information. Therefore, there is a need for a system that can evaluate the reliability of review comments and provide users with accurate review information.

[0512] 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.

[0513] In this invention, the server includes a means for collecting review comments, a means for performing data preprocessing of the review comments, a means for comparing the review comments with a feature word list to determine product relevance, a means for analyzing patterns of the review comments to determine unnatural reviews, a means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, and a means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, thereby enabling users to select products based on highly reliable review information.

[0514] "Review comments" are opinions and evaluations posted by users about products on online shopping platforms.

[0515] "Data preprocessing" refers to the process of removing unnecessary spaces and special characters from the text data of collected review comments, and performing tokenization and language standardization.

[0516] The "characteristic word list" is a list of keywords based on product attributes that are set to evaluate the reliability of review comments about products.

[0517] "Product relevance" is an index that indicates the degree to which a review comment mentions a specific product and the degree of match with the feature word list.

[0518] "Pattern analysis" is the process of analyzing the posting patterns of review comments and determining whether the same or similar content has been posted consecutively within a certain period of time.

[0519] An "unnatural review" is a review that is deemed to deviate from normal user behavior, such as when similar comments are posted consecutively within a short period of time.

[0520] A "copy-and-paste review" is a comment that is posted by simply copying an existing review comment, and refers to a review whose content is exactly the same.

[0521] "Feedback" is a process that notifies users and administrators of the relevance, unnaturalness, and copy-and-paste judgment results of review comments, and allows only appropriate reviews.

[0522] The present invention provides a system for evaluating the reliability of review comments on an online shopping platform, thereby enabling users to obtain more accurate review information. Specific embodiments are described in detail below.

[0523] Collecting review comments

[0524] The server periodically retrieves all review comments using the shopping platform's API. This periodically saves the latest review comments, including the most recent ones, to the database. The specific software used is the Python requests library.

[0525] Data Preprocessing

[0526] The server preprocesses the collected review comment data, including removing unnecessary whitespace and special characters, tokenizing the text data, and unifying the language. This process uses the Python nltk library and re module.

[0527] Product relevance determination

[0528] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a feature word list such as "image quality," "battery," and "operability" is set. The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is deemed to be "highly relevant to the product." The specific software used is the Python scikit-learn library.

[0529] Specific examples

[0530] Product: Smartphone

[0531] Review comment: "The image quality of this phone is excellent."

[0532] Since the word list includes "image quality," it is judged to have a high score.

[0533] Identifying unnatural reviews

[0534] The server analyzes patterns in review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time. If the server detects an unnatural pattern, it flags those reviews. This process uses the Python Pandas library and time series analysis.

[0535] Specific examples

[0536] Review 1: "Great product!"

[0537] Review 2: "Great product!"

[0538] Review 3: "Great product!"

[0539] The server will flag these as unnatural reviews.

[0540] Judgment on copy-paste reviews

[0541] The server compares each new review comment submitted with all existing review comments to see if there are any identical comments. If the server finds a match, it flags the review as a "copy-and-paste review." The software uses the Python difflib library.

[0542] Specific examples

[0543] Review 1: "This product is amazing!"

[0544] Review 2: "This product is amazing!"

[0545] The server determines that Review 2 is a copy-and-paste review and flags it.

[0546] feedback

[0547] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. Inappropriate reviews are filtered out and an error message is displayed to the user. The server also notifies the administrator of the review content and the result. This process uses the Python Flask or Django framework and a mail server (SMTP) for notifications.

[0548] Prompt Sentence Examples

[0549] Please rate the user-submitted review comment "This phone has excellent battery life" and provide feedback based on the following criteria:

[0550] conditions:

[0551] 1. To determine whether a review is highly relevant to the product, it is scored based on a list of characteristic words such as "battery," "duty," and "excellent."

[0552] 2. Check to see if similar reviews have been posted consecutively within a certain period of time.

[0553] 3. Compare with existing review comments to check whether the review is a copy-and-paste one.

[0554] This system will enable users to select products based on reliable review information, and is expected to improve the evaluation quality of the entire online shopping platform.

[0555] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0556] Step 1:

[0557] Collects review comments. The server periodically retrieves all review comments through the shopping platform's API. The input is the API request, and the output is the retrieved review comment data in JSON format. The server sends the API request and stores the received data in a database.

[0558] Step 2:

[0559] Perform data preprocessing. The server performs preprocessing on the collected review comment data. The input is JSON-formatted review comment data, and the output is preprocessed text data. The server removes unnecessary whitespace and special characters, tokenizes the text data, and unifies the language.

[0560] Step 3:

[0561] The server determines product relevance by comparing the preprocessed review comments with a feature word list based on product attributes. The input is the preprocessed text data and feature word list, and the output is scored review comment data. The server scores how often words included in the word list appear in the review comments, and if the score is above a certain level, it determines the review to be "highly relevant to the product."

[0562] Step 4:

[0563] The server determines whether a review is unnatural. It analyzes patterns in review comments posted within a certain period of time. The input is review comment data from a specific period of time, and the output is the determination result and flagged data. The server checks whether reviews with similar content have been posted consecutively within a short period of time, and if an unnatural pattern is detected, it flags those reviews.

[0564] Step 5:

[0565] The server determines whether a review is a copy-and-paste review. Whenever a new review comment is posted, it compares it with all existing review comments. The input is the new review comment and existing review comment data, and the output is the determination result and flagged data. If the server finds a matching comment, it flags the review as a "copy-and-paste review."

[0566] Step 6:

[0567] Feedback is provided. The server allows a review to be posted only if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. The input is the review comment and various judgment results, and the output is a notification to the user and administrator. The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the judgment results.

[0568] (Application example 1)

[0569] 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."

[0570] On conventional shopping platforms, many unreliable review comments are posted, causing users to make incorrect purchasing decisions. Furthermore, there is also the problem of similar reviews and copy-paste reviews being posted in a short period of time, which reduces the reliability of the reviews. Therefore, there is a need for the development of a system that provides users with reliable review information and helps them make more accurate purchasing decisions.

[0571] 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.

[0572] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a list of characteristic words to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, means for evaluating the reliability of the review comments in real time and providing feedback, and means for displaying reviews in order of reliability. This allows users to receive reliable reviews in real time and make more accurate purchasing decisions based on them.

[0573] A "review comment" is a written comment written by a user expressing their evaluation or opinion of a product or service.

[0574] "Means of collection" refers to the methods by which data is obtained from online platforms and entered into the system.

[0575] "Data preprocessing" is the process of removing unnecessary information and formatting acquired data to make it easier to analyze.

[0576] A "characteristic word list" is a list of important keywords related to a specific product or service.

[0577] "Product relevance" is the degree to which a review comment relates to a particular product or service.

[0578] "Means for analyzing patterns" are methods for detecting certain templates or similarities in review comments.

[0579] "Unnatural reviews" are comments that deviate from normal user behavior, such as those posted in large numbers in a short period of time or those that contain repeated, repeated comments.

[0580] A "copy-and-paste review" is a review comment in which the exact same content as another comment is copied and pasted.

[0581] "Feedback methods" are methods for notifying users and administrators of the evaluation results and encouraging them to take appropriate action.

[0582] "Means for assessing trustworthiness in real time" refers to a method for automatically determining the accuracy and trustworthiness of a review the moment it is posted.

[0583] "Displaying in order of reliability" is a method of displaying highly rated review comments at the top, making them easier for users to refer to.

[0584] The system of the present invention evaluates the reliability of review comments on a shopping platform, allowing users to obtain more accurate review information. The system is implemented through the following process.

[0585] Hardware used

[0586] Smartphone: Used as a device for users to post review comments.

[0587] Server: A central computer that collects review comments, performs data preprocessing, analysis, evaluation, and feedback.

[0588] API server: Used as an interface to obtain review comments from the shopping platform.

[0589] Software used

[0590] Python: Used for server-side data processing and analysis.

[0591] Flask: A Python web framework used to build API servers.

[0592] MySQL: Used as a database management system to store collected review comments.

[0593] Natural language processing library: Used to preprocess and parse review comments.

[0594] Data processing and calculation flow

[0595] 1. Collecting review comments

[0596] The server periodically retrieves review comments from the shopping platform via API. This API server is built using Flask, and the retrieved data is stored in a MySQL database.

[0597] 2. Data Preprocessing

[0598] The server preprocesses the retrieved review comment data using a Python natural language processing library, which includes removing unnecessary whitespace and special characters, splitting review comments, and standardizing the language.

[0599] 3. Product relevance determination

[0600] The server compares the review comment with a list of characteristic words based on product attributes. For example, a review about a smartphone would use keywords such as "image quality," "battery," and "operability." The server then scores the review comment, and if it scores above a certain level, it determines that the review comment is "highly relevant to the product."

[0601] 4. Identifying unnatural reviews

[0602] The server analyzes review comments posted within a certain period of time to check for similar reviews posted in succession within a short period of time, and if it detects any unnatural patterns, it flags those reviews.

[0603] 5. Judging copy-paste reviews

[0604] The server compares the newly submitted review comment with all existing review comments, checking for exact matches, and if an exact match is found, flagging it as a "copy-and-paste review."

[0605] 6. Feedback

[0606] The server allows reviews to be posted only if they are not flagged as being related to the product, unnatural, or copied and pasted, and provides real-time feedback to users. It also has a function to display reviews in order of reliability.

[0607] Examples of specific examples and prompts

[0608] As a concrete example, consider the following review comment:

[0609] Product: Smartphone

[0610] Reviewer: "This phone's battery lasts a long time."

[0611] An example of a prompt is:

[0612] "Please tell us if this review is relevant to the product: 'This phone's battery lasts a long time.'"

[0613] "Tell me if this review fits an unnatural pattern: 'This phone's battery lasts a long time.'"

[0614] "Please tell us if this review exactly matches an existing review: 'This phone's battery lasts a long time.'"

[0615] This allows users to choose products based on reliable reviews, improving the quality of ratings across online shopping platforms.

[0616] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0617] Step 1:

[0618] Collect review comments:

[0619] The server periodically collects review comments using the API provided by the shopping platform. The server receives the review comments obtained via the API in JSON format and stores them in a MySQL database. The input is the JSON data obtained from the API, and the output is the review comments stored in the database.

[0620] Step 2:

[0621] Data preprocessing:

[0622] The server preprocesses the collected review comment data. It uses Python's regular expression library to remove unnecessary whitespace and special characters, split the review comments into sentences, and converts data in different formats into a unified format. The input is the raw review comments, and the output is the preprocessed, clean data.

[0623] Step 3:

[0624] Product relevance determination:

[0625] The server compares the preprocessed review comments with a pre-defined feature word list. The server calculates the frequency of occurrence of words included in the feature word list, and determines that a product is highly relevant if the frequency is above a certain score. The input is the preprocessed review comments and the feature word list, and the output is data flagged as highly relevant reviews.

[0626] Step 4:

[0627] Unnatural review verdict:

[0628] The server compares the preprocessed review comments with other review comments posted within a certain period of time and analyzes patterns. If a large number of similar review comments are posted in a short period of time, the server determines that they are unnatural reviews and flags them. The input is the preprocessed review comments and past review comments, and the output is data flagged as unnatural reviews.

[0629] Step 5:

[0630] Copypaste review verdict:

[0631] The server compares the newly posted review comment with all existing review comments based on their content similarity. If the server finds a review comment with exactly the same content, it flags it as a copy-and-paste review. The input is the new review comment and the existing review comments, and the output is the data determined to be a copy-and-paste review.

[0632] Step 6:

[0633] feedback:

[0634] The server provides real-time feedback to reviewers based on the review's relevance, unnaturalness, and copy-and-paste judgment results. Reviews are displayed on the user's device in order of reliability. Reviews with low reliability are rejected, and the user is notified of the reason. The input is the review judgment results, and the output is a feedback message.

[0635] This will enable users to make more accurate purchasing decisions based on reliable review comments.

[0636] 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.

[0637] The system of the present invention evaluates the reliability of review comments on a shopping platform, and further optimizes review information by recognizing user sentiment. The program processing of this system is explained below in natural language.

[0638] Process Overview

[0639] 1. Collecting review comments

[0640] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[0641] 2. Data Preprocessing

[0642] The server cleans the collected review comments, removing unnecessary whitespace, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[0643] 3. Product relevance determination

[0644] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a smartphone review, a word list such as "image quality," "battery," and "operability" would be prepared.

[0645] The server scores the number of times the words in the word list appear in the review comments, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[0646] Examples:

[0647] Product: Smartphone

[0648] Review comment: "The image quality of this smartphone is outstanding."

[0649] Score: The word list includes "image quality," so the score is high.

[0650] 4. Identifying unnatural reviews

[0651] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, checking to see if reviews with similar content or writing style have been posted in succession within a short period of time.

[0652] If the server detects an unnatural pattern, it flags the review comment.

[0653] Examples:

[0654] Review 1: "Great product!"

[0655] Review 2: "Great product!"

[0656] Review 3: "Great product!"

[0657] Verdict: Unnatural review and flagging.

[0658] 5. Judging copy-paste reviews

[0659] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[0660] If the server finds a match, it flags the review comment as a "copy and paste review."

[0661] Examples:

[0662] Review 1: "This product is amazing!"

[0663] Review 2: "This product is amazing!"

[0664] Verdict: Copy-paste review and flagging.

[0665] 6. Emotion Recognition by Emotion Engine

[0666] The server analyzes the review comments through a sentiment engine and categorizes them into positive, negative, or neutral sentiment.

[0667] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[0668] Examples:

[0669] Review comment: "The battery on this phone dies so quickly it's useless."

[0670] Sentiment analysis result: Negative

[0671] Score: Reflected as a low rating.

[0672] 7. Feedback and Filtering

[0673] The server will only allow a review comment to be posted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the post.

[0674] The server notifies the user and the administrator of the feedback along with the emotion score from the emotion engine, while the administrator is notified to further review and take action.

[0675] As a result, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind. The introduction of the emotion engine further improves the quality and reliability of reviews, and product evaluation information is more accurately reflected.

[0676] The processing flow will be explained below.

[0677] Step 1:

[0678] Collecting review comments

[0679] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[0680] Step 2:

[0681] Data Preprocessing

[0682] The server cleans the collected review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences to make them analyzable.

[0683] Step 3:

[0684] Product relevance determination

[0685] The server compares review comments using a word list based on the features and attributes of each product. It compares the words in the review comments with the word list and generates a score. For example, a word list related to smartphones might include "image quality," "battery," and "operability."

[0686] The server scores the frequency with which words included in the word list appear, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[0687] Step 4:

[0688] Identifying unnatural reviews

[0689] The server analyzes all review comments posted within a certain period of time to identify patterns in content, determining whether similar reviews or writing styles have been posted in succession within a short period of time, particularly comparing sentence length, vocabulary, and sentence structure.

[0690] If the server detects an unnatural pattern, it flags the review comment.

[0691] Step 5:

[0692] Judgment on copy-paste reviews

[0693] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[0694] If the server finds a match, it flags the review comment as a "copy and paste review."

[0695] Step 6:

[0696] Emotion recognition by emotion engine

[0697] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral sentiments. Specifically, it uses a natural language processing algorithm to extract and score emotional expressions contained in the sentences.

[0698] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[0699] Step 7:

[0700] Feedback and Filtering

[0701] The server will only allow a review comment to be posted if it is not flagged for product relevance, unnaturalness, or copy-and-paste. If a flag is raised, the server will display an error message to the user explaining why the post was rejected.

[0702] The server filters out inappropriate reviews and notifies users and administrators, including the sentiment score generated by the sentiment engine. Administrators are sent detailed information, such as the review content, the reason for flagging, and the sentiment score, and are offered help in reviewing the review if necessary.

[0703] Based on these specific steps, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind.The integration of the sentiment engine further improves the quality and reliability of reviews, and more accurately reflects product evaluation information.

[0704] Example 2

[0705] 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."

[0706] Conventional shopping platforms lack an appropriate mechanism for determining the authenticity of review comments, which creates the risk of users making purchasing decisions based on inaccurate information. Therefore, there is a need to improve the reliability of review comments and provide a safe and secure environment for users. Furthermore, it is necessary to accurately capture the sentiment of review comments in order to more accurately reflect product evaluation information.

[0707] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for collecting review comments, means for performing data preprocessing of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing the patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for analyzing the sentiment of the review comments, means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results, and means for rejecting the posting of a review comment if it is unnatural or a copy-and-paste. This improves the reliability of the review comments, provides an environment where users can refer to reviews with confidence, and makes it possible to more accurately reflect product evaluation information.

[0708] A "review comment" is a text of an opinion or evaluation written by a user on a shopping platform regarding a product or service.

[0709] "Data preprocessing" is the process of removing unnecessary elements from collected review comments and converting them into a format suitable for analysis.

[0710] A "characteristic word list" is a list of keywords related to a particular product or service that is used to compare with review comments.

[0711] "Product relevance determination" is the process of determining whether a review comment is relevant to a particular product or service.

[0712] "Pattern analysis" is the process of analyzing the content and format of review comments to detect specific patterns or anomalies.

[0713] An "unnatural review" is a review comment that shows an unusual pattern, such as being posted repeatedly in a short period of time or in a different style.

[0714] A "copy-and-paste review" refers to a review comment that has exactly the same content as an existing review comment.

[0715] "Sentiment analysis" is the process of analyzing the content of review comments and classifying them into sentiments such as positive, negative, or neutral.

[0716] "Feedback" refers to the provision of information to notify the user or administrator of the results of analysis or judgment, and to encourage them to take necessary action.

[0717] "Posting refusal" refers to the action of not allowing a user to post in order to prevent unnatural reviews or copy-and-paste reviews from being posted.

[0718] "Server" refers to a computer system that manages and executes a series of processes, including collection of review comments, pre-processing, analysis, judgment, feedback, and rejection of submissions.

[0719] The system of the present invention evaluates the reliability of review comments on a shopping platform, recognizes user sentiment, and optimizes review information. Specific operations and processes for this purpose are described below.

[0720] 1. Collecting review comments

[0721] The system retrieves review comments through the shopping platform's API. The server periodically sends requests to the API endpoint and stores the retrieved review comments in a database. Specifically, MySQL or PostgreSQL can be used as the database.

[0722] For example, the server accesses the API endpoint "https: / / example.com / api / reviews" and stores the retrieved review comments in the "reviews" table.

[0723] 2. Data Preprocessing

[0724] The server cleans the retrieved review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences. This preprocessing can be performed using NLPK, a Python NLP library.

[0725] For example, "This product wonderful !" is text cleaned to "This product is great!" and then tokenized.

[0726] 3. Product relevance determination

[0727] The server generates a word list based on the characteristics and attributes of each product and compares it with review comments. For example, a word list containing keywords such as "image quality," "battery," and "operability" is used for smartphone reviews.

[0728] Specifically, the frequency of keywords in review comments is scored, and if the score is above a certain level, it is determined to be a "review with high product relevance." For example, the server analyzes a comment such as "The image quality of this smartphone is outstanding," and assigns a high score because "image quality" is included in the word list.

[0729] 4. Identifying unnatural reviews

[0730] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, specifically, whether review comments with similar content or format have been posted consecutively within a short period of time.

[0731] If an unnatural pattern is detected, the review comments are flagged. For example, if the server sees multiple reviews with the same content saying "Great product!" posted within a short period of time, it will flag them as unnatural.

[0732] 5. Judging copy-paste reviews

[0733] Each time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms (e.g., Python's difflib library) to determine whether they match.

[0734] If a match is found, the server flags the review comment as a "copy-and-paste review." For example, if a comment already exists that says "This product is great!", the server adds a flag.

[0735] 6. Emotion Recognition by Emotion Engine

[0736] The server feeds the review comments into a sentiment analysis engine and classifies them into positive, negative, or neutral sentiments. Specifically, sentiment analysis libraries such as TextBlob and VADER can be used.

[0737] This allows us to calculate an emotion score and reflect that score in the overall rating of the product. For example, we can use TextBlob to analyze a comment like "This phone's battery runs out quickly, so it's useless" and assign it a negative score.

[0738] 7. Feedback and Filtering

[0739] When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user explaining the reason for the post.

[0740] The emotion engine also notifies users and administrators of the emotion score, aiming to improve the reliability of reviews. For example, the server may notify users that "this review has been flagged as unnatural" and encourage them to resubmit. It may also notify administrators that "a new unnatural review has been detected. Please check it."

[0741] Example prompts for generative AI models

[0742] "Category this review comment as it relates to a specific product and its sentiment. Example: 'The image quality of this phone is outstanding.'"

[0743] "Please judge whether the review comments for this listing are unnatural. Example: 'Great product!', 'Great product!'"

[0744] "Check if this review comment matches an existing comment or is a copypasta review. Example: 'This product is amazing!'"

[0745] The above is an embodiment of the system of the present invention.

[0746] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0747] Step 1:

[0748] Collect review comments:

[0749] Input: Shopping platform API endpoint URL

[0750] Specific operation: The server periodically sends a request to an API endpoint to retrieve review comments. For example, it accesses the API endpoint "https: / / example.com / api / reviews".

[0751] Data processing: Metadata (user ID, product ID, posting date and time, etc.) is added to the acquired review comments.

[0752] Output: A dataset containing review comments and metadata

[0753] Specific operation: The server stores the acquired dataset in a database (e.g., MySQL or PostgreSQL).

[0754] Step 2:

[0755] Data preprocessing:

[0756] Input: Review comments stored in the database

[0757] What happens: The server cleans the review comments, specifically using a Python NLP library (e.g., NLPK) to remove unnecessary whitespace, special characters, and HTML tags.

[0758] Data processing: Review comments are tokenized into words and sentences.

[0759] Output: Clean, tokenized review comments

[0760] Specific actions: For example, "This product wonderful !" to "This product is great!" and then split into individual words.

[0761] Step 3:

[0762] Product relevance determination:

[0763] Input: Clean and tokenized review comments, product feature wordlist

[0764] Specific operation: The server compares the review comments with the feature word list, which includes keywords based on product attributes such as "image quality," "battery," and "operability."

[0765] Data calculation: Score the frequency of occurrence of keywords in review comments.

[0766] Output: Product relevance score

[0767] Specific operation: For example, a comment such as "The image quality of this smartphone is outstanding" will be given a high score because the word list includes "image quality."

[0768] Step 4:

[0769] Unnatural review verdict:

[0770] Input: All review comments posted within a certain period

[0771] Specific operation: The server analyzes the content patterns of review comments, checking whether reviews with the same content or writing style have been posted consecutively.

[0772] Data arithmetic: Analysis to detect unnatural patterns

[0773] Output: Review comments flagged as unnatural

[0774] What happens: For example, if multiple reviews with the same content, such as "Great product!", are posted within a short period of time, we flag them as unusual.

[0775] Step 5:

[0776] Copypaste review verdict:

[0777] Input: New review comment, Existing review comment

[0778] What happens: The server uses advanced text comparison algorithms (e.g., Python's difflib library) to compare the new comment with existing review comments.

[0779] Data calculation: The process of checking the content of comments for consistency

[0780] Output: Review comments flagged as copypasta reviews

[0781] Specific behavior: For example, if a comment saying "This product is amazing!" already exists, add a flag.

[0782] Step 6:

[0783] Emotion Recognition with Emotion Engine:

[0784] Input: Review comment

[0785] Specific operation: The server inputs the review comments into a sentiment analysis engine (e.g., TextBlob or VADER) and classifies the sentiment.

[0786] Data arithmetic: Calculating sentiment scores

[0787] Output: Review comments with sentiment scores

[0788] Specific behavior: For example, a comment such as "This phone's battery runs out quickly, making it useless" is scored as negative.

[0789] Step 7:

[0790] Feedback and Filtering:

[0791] Input: Review comments (flagged for relevance, unnaturalness, and copy-paste), sentiment score

[0792] Specific behavior: When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user.

[0793] Data Calculation: Feedback and Submission Judgment

[0794] Output: Notification to users and administrators

[0795] Specific behavior: For example, the server notifies the user that "this review has been flagged as unnatural" and encourages them to resubmit, while simultaneously notifying the administrator that "a new unnatural review has been detected."

[0796] The above are the specific processing steps of the program of this system.

[0797] (Application example 2)

[0798] 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."

[0799] Conventional review comment evaluation systems often lack the functionality to fully guarantee the reliability of reviews. For example, reviews that are unrelated to the product, unnatural reviews, or copy-and-paste reviews are often mixed in, making it difficult for users to obtain accurate information. Furthermore, there is a demand for more accurate product evaluations by analyzing the sentiment of review comments. However, current systems lack sentiment analysis functionality, making it impossible to improve the quality and reliability of reviews.

[0800] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0801] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for performing sentiment analysis of the review comments, and means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results. This allows users to obtain highly reliable review information and evaluate products more accurately.

[0802] "Review comments" refer to impressions and evaluations written by users about products and services.

[0803] "Means of collection" refers to the process or function of obtaining specific information from the internet or databases.

[0804] "Data preprocessing" refers to the process of analyzing collected data or performing pre-analysis processing to remove unnecessary parts and format the data.

[0805] A "characteristic word list" is a list of important keywords related to a specific product or category.

[0806] "Method for determining product relevance" refers to the method or algorithm used to evaluate how relevant a review comment is to a particular product.

[0807] "Methods for analyzing patterns to identify unnatural reviews" refers to methods for analyzing the content and posting patterns of review comments to detect unnatural content and typical actions.

[0808] A "copy-and-paste review" refers to a review comment with the same content, copied and pasted exactly from another review comment.

[0809] "Sentiment analysis" refers to the technology of analyzing user emotions and intentions from text data and classifying them as positive, negative, or neutral.

[0810] "Feedback" refers to the process of returning analysis and evaluation results to the user or system.

[0811] MODE FOR CARRYING OUT THE INVENTION

[0812] This invention is a system for improving the reliability of review comments on shopping platforms and providing optimal information to users. This system collects review comments, performs data preprocessing, and performs product relevance, unnaturalness, copy-and-paste, and sentiment analysis before providing feedback.

[0813] System Overview

[0814] The server includes the following means:

[0815] 1. How to collect review comments:

[0816] The server periodically retrieves all review comments from the shopping platform via API, and the collected review comments are stored in a database.

[0817] 2. Methods for data preprocessing of review comments:

[0818] The server cleans the collected review comments by removing unnecessary spaces, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[0819] 3. A method for determining product relevance by comparing review comments with a feature word list:

[0820] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is prepared. The server scores the number of times words included in the word list appear in the review comments, and if the score is above a certain level, it is determined to be a "review highly relevant to the product."

[0821] 4. Methods for analyzing review comment patterns and identifying unnatural reviews:

[0822] The server analyzes all review comments posted within a certain period of time to check for patterns in content, checking to see if similar reviews or writing styles have been posted in succession within a short period of time, and if an unnatural pattern is detected, flagging the review comment.

[0823] 5. How to compare review comments and identify comments with identical content as copy-paste reviews:

[0824] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to check whether an identical review comment already exists, and if a match is found, flags the review comment as a "copy-and-paste review."

[0825] 6. Methods for sentiment analysis of review comments:

[0826] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral emotions. Based on the scoring by the emotion engine, the server calculates an emotion score for the review comments and reflects it in the overall rating of the product.

[0827] 7. Feedback on review relevance, unnaturalness, copy-paste determination, and sentiment analysis results:

[0828] When a review comment is attempted to be posted, the server allows it to be posted only if it is not flagged as being related to the product, unnatural, or copied and pasted. If a flag is raised, an error message is displayed to the user explaining the reason for the post. Feedback is also provided to the user and administrator along with the emotion score calculated by the emotion engine, and administrators are notified of the review and prompted to take further action.

[0829] Hardware and software used

[0830] Hardware: Servers, cloud storage devices

[0831] Software: Python, requests, BeautifulSoup, re, nltk, textblob, scikit-learn, emotion engine (TextBlob, etc.)

[0832] Specific examples

[0833] As a concrete example, the procedure for processing a review comment such as "The image quality of this smartphone is outstanding" is shown below. This comment is determined to be highly relevant to the product because it contains the keyword "image quality," and is approved because it does not match unnatural reviews or copy-paste reviews.

[0834] Prompt Sentence Examples

[0835] Example prompts to input to a generative AI model:

[0836] Let's analyze the new review comment, "The image quality of this phone is outstanding." We'll rate it based on the following criteria:

[0837] 1. Product Relevance

[0838] 2. Unnatural patterns

[0839] 3. Copy and paste

[0840] 4. Sentiment analysis

[0841] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0842] Step 1:

[0843] Collecting review comments

[0844] The server periodically calls the shopping platform's API to retrieve review comments for each product. As input, it uses the API endpoint URL and authentication information. As output, it generates JSON data containing the retrieved review comments. This data is stored in the server's database.

[0845] Step 2:

[0846] Data Preprocessing

[0847] The server cleans the collected review comments. As input, it uses the text data of the review comments stored in the database. Specifically, it removes unnecessary whitespace, special characters, and HTML tags, and tokenizes the review comments into words and sentences. As output, clean text data is generated and passed to the next step.

[0848] Step 3:

[0849] Product relevance determination

[0850] The server evaluates the relevance of review comments using a word list based on the features and attributes of each product. As input, it uses the clean review comments and the feature word list. Specifically, it scores how often words included in the feature word list appear in the review comments. As output, it generates a relevance score for each review comment.

[0851] Step 4:

[0852] Identifying unnatural reviews

[0853] The server analyzes all review comments posted within a certain period of time and checks for patterns. As input, it uses text data from review comments over a certain period of time and the current review comment. Specifically, it applies a text pattern matching algorithm to evaluate the similarity. As output, it generates a flag indicating whether the review is unnatural.

[0854] Step 5:

[0855] Judgment on copy-paste reviews

[0856] Each time a new review comment is posted, the server compares it with all existing review comments. As input, it uses data from the new review comment and the existing review comments. Specifically, it uses advanced text comparison algorithms (e.g., cosine similarity) to check whether there are any identical review comments. As output, it generates a flag indicating whether the review is a copy-and-paste review.

[0857] Step 6:

[0858] sentiment analysis

[0859] The server parses the review comments into a sentiment engine. It uses the clean review comments as input. It sends the text to the sentiment engine (e.g., TextBlob) and gets a sentiment score: positive, negative, or neutral. It generates a sentiment score for each review comment as output.

[0860] Step 7:

[0861] Feedback and Filtering

[0862] When an attempt is made to post a review comment, the server provides feedback based on the judgment results obtained in the previous steps. All judgment results (relevance, unnaturalness, copy-paste judgment results, and sentiment score) are used as input. Specific behavior is to allow posting only if no flags are raised, and to display an error message to the user if a flag is raised. More detailed feedback is notified to the administrator. As output, an error message or approval message is displayed to the user, and the review comment is saved in the database.

[0863] 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.

[0864] 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.

[0865] 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.

[0866] [Third embodiment]

[0867] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0868] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[0869] 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).

[0870] 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.

[0871] 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.

[0872] 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).

[0873] 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.

[0874] 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.

[0875] 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.

[0876] 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.

[0877] 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.

[0878] 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."

[0879] The system of the present invention evaluates the reliability of review comments on shopping platforms and provides appropriate feedback to enable users to obtain more accurate review information. The program processing of this system is explained below in natural language.

[0880] Process Overview

[0881] 1. Collecting review comments

[0882] The server periodically retrieves all review comments from the shopping platform via API, which allows new review comments to be added to the system.

[0883] 2. Data Preprocessing

[0884] The server preprocesses the collected review comment data, which includes removing unnecessary spaces and special characters, splitting review comments, and standardizing the language.

[0885] 3. Product relevance determination

[0886] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is set.

[0887] The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is determined to be a "review with high product relevance."

[0888] Examples:

[0889] Product: Smartphone

[0890] Review comment: "The image quality of this phone is excellent."

[0891] Score: The word list includes "image quality," so the score is high.

[0892] 4. Identifying unnatural reviews

[0893] The server analyzes the patterns of review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time.

[0894] The server flags reviews if it detects any unnatural patterns.

[0895] Examples:

[0896] Review 1: "Great product!"

[0897] Review 2: "Great product!"

[0898] Review 3: "Great product!"

[0899] Verdict: Unnatural review and flagging.

[0900] 5. Judging copy-paste reviews

[0901] Every time a new review comment is posted, the server compares it with all existing review comments to see if there are any that are exactly the same.

[0902] If the server finds an exact match, it flags the review as a "copy-and-paste review."

[0903] Examples:

[0904] Review 1: "This product is amazing!"

[0905] Review 2: "This product is amazing!"

[0906] Verdict: Copy-paste review and flagging.

[0907] 6. Feedback

[0908] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being related to the product, unnatural, or copied and pasted.

[0909] The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the result of the judgment.

[0910] This system allows users to select products based on reliable review information, improving the evaluation quality of the entire online shopping platform.

[0911] The processing flow will be explained below.

[0912] Step 1:

[0913] Collecting review comments

[0914] The server uses the API from the shopping platform to periodically retrieve all review comments. This collection includes review comments for all products.

[0915] Step 2:

[0916] Data Preprocessing

[0917] The server stores the collected review comments as text data. Next, it performs a cleaning process to remove unnecessary spaces, special characters, and HTML tags from the review comments. It then tokenizes the review comments into words and sentences to prepare them for analysis.

[0918] Step 3:

[0919] Product relevance determination

[0920] The server pre-sets a word list based on the characteristics and attributes of each product. When a review comment is posted, the server compares the words in the comment with the word list. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is used. The server scores the frequency with which words from the word list appear in the comment, and if the score is above a certain level, it determines that the review is "highly relevant to the product."

[0921] Step 4:

[0922] Identifying unnatural reviews

[0923] The server analyzes all review comments posted within a certain period of time to check for patterns in content, determining whether reviews with similar content or writing style have been posted in succession within a short period of time. If the server detects an unnatural pattern, it flags the review comment.

[0924] Step 5:

[0925] Judgment on copy-paste reviews

[0926] Each time a new review comment is submitted, the server compares it with all existing review comments. Using sophisticated text comparison algorithms, the server determines whether a review comment with the exact same content already exists. If the server finds a match, it flags the review comment as a "copy-and-paste review."

[0927] Step 6:

[0928] Feedback and Filtering

[0929] The server will allow a review comment to be posted only if it is not flagged as irrelevant, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the posting. The server will also notify the administrator of the fraudulent review and help the administrator review the review.

[0930] Based on these concrete steps, we will be able to provide users with reliable review information and improve the quality of the entire shopping platform.

[0931] Example 1

[0932] 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."

[0933] On online shopping platforms, the reliability of review comments is important for users to use when selecting products. However, fake reviews, copy-paste reviews, and reviews with unnatural patterns are rampant, making it difficult for users to obtain accurate information. Therefore, there is a need for a system that can evaluate the reliability of review comments and provide users with accurate review information.

[0934] 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.

[0935] In this invention, the server includes a means for collecting review comments, a means for performing data preprocessing of the review comments, a means for comparing the review comments with a feature word list to determine product relevance, a means for analyzing patterns of the review comments to determine unnatural reviews, a means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, and a means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, thereby enabling users to select products based on highly reliable review information.

[0936] "Review comments" are opinions and evaluations posted by users about products on online shopping platforms.

[0937] "Data preprocessing" refers to the process of removing unnecessary spaces and special characters from the text data of collected review comments, and performing tokenization and language standardization.

[0938] The "characteristic word list" is a list of keywords based on product attributes that are set to evaluate the reliability of review comments about products.

[0939] "Product relevance" is an index that indicates the degree to which a review comment mentions a specific product and the degree of match with the feature word list.

[0940] "Pattern analysis" is the process of analyzing the posting patterns of review comments and determining whether the same or similar content has been posted consecutively within a certain period of time.

[0941] An "unnatural review" is a review that is deemed to deviate from normal user behavior, such as when similar comments are posted consecutively within a short period of time.

[0942] A "copy-and-paste review" is a comment that is posted by simply copying an existing review comment, and refers to a review whose content is exactly the same.

[0943] "Feedback" is a process that notifies users and administrators of the relevance, unnaturalness, and copy-and-paste judgment results of review comments, and allows only appropriate reviews.

[0944] The present invention provides a system for evaluating the reliability of review comments on an online shopping platform, thereby enabling users to obtain more accurate review information. Specific embodiments are described in detail below.

[0945] Collecting review comments

[0946] The server periodically retrieves all review comments using the shopping platform's API. This periodically saves the latest review comments, including the most recent ones, to the database. The specific software used is the Python requests library.

[0947] Data Preprocessing

[0948] The server preprocesses the collected review comment data, including removing unnecessary whitespace and special characters, tokenizing the text data, and unifying the language. This process uses the Python nltk library and re module.

[0949] Product relevance determination

[0950] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a feature word list such as "image quality," "battery," and "operability" is set. The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is deemed to be "highly relevant to the product." The specific software used is the Python scikit-learn library.

[0951] Specific examples

[0952] Product: Smartphone

[0953] Review comment: "The image quality of this phone is excellent."

[0954] Since the word list includes "image quality," it is judged to have a high score.

[0955] Identifying unnatural reviews

[0956] The server analyzes patterns in review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time. If the server detects an unnatural pattern, it flags those reviews. This process uses the Python Pandas library and time series analysis.

[0957] Specific examples

[0958] Review 1: "Great product!"

[0959] Review 2: "Great product!"

[0960] Review 3: "Great product!"

[0961] The server will flag these as unnatural reviews.

[0962] Judgment on copy-paste reviews

[0963] The server compares each new review comment submitted with all existing review comments to see if there are any identical comments. If the server finds a match, it flags the review as a "copy-and-paste review." The software uses the Python difflib library.

[0964] Specific examples

[0965] Review 1: "This product is amazing!"

[0966] Review 2: "This product is amazing!"

[0967] The server determines that Review 2 is a copy-and-paste review and flags it.

[0968] feedback

[0969] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. Inappropriate reviews are filtered out and an error message is displayed to the user. The server also notifies the administrator of the review content and the result. This process uses the Python Flask or Django framework and a mail server (SMTP) for notifications.

[0970] Prompt Sentence Examples

[0971] Please rate the user-submitted review comment "This phone has excellent battery life" and provide feedback based on the following criteria:

[0972] conditions:

[0973] 1. To determine whether a review is highly relevant to the product, it is scored based on a list of characteristic words such as "battery," "duty," and "excellent."

[0974] 2. Check to see if similar reviews have been posted consecutively within a certain period of time.

[0975] 3. Compare with existing review comments to check whether the review is a copy-and-paste one.

[0976] This system will enable users to select products based on reliable review information, and is expected to improve the evaluation quality of the entire online shopping platform.

[0977] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0978] Step 1:

[0979] Collects review comments. The server periodically retrieves all review comments through the shopping platform's API. The input is the API request, and the output is the retrieved review comment data in JSON format. The server sends the API request and stores the received data in a database.

[0980] Step 2:

[0981] Perform data preprocessing. The server performs preprocessing on the collected review comment data. The input is JSON-formatted review comment data, and the output is preprocessed text data. The server removes unnecessary whitespace and special characters, tokenizes the text data, and unifies the language.

[0982] Step 3:

[0983] The server determines product relevance by comparing the preprocessed review comments with a feature word list based on product attributes. The input is the preprocessed text data and feature word list, and the output is scored review comment data. The server scores how often words included in the word list appear in the review comments, and if the score is above a certain level, it determines the review to be "highly relevant to the product."

[0984] Step 4:

[0985] The server determines whether a review is unnatural. It analyzes patterns in review comments posted within a certain period of time. The input is review comment data from a specific period of time, and the output is the determination result and flagged data. The server checks whether reviews with similar content have been posted consecutively within a short period of time, and if an unnatural pattern is detected, it flags those reviews.

[0986] Step 5:

[0987] The server determines whether a review is a copy-and-paste review. Whenever a new review comment is posted, it compares it with all existing review comments. The input is the new review comment and existing review comment data, and the output is the determination result and flagged data. If the server finds a matching comment, it flags the review as a "copy-and-paste review."

[0988] Step 6:

[0989] Feedback is provided. The server allows a review to be posted only if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. The input is the review comment and various judgment results, and the output is a notification to the user and administrator. The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the judgment results.

[0990] (Application example 1)

[0991] 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."

[0992] On conventional shopping platforms, many unreliable review comments are posted, causing users to make incorrect purchasing decisions. Furthermore, there is also the problem of similar reviews and copy-paste reviews being posted in a short period of time, which reduces the reliability of the reviews. Therefore, there is a need for the development of a system that provides users with reliable review information and helps them make more accurate purchasing decisions.

[0993] 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.

[0994] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a list of characteristic words to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, means for evaluating the reliability of the review comments in real time and providing feedback, and means for displaying reviews in order of reliability. This allows users to receive reliable reviews in real time and make more accurate purchasing decisions based on them.

[0995] A "review comment" is a written comment written by a user expressing their evaluation or opinion of a product or service.

[0996] "Means of collection" refers to the methods by which data is obtained from online platforms and entered into the system.

[0997] "Data preprocessing" is the process of removing unnecessary information and formatting acquired data to make it easier to analyze.

[0998] A "characteristic word list" is a list of important keywords related to a specific product or service.

[0999] "Product relevance" is the degree to which a review comment relates to a particular product or service.

[1000] "Means for analyzing patterns" are methods for detecting certain templates or similarities in review comments.

[1001] "Unnatural reviews" are comments that deviate from normal user behavior, such as those posted in large numbers in a short period of time or those that contain repeated, repeated comments.

[1002] A "copy-and-paste review" is a review comment in which the exact same content as another comment is copied and pasted.

[1003] "Feedback methods" are methods for notifying users and administrators of the evaluation results and encouraging them to take appropriate action.

[1004] "Means for assessing trustworthiness in real time" refers to a method for automatically determining the accuracy and trustworthiness of a review the moment it is posted.

[1005] "Displaying in order of reliability" is a method of displaying highly rated review comments at the top, making them easier for users to refer to.

[1006] The system of the present invention evaluates the reliability of review comments on a shopping platform, allowing users to obtain more accurate review information. The system is implemented through the following process.

[1007] Hardware used

[1008] Smartphone: Used as a device for users to post review comments.

[1009] Server: A central computer that collects review comments, performs data preprocessing, analysis, evaluation, and feedback.

[1010] API server: Used as an interface to obtain review comments from the shopping platform.

[1011] Software used

[1012] Python: Used for server-side data processing and analysis.

[1013] Flask: A Python web framework used to build API servers.

[1014] MySQL: Used as a database management system to store collected review comments.

[1015] Natural language processing library: Used to preprocess and parse review comments.

[1016] Data processing and calculation flow

[1017] 1. Collecting review comments

[1018] The server periodically retrieves review comments from the shopping platform via API. This API server is built using Flask, and the retrieved data is stored in a MySQL database.

[1019] 2. Data Preprocessing

[1020] The server preprocesses the retrieved review comment data using a Python natural language processing library, which includes removing unnecessary whitespace and special characters, splitting review comments, and standardizing the language.

[1021] 3. Product relevance determination

[1022] The server compares the review comment with a list of characteristic words based on product attributes. For example, a review about a smartphone would use keywords such as "image quality," "battery," and "operability." The server then scores the review comment, and if it scores above a certain level, it determines that the review comment is "highly relevant to the product."

[1023] 4. Identifying unnatural reviews

[1024] The server analyzes review comments posted within a certain period of time to check for similar reviews posted in succession within a short period of time, and if it detects any unnatural patterns, it flags those reviews.

[1025] 5. Judging copy-paste reviews

[1026] The server compares the newly submitted review comment with all existing review comments, checking for exact matches, and if an exact match is found, flagging it as a "copy-and-paste review."

[1027] 6. Feedback

[1028] The server allows reviews to be posted only if they are not flagged as being related to the product, unnatural, or copied and pasted, and provides real-time feedback to users. It also has a function to display reviews in order of reliability.

[1029] Examples of specific examples and prompts

[1030] As a concrete example, consider the following review comment:

[1031] Product: Smartphone

[1032] Reviewer: "This phone's battery lasts a long time."

[1033] An example of a prompt is:

[1034] "Please tell us if this review is relevant to the product: 'This phone's battery lasts a long time.'"

[1035] "Tell me if this review fits an unnatural pattern: 'This phone's battery lasts a long time.'"

[1036] "Please tell us if this review exactly matches an existing review: 'This phone's battery lasts a long time.'"

[1037] This allows users to choose products based on reliable reviews, improving the quality of ratings across online shopping platforms.

[1038] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1039] Step 1:

[1040] Collect review comments:

[1041] The server periodically collects review comments using the API provided by the shopping platform. The server receives the review comments obtained via the API in JSON format and stores them in a MySQL database. The input is the JSON data obtained from the API, and the output is the review comments stored in the database.

[1042] Step 2:

[1043] Data preprocessing:

[1044] The server preprocesses the collected review comment data. It uses Python's regular expression library to remove unnecessary whitespace and special characters, split the review comments into sentences, and converts data in different formats into a unified format. The input is the raw review comments, and the output is the preprocessed, clean data.

[1045] Step 3:

[1046] Product relevance determination:

[1047] The server compares the preprocessed review comments with a pre-defined feature word list. The server calculates the frequency of occurrence of words included in the feature word list, and determines that a product is highly relevant if the frequency is above a certain score. The input is the preprocessed review comments and the feature word list, and the output is data flagged as highly relevant reviews.

[1048] Step 4:

[1049] Unnatural review verdict:

[1050] The server compares the preprocessed review comments with other review comments posted within a certain period of time and analyzes patterns. If a large number of similar review comments are posted in a short period of time, the server determines that they are unnatural reviews and flags them. The input is the preprocessed review comments and past review comments, and the output is data flagged as unnatural reviews.

[1051] Step 5:

[1052] Copypaste review verdict:

[1053] The server compares the newly posted review comment with all existing review comments based on their content similarity. If the server finds a review comment with exactly the same content, it flags it as a copy-and-paste review. The input is the new review comment and the existing review comments, and the output is the data determined to be a copy-and-paste review.

[1054] Step 6:

[1055] feedback:

[1056] The server provides real-time feedback to reviewers based on the review's relevance, unnaturalness, and copy-and-paste judgment results. Reviews are displayed on the user's device in order of reliability. Reviews with low reliability are rejected, and the user is notified of the reason. The input is the review judgment results, and the output is a feedback message.

[1057] This will enable users to make more accurate purchasing decisions based on reliable review comments.

[1058] 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.

[1059] The system of the present invention evaluates the reliability of review comments on a shopping platform, and further optimizes review information by recognizing user sentiment. The program processing of this system is explained below in natural language.

[1060] Process Overview

[1061] 1. Collecting review comments

[1062] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[1063] 2. Data Preprocessing

[1064] The server cleans the collected review comments, removing unnecessary whitespace, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[1065] 3. Product relevance determination

[1066] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a smartphone review, a word list such as "image quality," "battery," and "operability" would be prepared.

[1067] The server scores the number of times the words in the word list appear in the review comments, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[1068] Examples:

[1069] Product: Smartphone

[1070] Review comment: "The image quality of this smartphone is outstanding."

[1071] Score: The word list includes "image quality," so the score is high.

[1072] 4. Identifying unnatural reviews

[1073] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, checking to see if reviews with similar content or writing style have been posted in succession within a short period of time.

[1074] If the server detects an unnatural pattern, it flags the review comment.

[1075] Examples:

[1076] Review 1: "Great product!"

[1077] Review 2: "Great product!"

[1078] Review 3: "Great product!"

[1079] Verdict: Unnatural review and flagging.

[1080] 5. Judging copy-paste reviews

[1081] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[1082] If the server finds a match, it flags the review comment as a "copy and paste review."

[1083] Examples:

[1084] Review 1: "This product is amazing!"

[1085] Review 2: "This product is amazing!"

[1086] Verdict: Copy-paste review and flagging.

[1087] 6. Emotion Recognition by Emotion Engine

[1088] The server analyzes the review comments through a sentiment engine and categorizes them into positive, negative, or neutral sentiment.

[1089] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[1090] Examples:

[1091] Review comment: "The battery on this phone dies so quickly it's useless."

[1092] Sentiment analysis result: Negative

[1093] Score: Reflected as a low rating.

[1094] 7. Feedback and Filtering

[1095] The server will only allow a review comment to be posted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the post.

[1096] The server notifies the user and the administrator of the feedback along with the emotion score from the emotion engine, while the administrator is notified to further review and take action.

[1097] As a result, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind. The introduction of the emotion engine further improves the quality and reliability of reviews, and product evaluation information is more accurately reflected.

[1098] The processing flow will be explained below.

[1099] Step 1:

[1100] Collecting review comments

[1101] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[1102] Step 2:

[1103] Data Preprocessing

[1104] The server cleans the collected review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences to make them analyzable.

[1105] Step 3:

[1106] Product relevance determination

[1107] The server compares review comments using a word list based on the features and attributes of each product. It compares the words in the review comments with the word list and generates a score. For example, a word list related to smartphones might include "image quality," "battery," and "operability."

[1108] The server scores the frequency with which words included in the word list appear, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[1109] Step 4:

[1110] Identifying unnatural reviews

[1111] The server analyzes all review comments posted within a certain period of time to identify patterns in content, determining whether similar reviews or writing styles have been posted in succession within a short period of time, particularly comparing sentence length, vocabulary, and sentence structure.

[1112] If the server detects an unnatural pattern, it flags the review comment.

[1113] Step 5:

[1114] Judgment on copy-paste reviews

[1115] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[1116] If the server finds a match, it flags the review comment as a "copy and paste review."

[1117] Step 6:

[1118] Emotion recognition by emotion engine

[1119] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral sentiments. Specifically, it uses a natural language processing algorithm to extract and score emotional expressions contained in the sentences.

[1120] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[1121] Step 7:

[1122] Feedback and Filtering

[1123] The server will only allow a review comment to be posted if it is not flagged for product relevance, unnaturalness, or copy-and-paste. If a flag is raised, the server will display an error message to the user explaining why the post was rejected.

[1124] The server filters out inappropriate reviews and notifies users and administrators, including the sentiment score generated by the sentiment engine. Administrators are sent detailed information, such as the review content, the reason for flagging, and the sentiment score, and are offered help in reviewing the review if necessary.

[1125] Based on these specific steps, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind.The integration of the sentiment engine further improves the quality and reliability of reviews, and more accurately reflects product evaluation information.

[1126] Example 2

[1127] 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."

[1128] Conventional shopping platforms lack an appropriate mechanism for determining the authenticity of review comments, which creates the risk of users making purchasing decisions based on inaccurate information. Therefore, there is a need to improve the reliability of review comments and provide a safe and secure environment for users. Furthermore, it is necessary to accurately capture the sentiment of review comments in order to more accurately reflect product evaluation information.

[1129] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for collecting review comments, means for performing data preprocessing of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing the patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for analyzing the sentiment of the review comments, means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results, and means for rejecting the posting of a review comment if it is unnatural or a copy-and-paste. This improves the reliability of the review comments, provides an environment where users can refer to reviews with confidence, and makes it possible to more accurately reflect product evaluation information.

[1130] A "review comment" is a text of an opinion or evaluation written by a user on a shopping platform regarding a product or service.

[1131] "Data preprocessing" is the process of removing unnecessary elements from collected review comments and converting them into a format suitable for analysis.

[1132] A "characteristic word list" is a list of keywords related to a particular product or service that is used to compare with review comments.

[1133] "Product relevance determination" is the process of determining whether a review comment is relevant to a particular product or service.

[1134] "Pattern analysis" is the process of analyzing the content and format of review comments to detect specific patterns or anomalies.

[1135] An "unnatural review" is a review comment that shows an unusual pattern, such as being posted repeatedly in a short period of time or in a different style.

[1136] A "copy-and-paste review" refers to a review comment that has exactly the same content as an existing review comment.

[1137] "Sentiment analysis" is the process of analyzing the content of review comments and classifying them into sentiments such as positive, negative, or neutral.

[1138] "Feedback" refers to the provision of information to notify the user or administrator of the results of analysis or judgment, and to encourage them to take necessary action.

[1139] "Posting refusal" refers to the action of not allowing a user to post in order to prevent unnatural reviews or copy-and-paste reviews from being posted.

[1140] "Server" refers to a computer system that manages and executes a series of processes, including collection of review comments, pre-processing, analysis, judgment, feedback, and rejection of submissions.

[1141] The system of the present invention evaluates the reliability of review comments on a shopping platform, recognizes user sentiment, and optimizes review information. Specific operations and processes for this purpose are described below.

[1142] 1. Collecting review comments

[1143] The system retrieves review comments through the shopping platform's API. The server periodically sends requests to the API endpoint and stores the retrieved review comments in a database. Specifically, MySQL or PostgreSQL can be used as the database.

[1144] For example, the server accesses the API endpoint "https: / / example.com / api / reviews" and stores the retrieved review comments in the "reviews" table.

[1145] 2. Data Preprocessing

[1146] The server cleans the retrieved review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences. This preprocessing can be performed using NLPK, a Python NLP library.

[1147] For example, "This product wonderful !" is text cleaned to "This product is great!" and then tokenized.

[1148] 3. Product relevance determination

[1149] The server generates a word list based on the characteristics and attributes of each product and compares it with review comments. For example, a word list containing keywords such as "image quality," "battery," and "operability" is used for smartphone reviews.

[1150] Specifically, the frequency of keywords in review comments is scored, and if the score is above a certain level, it is determined to be a "review with high product relevance." For example, the server analyzes a comment such as "The image quality of this smartphone is outstanding," and assigns a high score because "image quality" is included in the word list.

[1151] 4. Identifying unnatural reviews

[1152] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, specifically, whether review comments with similar content or format have been posted consecutively within a short period of time.

[1153] If an unnatural pattern is detected, the review comments are flagged. For example, if the server sees multiple reviews with the same content saying "Great product!" posted within a short period of time, it will flag them as unnatural.

[1154] 5. Judging copy-paste reviews

[1155] Each time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms (e.g., Python's difflib library) to determine whether they match.

[1156] If a match is found, the server flags the review comment as a "copy-and-paste review." For example, if a comment already exists that says "This product is great!", the server adds a flag.

[1157] 6. Emotion Recognition by Emotion Engine

[1158] The server feeds the review comments into a sentiment analysis engine and classifies them into positive, negative, or neutral sentiments. Specifically, sentiment analysis libraries such as TextBlob and VADER can be used.

[1159] This allows us to calculate an emotion score and reflect that score in the overall rating of the product. For example, we can use TextBlob to analyze a comment like "This phone's battery runs out quickly, so it's useless" and assign it a negative score.

[1160] 7. Feedback and Filtering

[1161] When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user explaining the reason for the post.

[1162] The emotion engine also notifies users and administrators of the emotion score, aiming to improve the reliability of reviews. For example, the server may notify users that "this review has been flagged as unnatural" and encourage them to resubmit. It may also notify administrators that "a new unnatural review has been detected. Please check it."

[1163] Example prompts for generative AI models

[1164] "Category this review comment as it relates to a specific product and its sentiment. Example: 'The image quality of this phone is outstanding.'"

[1165] "Please judge whether the review comments for this listing are unnatural. Example: 'Great product!', 'Great product!'"

[1166] "Check if this review comment matches an existing comment or is a copypasta review. Example: 'This product is amazing!'"

[1167] The above is an embodiment of the system of the present invention.

[1168] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1169] Step 1:

[1170] Collect review comments:

[1171] Input: Shopping platform API endpoint URL

[1172] Specific operation: The server periodically sends a request to an API endpoint to retrieve review comments. For example, it accesses the API endpoint "https: / / example.com / api / reviews".

[1173] Data processing: Metadata (user ID, product ID, posting date and time, etc.) is added to the acquired review comments.

[1174] Output: A dataset containing review comments and metadata

[1175] Specific operation: The server stores the acquired dataset in a database (e.g., MySQL or PostgreSQL).

[1176] Step 2:

[1177] Data preprocessing:

[1178] Input: Review comments stored in the database

[1179] What happens: The server cleans the review comments, specifically using a Python NLP library (e.g., NLPK) to remove unnecessary whitespace, special characters, and HTML tags.

[1180] Data processing: Review comments are tokenized into words and sentences.

[1181] Output: Clean, tokenized review comments

[1182] Specific actions: For example, "This product wonderful !" to "This product is great!" and then split into individual words.

[1183] Step 3:

[1184] Product relevance determination:

[1185] Input: Clean and tokenized review comments, product feature wordlist

[1186] Specific operation: The server compares the review comments with the feature word list, which includes keywords based on product attributes such as "image quality," "battery," and "operability."

[1187] Data calculation: Score the frequency of occurrence of keywords in review comments.

[1188] Output: Product relevance score

[1189] Specific operation: For example, a comment such as "The image quality of this smartphone is outstanding" will be given a high score because the word list includes "image quality."

[1190] Step 4:

[1191] Unnatural review verdict:

[1192] Input: All review comments posted within a certain period

[1193] Specific operation: The server analyzes the content patterns of review comments, checking whether reviews with the same content or writing style have been posted consecutively.

[1194] Data arithmetic: Analysis to detect unnatural patterns

[1195] Output: Review comments flagged as unnatural

[1196] What happens: For example, if multiple reviews with the same content, such as "Great product!", are posted within a short period of time, we flag them as unusual.

[1197] Step 5:

[1198] Copypaste review verdict:

[1199] Input: New review comment, Existing review comment

[1200] What happens: The server uses advanced text comparison algorithms (e.g., Python's difflib library) to compare the new comment with existing review comments.

[1201] Data calculation: The process of checking the content of comments for consistency

[1202] Output: Review comments flagged as copypasta reviews

[1203] Specific behavior: For example, if a comment saying "This product is amazing!" already exists, add a flag.

[1204] Step 6:

[1205] Emotion Recognition with Emotion Engine:

[1206] Input: Review comment

[1207] Specific operation: The server inputs the review comments into a sentiment analysis engine (e.g., TextBlob or VADER) and classifies the sentiment.

[1208] Data arithmetic: Calculating sentiment scores

[1209] Output: Review comments with sentiment scores

[1210] Specific behavior: For example, a comment such as "This phone's battery runs out quickly, making it useless" is scored as negative.

[1211] Step 7:

[1212] Feedback and Filtering:

[1213] Input: Review comments (flagged for relevance, unnaturalness, and copy-paste), sentiment score

[1214] Specific behavior: When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user.

[1215] Data Calculation: Feedback and Submission Judgment

[1216] Output: Notification to users and administrators

[1217] Specific behavior: For example, the server notifies the user that "this review has been flagged as unnatural" and encourages them to resubmit, while simultaneously notifying the administrator that "a new unnatural review has been detected."

[1218] The above are the specific processing steps of the program of this system.

[1219] (Application example 2)

[1220] 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."

[1221] Conventional review comment evaluation systems often lack the functionality to fully guarantee the reliability of reviews. For example, reviews that are unrelated to the product, unnatural reviews, or copy-and-paste reviews are often mixed in, making it difficult for users to obtain accurate information. Furthermore, there is a demand for more accurate product evaluations by analyzing the sentiment of review comments. However, current systems lack sentiment analysis functionality, making it impossible to improve the quality and reliability of reviews.

[1222] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1223] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for performing sentiment analysis of the review comments, and means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results. This allows users to obtain highly reliable review information and evaluate products more accurately.

[1224] "Review comments" refer to impressions and evaluations written by users about products and services.

[1225] "Means of collection" refers to the process or function of obtaining specific information from the internet or databases.

[1226] "Data preprocessing" refers to the process of analyzing collected data or performing pre-analysis processing to remove unnecessary parts and format the data.

[1227] A "characteristic word list" is a list of important keywords related to a specific product or category.

[1228] "Method for determining product relevance" refers to the method or algorithm used to evaluate how relevant a review comment is to a particular product.

[1229] "Methods for analyzing patterns to identify unnatural reviews" refers to methods for analyzing the content and posting patterns of review comments to detect unnatural content and typical actions.

[1230] A "copy-and-paste review" refers to a review comment with the same content, copied and pasted exactly from another review comment.

[1231] "Sentiment analysis" refers to the technology of analyzing user emotions and intentions from text data and classifying them as positive, negative, or neutral.

[1232] "Feedback" refers to the process of returning analysis and evaluation results to the user or system.

[1233] MODE FOR CARRYING OUT THE INVENTION

[1234] This invention is a system for improving the reliability of review comments on shopping platforms and providing optimal information to users. This system collects review comments, performs data preprocessing, and performs product relevance, unnaturalness, copy-and-paste, and sentiment analysis before providing feedback.

[1235] System Overview

[1236] The server includes the following means:

[1237] 1. How to collect review comments:

[1238] The server periodically retrieves all review comments from the shopping platform via API, and the collected review comments are stored in a database.

[1239] 2. Methods for data preprocessing of review comments:

[1240] The server cleans the collected review comments by removing unnecessary spaces, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[1241] 3. A method for determining product relevance by comparing review comments with a feature word list:

[1242] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is prepared. The server scores the number of times words included in the word list appear in the review comments, and if the score is above a certain level, it is determined to be a "review highly relevant to the product."

[1243] 4. Methods for analyzing review comment patterns and identifying unnatural reviews:

[1244] The server analyzes all review comments posted within a certain period of time to check for patterns in content, checking to see if similar reviews or writing styles have been posted in succession within a short period of time, and if an unnatural pattern is detected, flagging the review comment.

[1245] 5. How to compare review comments and identify comments with identical content as copy-paste reviews:

[1246] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to check whether an identical review comment already exists, and if a match is found, flags the review comment as a "copy-and-paste review."

[1247] 6. Methods for sentiment analysis of review comments:

[1248] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral emotions. Based on the scoring by the emotion engine, the server calculates an emotion score for the review comments and reflects it in the overall rating of the product.

[1249] 7. Feedback on review relevance, unnaturalness, copy-paste determination, and sentiment analysis results:

[1250] When a review comment is attempted to be posted, the server allows it to be posted only if it is not flagged as being related to the product, unnatural, or copied and pasted. If a flag is raised, an error message is displayed to the user explaining the reason for the post. Feedback is also provided to the user and administrator along with the emotion score calculated by the emotion engine, and administrators are notified of the review and prompted to take further action.

[1251] Hardware and software used

[1252] Hardware: Servers, cloud storage devices

[1253] Software: Python, requests, BeautifulSoup, re, nltk, textblob, scikit-learn, emotion engine (TextBlob, etc.)

[1254] Specific examples

[1255] As a concrete example, the procedure for processing a review comment such as "The image quality of this smartphone is outstanding" is shown below. This comment is determined to be highly relevant to the product because it contains the keyword "image quality," and is approved because it does not match unnatural reviews or copy-paste reviews.

[1256] Prompt Sentence Examples

[1257] Example prompts to input to a generative AI model:

[1258] Let's analyze the new review comment, "The image quality of this phone is outstanding." We'll rate it based on the following criteria:

[1259] 1. Product Relevance

[1260] 2. Unnatural patterns

[1261] 3. Copy and paste

[1262] 4. Sentiment analysis

[1263] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1264] Step 1:

[1265] Collecting review comments

[1266] The server periodically calls the shopping platform's API to retrieve review comments for each product. As input, it uses the API endpoint URL and authentication information. As output, it generates JSON data containing the retrieved review comments. This data is stored in the server's database.

[1267] Step 2:

[1268] Data Preprocessing

[1269] The server cleans the collected review comments. As input, it uses the text data of the review comments stored in the database. Specifically, it removes unnecessary whitespace, special characters, and HTML tags, and tokenizes the review comments into words and sentences. As output, clean text data is generated and passed to the next step.

[1270] Step 3:

[1271] Product relevance determination

[1272] The server evaluates the relevance of review comments using a word list based on the features and attributes of each product. As input, it uses the clean review comments and the feature word list. Specifically, it scores how often words included in the feature word list appear in the review comments. As output, it generates a relevance score for each review comment.

[1273] Step 4:

[1274] Identifying unnatural reviews

[1275] The server analyzes all review comments posted within a certain period of time and checks for patterns. As input, it uses text data from review comments over a certain period of time and the current review comment. Specifically, it applies a text pattern matching algorithm to evaluate the similarity. As output, it generates a flag indicating whether the review is unnatural.

[1276] Step 5:

[1277] Judgment on copy-paste reviews

[1278] Each time a new review comment is posted, the server compares it with all existing review comments. As input, it uses data from the new review comment and the existing review comments. Specifically, it uses advanced text comparison algorithms (e.g., cosine similarity) to check whether there are any identical review comments. As output, it generates a flag indicating whether the review is a copy-and-paste review.

[1279] Step 6:

[1280] sentiment analysis

[1281] The server parses the review comments into a sentiment engine. It uses the clean review comments as input. It sends the text to the sentiment engine (e.g., TextBlob) and gets a sentiment score: positive, negative, or neutral. It generates a sentiment score for each review comment as output.

[1282] Step 7:

[1283] Feedback and Filtering

[1284] When an attempt is made to post a review comment, the server provides feedback based on the judgment results obtained in the previous steps. All judgment results (relevance, unnaturalness, copy-paste judgment results, and sentiment score) are used as input. Specific behavior is to allow posting only if no flags are raised, and to display an error message to the user if a flag is raised. More detailed feedback is notified to the administrator. As output, an error message or approval message is displayed to the user, and the review comment is saved in the database.

[1285] 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.

[1286] 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.

[1287] 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.

[1288] [Fourth embodiment]

[1289] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1290] 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.

[1291] 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).

[1292] 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.

[1293] 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.

[1294] 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).

[1295] 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.

[1296] 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.

[1297] 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.

[1298] 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.

[1299] 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.

[1300] 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.

[1301] 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."

[1302] The system of the present invention evaluates the reliability of review comments on shopping platforms and provides appropriate feedback to enable users to obtain more accurate review information. The program processing of this system is explained below in natural language.

[1303] Process Overview

[1304] 1. Collecting review comments

[1305] The server periodically retrieves all review comments from the shopping platform via API, which allows new review comments to be added to the system.

[1306] 2. Data Preprocessing

[1307] The server preprocesses the collected review comment data, which includes removing unnecessary spaces and special characters, splitting review comments, and standardizing the language.

[1308] 3. Product relevance determination

[1309] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is set.

[1310] The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is determined to be a "review with high product relevance."

[1311] Examples:

[1312] Product: Smartphone

[1313] Review comment: "The image quality of this phone is excellent."

[1314] Score: The word list includes "image quality," so the score is high.

[1315] 4. Identifying unnatural reviews

[1316] The server analyzes the patterns of review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time.

[1317] The server flags reviews if it detects any unnatural patterns.

[1318] Examples:

[1319] Review 1: "Great product!"

[1320] Review 2: "Great product!"

[1321] Review 3: "Great product!"

[1322] Verdict: Unnatural review and flagging.

[1323] 5. Judging copy-paste reviews

[1324] Every time a new review comment is posted, the server compares it with all existing review comments to see if there are any that are exactly the same.

[1325] If the server finds an exact match, it flags the review as a "copy-and-paste review."

[1326] Examples:

[1327] Review 1: "This product is amazing!"

[1328] Review 2: "This product is amazing!"

[1329] Verdict: Copy-paste review and flagging.

[1330] 6. Feedback

[1331] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being related to the product, unnatural, or copied and pasted.

[1332] The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the result of the judgment.

[1333] This system allows users to select products based on reliable review information, improving the evaluation quality of the entire online shopping platform.

[1334] The processing flow will be explained below.

[1335] Step 1:

[1336] Collecting review comments

[1337] The server uses the API from the shopping platform to periodically retrieve all review comments. This collection includes review comments for all products.

[1338] Step 2:

[1339] Data Preprocessing

[1340] The server stores the collected review comments as text data. Next, it performs a cleaning process to remove unnecessary spaces, special characters, and HTML tags from the review comments. It then tokenizes the review comments into words and sentences to prepare them for analysis.

[1341] Step 3:

[1342] Product relevance determination

[1343] The server pre-sets a word list based on the characteristics and attributes of each product. When a review comment is posted, the server compares the words in the comment with the word list. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is used. The server scores the frequency with which words from the word list appear in the comment, and if the score is above a certain level, it determines that the review is "highly relevant to the product."

[1344] Step 4:

[1345] Identifying unnatural reviews

[1346] The server analyzes all review comments posted within a certain period of time to check for patterns in content, determining whether reviews with similar content or writing style have been posted in succession within a short period of time. If the server detects an unnatural pattern, it flags the review comment.

[1347] Step 5:

[1348] Judgment on copy-paste reviews

[1349] Each time a new review comment is submitted, the server compares it with all existing review comments. Using sophisticated text comparison algorithms, the server determines whether a review comment with the exact same content already exists. If the server finds a match, it flags the review comment as a "copy-and-paste review."

[1350] Step 6:

[1351] Feedback and Filtering

[1352] The server will allow a review comment to be posted only if it is not flagged as irrelevant, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the posting. The server will also notify the administrator of the fraudulent review and help the administrator review the review.

[1353] Based on these concrete steps, we will be able to provide users with reliable review information and improve the quality of the entire shopping platform.

[1354] Example 1

[1355] 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."

[1356] On online shopping platforms, the reliability of review comments is important for users to use when selecting products. However, fake reviews, copy-paste reviews, and reviews with unnatural patterns are rampant, making it difficult for users to obtain accurate information. Therefore, there is a need for a system that can evaluate the reliability of review comments and provide users with accurate review information.

[1357] 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.

[1358] In this invention, the server includes a means for collecting review comments, a means for performing data preprocessing of the review comments, a means for comparing the review comments with a feature word list to determine product relevance, a means for analyzing patterns of the review comments to determine unnatural reviews, a means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, and a means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, thereby enabling users to select products based on highly reliable review information.

[1359] "Review comments" are opinions and evaluations posted by users about products on online shopping platforms.

[1360] "Data preprocessing" refers to the process of removing unnecessary spaces and special characters from the text data of collected review comments, and performing tokenization and language standardization.

[1361] The "characteristic word list" is a list of keywords based on product attributes that are set to evaluate the reliability of review comments about products.

[1362] "Product relevance" is an index that indicates the degree to which a review comment mentions a specific product and the degree of match with the feature word list.

[1363] "Pattern analysis" is the process of analyzing the posting patterns of review comments and determining whether the same or similar content has been posted consecutively within a certain period of time.

[1364] An "unnatural review" is a review that is deemed to deviate from normal user behavior, such as when similar comments are posted consecutively within a short period of time.

[1365] A "copy-and-paste review" is a comment that is posted by simply copying an existing review comment, and refers to a review whose content is exactly the same.

[1366] "Feedback" is a process that notifies users and administrators of the relevance, unnaturalness, and copy-and-paste judgment results of review comments, and allows only appropriate reviews.

[1367] The present invention provides a system for evaluating the reliability of review comments on an online shopping platform, thereby enabling users to obtain more accurate review information. Specific embodiments are described in detail below.

[1368] Collecting review comments

[1369] The server periodically retrieves all review comments using the shopping platform's API. This periodically saves the latest review comments, including the most recent ones, to the database. The specific software used is the Python requests library.

[1370] Data Preprocessing

[1371] The server preprocesses the collected review comment data, including removing unnecessary whitespace and special characters, tokenizing the text data, and unifying the language. This process uses the Python nltk library and re module.

[1372] Product relevance determination

[1373] The server compares the review comments with a feature word list based on product attributes. For example, for a review about a smartphone, a feature word list such as "image quality," "battery," and "operability" is set. The server scores the frequency with which words included in the word list appear in the review comments, and if the score is above a certain level, the review comment is deemed to be "highly relevant to the product." The specific software used is the Python scikit-learn library.

[1374] Specific examples

[1375] Product: Smartphone

[1376] Review comment: "The image quality of this phone is excellent."

[1377] Since the word list includes "image quality," it is judged to have a high score.

[1378] Identifying unnatural reviews

[1379] The server analyzes patterns in review comments posted within a certain period of time, checking to see if similar reviews have been posted consecutively within a short period of time. If the server detects an unnatural pattern, it flags those reviews. This process uses the Python Pandas library and time series analysis.

[1380] Specific examples

[1381] Review 1: "Great product!"

[1382] Review 2: "Great product!"

[1383] Review 3: "Great product!"

[1384] The server will flag these as unnatural reviews.

[1385] Judgment on copy-paste reviews

[1386] The server compares each new review comment submitted with all existing review comments to see if there are any identical comments. If the server finds a match, it flags the review as a "copy-and-paste review." The software uses the Python difflib library.

[1387] Specific examples

[1388] Review 1: "This product is amazing!"

[1389] Review 2: "This product is amazing!"

[1390] The server determines that Review 2 is a copy-and-paste review and flags it.

[1391] feedback

[1392] When a review is submitted, the server will only allow it to be submitted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. Inappropriate reviews are filtered out and an error message is displayed to the user. The server also notifies the administrator of the review content and the result. This process uses the Python Flask or Django framework and a mail server (SMTP) for notifications.

[1393] Prompt Sentence Examples

[1394] Please rate the user-submitted review comment "This phone has excellent battery life" and provide feedback based on the following criteria:

[1395] conditions:

[1396] 1. To determine whether a review is highly relevant to the product, it is scored based on a list of characteristic words such as "battery," "duty," and "excellent."

[1397] 2. Check to see if similar reviews have been posted consecutively within a certain period of time.

[1398] 3. Compare with existing review comments to check whether the review is a copy-and-paste one.

[1399] This system will enable users to select products based on reliable review information, and is expected to improve the evaluation quality of the entire online shopping platform.

[1400] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1401] Step 1:

[1402] Collects review comments. The server periodically retrieves all review comments through the shopping platform's API. The input is the API request, and the output is the retrieved review comment data in JSON format. The server sends the API request and stores the received data in a database.

[1403] Step 2:

[1404] Perform data preprocessing. The server performs preprocessing on the collected review comment data. The input is JSON-formatted review comment data, and the output is preprocessed text data. The server removes unnecessary whitespace and special characters, tokenizes the text data, and unifies the language.

[1405] Step 3:

[1406] The server determines product relevance by comparing the preprocessed review comments with a feature word list based on product attributes. The input is the preprocessed text data and feature word list, and the output is scored review comment data. The server scores how often words included in the word list appear in the review comments, and if the score is above a certain level, it determines the review to be "highly relevant to the product."

[1407] Step 4:

[1408] The server determines whether a review is unnatural. It analyzes patterns in review comments posted within a certain period of time. The input is review comment data from a specific period of time, and the output is the determination result and flagged data. The server checks whether reviews with similar content have been posted consecutively within a short period of time, and if an unnatural pattern is detected, it flags those reviews.

[1409] Step 5:

[1410] The server determines whether a review is a copy-and-paste review. Whenever a new review comment is posted, it compares it with all existing review comments. The input is the new review comment and existing review comment data, and the output is the determination result and flagged data. If the server finds a matching comment, it flags the review as a "copy-and-paste review."

[1411] Step 6:

[1412] Feedback is provided. The server allows a review to be posted only if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. The input is the review comment and various judgment results, and the output is a notification to the user and administrator. The server filters out inappropriate reviews, displays an error message to the user, and notifies the administrator of the review content and the judgment results.

[1413] (Application example 1)

[1414] 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."

[1415] On conventional shopping platforms, many unreliable review comments are posted, causing users to make incorrect purchasing decisions. Furthermore, there is also the problem of similar reviews and copy-paste reviews being posted in a short period of time, which reduces the reliability of the reviews. Therefore, there is a need for the development of a system that provides users with reliable review information and helps them make more accurate purchasing decisions.

[1416] 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.

[1417] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a list of characteristic words to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for providing feedback on the relevance, unnaturalness, and copy-and-paste determination results of the reviews, means for evaluating the reliability of the review comments in real time and providing feedback, and means for displaying reviews in order of reliability. This allows users to receive reliable reviews in real time and make more accurate purchasing decisions based on them.

[1418] A "review comment" is a written comment written by a user expressing their evaluation or opinion of a product or service.

[1419] "Means of collection" refers to the methods by which data is obtained from online platforms and entered into the system.

[1420] "Data preprocessing" is the process of removing unnecessary information and formatting acquired data to make it easier to analyze.

[1421] A "characteristic word list" is a list of important keywords related to a specific product or service.

[1422] "Product relevance" is the degree to which a review comment relates to a particular product or service.

[1423] "Means for analyzing patterns" are methods for detecting certain templates or similarities in review comments.

[1424] "Unnatural reviews" are comments that deviate from normal user behavior, such as those posted in large numbers in a short period of time or those that contain repeated, repeated comments.

[1425] A "copy-and-paste review" is a review comment in which the exact same content as another comment is copied and pasted.

[1426] "Feedback methods" are methods for notifying users and administrators of the evaluation results and encouraging them to take appropriate action.

[1427] "Means for assessing trustworthiness in real time" refers to a method for automatically determining the accuracy and trustworthiness of a review the moment it is posted.

[1428] "Displaying in order of reliability" is a method of displaying highly rated review comments at the top, making them easier for users to refer to.

[1429] The system of the present invention evaluates the reliability of review comments on a shopping platform, allowing users to obtain more accurate review information. The system is implemented through the following process.

[1430] Hardware used

[1431] Smartphone: Used as a device for users to post review comments.

[1432] Server: A central computer that collects review comments, performs data preprocessing, analysis, evaluation, and feedback.

[1433] API server: Used as an interface to obtain review comments from the shopping platform.

[1434] Software used

[1435] Python: Used for server-side data processing and analysis.

[1436] Flask: A Python web framework used to build API servers.

[1437] MySQL: Used as a database management system to store collected review comments.

[1438] Natural language processing library: Used to preprocess and parse review comments.

[1439] Data processing and calculation flow

[1440] 1. Collecting review comments

[1441] The server periodically retrieves review comments from the shopping platform via API. This API server is built using Flask, and the retrieved data is stored in a MySQL database.

[1442] 2. Data Preprocessing

[1443] The server preprocesses the retrieved review comment data using a Python natural language processing library, which includes removing unnecessary whitespace and special characters, splitting review comments, and standardizing the language.

[1444] 3. Product relevance determination

[1445] The server compares the review comment with a list of characteristic words based on product attributes. For example, a review about a smartphone would use keywords such as "image quality," "battery," and "operability." The server then scores the review comment, and if it scores above a certain level, it determines that the review comment is "highly relevant to the product."

[1446] 4. Identifying unnatural reviews

[1447] The server analyzes review comments posted within a certain period of time to check for similar reviews posted in succession within a short period of time, and if it detects any unnatural patterns, it flags those reviews.

[1448] 5. Judging copy-paste reviews

[1449] The server compares the newly submitted review comment with all existing review comments, checking for exact matches, and if an exact match is found, flagging it as a "copy-and-paste review."

[1450] 6. Feedback

[1451] The server allows reviews to be posted only if they are not flagged as being related to the product, unnatural, or copied and pasted, and provides real-time feedback to users. It also has a function to display reviews in order of reliability.

[1452] Examples of specific examples and prompts

[1453] As a concrete example, consider the following review comment:

[1454] Product: Smartphone

[1455] Reviewer: "This phone's battery lasts a long time."

[1456] An example of a prompt is:

[1457] "Please tell us if this review is relevant to the product: 'This phone's battery lasts a long time.'"

[1458] "Tell me if this review fits an unnatural pattern: 'This phone's battery lasts a long time.'"

[1459] "Please tell us if this review exactly matches an existing review: 'This phone's battery lasts a long time.'"

[1460] This allows users to choose products based on reliable reviews, improving the quality of ratings across online shopping platforms.

[1461] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1462] Step 1:

[1463] Collect review comments:

[1464] The server periodically collects review comments using the API provided by the shopping platform. The server receives the review comments obtained via the API in JSON format and stores them in a MySQL database. The input is the JSON data obtained from the API, and the output is the review comments stored in the database.

[1465] Step 2:

[1466] Data preprocessing:

[1467] The server preprocesses the collected review comment data. It uses Python's regular expression library to remove unnecessary whitespace and special characters, split the review comments into sentences, and converts data in different formats into a unified format. The input is the raw review comments, and the output is the preprocessed, clean data.

[1468] Step 3:

[1469] Product relevance determination:

[1470] The server compares the preprocessed review comments with a pre-defined feature word list. The server calculates the frequency of occurrence of words included in the feature word list, and determines that a product is highly relevant if the frequency is above a certain score. The input is the preprocessed review comments and the feature word list, and the output is data flagged as highly relevant reviews.

[1471] Step 4:

[1472] Unnatural review verdict:

[1473] The server compares the preprocessed review comments with other review comments posted within a certain period of time and analyzes patterns. If a large number of similar review comments are posted in a short period of time, the server determines that they are unnatural reviews and flags them. The input is the preprocessed review comments and past review comments, and the output is data flagged as unnatural reviews.

[1474] Step 5:

[1475] Copypaste review verdict:

[1476] The server compares the newly posted review comment with all existing review comments based on their content similarity. If the server finds a review comment with exactly the same content, it flags it as a copy-and-paste review. The input is the new review comment and the existing review comments, and the output is the data determined to be a copy-and-paste review.

[1477] Step 6:

[1478] feedback:

[1479] The server provides real-time feedback to reviewers based on the review's relevance, unnaturalness, and copy-and-paste judgment results. Reviews are displayed on the user's device in order of reliability. Reviews with low reliability are rejected, and the user is notified of the reason. The input is the review judgment results, and the output is a feedback message.

[1480] This will enable users to make more accurate purchasing decisions based on reliable review comments.

[1481] 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.

[1482] The system of the present invention evaluates the reliability of review comments on a shopping platform, and further optimizes review information by recognizing user sentiment. The program processing of this system is explained below in natural language.

[1483] Process Overview

[1484] 1. Collecting review comments

[1485] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[1486] 2. Data Preprocessing

[1487] The server cleans the collected review comments, removing unnecessary whitespace, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[1488] 3. Product relevance determination

[1489] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a smartphone review, a word list such as "image quality," "battery," and "operability" would be prepared.

[1490] The server scores the number of times the words in the word list appear in the review comments, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[1491] Examples:

[1492] Product: Smartphone

[1493] Review comment: "The image quality of this smartphone is outstanding."

[1494] Score: The word list includes "image quality," so the score is high.

[1495] 4. Identifying unnatural reviews

[1496] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, checking to see if reviews with similar content or writing style have been posted in succession within a short period of time.

[1497] If the server detects an unnatural pattern, it flags the review comment.

[1498] Examples:

[1499] Review 1: "Great product!"

[1500] Review 2: "Great product!"

[1501] Review 3: "Great product!"

[1502] Verdict: Unnatural review and flagging.

[1503] 5. Judging copy-paste reviews

[1504] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[1505] If the server finds a match, it flags the review comment as a "copy and paste review."

[1506] Examples:

[1507] Review 1: "This product is amazing!"

[1508] Review 2: "This product is amazing!"

[1509] Verdict: Copy-paste review and flagging.

[1510] 6. Emotion Recognition by Emotion Engine

[1511] The server analyzes the review comments through a sentiment engine and categorizes them into positive, negative, or neutral sentiment.

[1512] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[1513] Examples:

[1514] Review comment: "The battery on this phone dies so quickly it's useless."

[1515] Sentiment analysis result: Negative

[1516] Score: Reflected as a low rating.

[1517] 7. Feedback and Filtering

[1518] The server will only allow a review comment to be posted if it is not flagged as being unrelated to the product, unnatural, or copied and pasted. If a flag is raised, the server will display an error message to the user explaining the reason for the post.

[1519] The server notifies the user and the administrator of the feedback along with the emotion score from the emotion engine, while the administrator is notified to further review and take action.

[1520] As a result, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind. The introduction of the emotion engine further improves the quality and reliability of reviews, and product evaluation information is more accurately reflected.

[1521] The processing flow will be explained below.

[1522] Step 1:

[1523] Collecting review comments

[1524] The server periodically retrieves all review comments from the shopping platform via API, and stores the collected review comments in a database.

[1525] Step 2:

[1526] Data Preprocessing

[1527] The server cleans the collected review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences to make them analyzable.

[1528] Step 3:

[1529] Product relevance determination

[1530] The server compares review comments using a word list based on the features and attributes of each product. It compares the words in the review comments with the word list and generates a score. For example, a word list related to smartphones might include "image quality," "battery," and "operability."

[1531] The server scores the frequency with which words included in the word list appear, and if the score is above a certain level, it determines that the review is highly relevant to the product.

[1532] Step 4:

[1533] Identifying unnatural reviews

[1534] The server analyzes all review comments posted within a certain period of time to identify patterns in content, determining whether similar reviews or writing styles have been posted in succession within a short period of time, particularly comparing sentence length, vocabulary, and sentence structure.

[1535] If the server detects an unnatural pattern, it flags the review comment.

[1536] Step 5:

[1537] Judgment on copy-paste reviews

[1538] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to see if a review comment with the exact same content already exists.

[1539] If the server finds a match, it flags the review comment as a "copy and paste review."

[1540] Step 6:

[1541] Emotion recognition by emotion engine

[1542] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral sentiments. Specifically, it uses a natural language processing algorithm to extract and score emotional expressions contained in the sentences.

[1543] The server calculates the emotion score of the review comments based on the scoring by the emotion engine and reflects it in the overall rating of the product.

[1544] Step 7:

[1545] Feedback and Filtering

[1546] The server will only allow a review comment to be posted if it is not flagged for product relevance, unnaturalness, or copy-and-paste. If a flag is raised, the server will display an error message to the user explaining why the post was rejected.

[1547] The server filters out inappropriate reviews and notifies users and administrators, including the sentiment score generated by the sentiment engine. Administrators are sent detailed information, such as the review content, the reason for flagging, and the sentiment score, and are offered help in reviewing the review if necessary.

[1548] Based on these specific steps, the system of the present invention ensures the reliability of review comments and provides a platform that users can use with peace of mind.The integration of the sentiment engine further improves the quality and reliability of reviews, and more accurately reflects product evaluation information.

[1549] Example 2

[1550] 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."

[1551] Conventional shopping platforms lack an appropriate mechanism for determining the authenticity of review comments, which creates the risk of users making purchasing decisions based on inaccurate information. Therefore, there is a need to improve the reliability of review comments and provide a safe and secure environment for users. Furthermore, it is necessary to accurately capture the sentiment of review comments in order to more accurately reflect product evaluation information.

[1552] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for collecting review comments, means for performing data preprocessing of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing the patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for analyzing the sentiment of the review comments, means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results, and means for rejecting the posting of a review comment if it is unnatural or a copy-and-paste. This improves the reliability of the review comments, provides an environment where users can refer to reviews with confidence, and makes it possible to more accurately reflect product evaluation information.

[1553] A "review comment" is a text of an opinion or evaluation written by a user on a shopping platform regarding a product or service.

[1554] "Data preprocessing" is the process of removing unnecessary elements from collected review comments and converting them into a format suitable for analysis.

[1555] A "characteristic word list" is a list of keywords related to a particular product or service that is used to compare with review comments.

[1556] "Product relevance determination" is the process of determining whether a review comment is relevant to a particular product or service.

[1557] "Pattern analysis" is the process of analyzing the content and format of review comments to detect specific patterns or anomalies.

[1558] An "unnatural review" is a review comment that shows an unusual pattern, such as being posted repeatedly in a short period of time or in a different style.

[1559] A "copy-and-paste review" refers to a review comment that has exactly the same content as an existing review comment.

[1560] "Sentiment analysis" is the process of analyzing the content of review comments and classifying them into sentiments such as positive, negative, or neutral.

[1561] "Feedback" refers to the provision of information to notify the user or administrator of the results of analysis or judgment, and to encourage them to take necessary action.

[1562] "Posting refusal" refers to the action of not allowing a user to post in order to prevent unnatural reviews or copy-and-paste reviews from being posted.

[1563] "Server" refers to a computer system that manages and executes a series of processes, including collection of review comments, pre-processing, analysis, judgment, feedback, and rejection of submissions.

[1564] The system of the present invention evaluates the reliability of review comments on a shopping platform, recognizes user sentiment, and optimizes review information. Specific operations and processes for this purpose are described below.

[1565] 1. Collecting review comments

[1566] The system retrieves review comments through the shopping platform's API. The server periodically sends requests to the API endpoint and stores the retrieved review comments in a database. Specifically, MySQL or PostgreSQL can be used as the database.

[1567] For example, the server accesses the API endpoint "https: / / example.com / api / reviews" and stores the retrieved review comments in the "reviews" table.

[1568] 2. Data Preprocessing

[1569] The server cleans the retrieved review comments. Specifically, it removes unnecessary whitespace, special characters, and HTML tags. It also tokenizes the review comments into words and sentences. This preprocessing can be performed using NLPK, a Python NLP library.

[1570] For example, "This product wonderful !" is text cleaned to "This product is great!" and then tokenized.

[1571] 3. Product relevance determination

[1572] The server generates a word list based on the characteristics and attributes of each product and compares it with review comments. For example, a word list containing keywords such as "image quality," "battery," and "operability" is used for smartphone reviews.

[1573] Specifically, the frequency of keywords in review comments is scored, and if the score is above a certain level, it is determined to be a "review with high product relevance." For example, the server analyzes a comment such as "The image quality of this smartphone is outstanding," and assigns a high score because "image quality" is included in the word list.

[1574] 4. Identifying unnatural reviews

[1575] The server analyzes all review comments posted within a certain period of time and checks for patterns in content, specifically, whether review comments with similar content or format have been posted consecutively within a short period of time.

[1576] If an unnatural pattern is detected, the review comments are flagged. For example, if the server sees multiple reviews with the same content saying "Great product!" posted within a short period of time, it will flag them as unnatural.

[1577] 5. Judging copy-paste reviews

[1578] Each time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms (e.g., Python's difflib library) to determine whether they match.

[1579] If a match is found, the server flags the review comment as a "copy-and-paste review." For example, if a comment already exists that says "This product is great!", the server adds a flag.

[1580] 6. Emotion Recognition by Emotion Engine

[1581] The server feeds the review comments into a sentiment analysis engine and classifies them into positive, negative, or neutral sentiments. Specifically, sentiment analysis libraries such as TextBlob and VADER can be used.

[1582] This allows us to calculate an emotion score and reflect that score in the overall rating of the product. For example, we can use TextBlob to analyze a comment like "This phone's battery runs out quickly, so it's useless" and assign it a negative score.

[1583] 7. Feedback and Filtering

[1584] When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user explaining the reason for the post.

[1585] The emotion engine also notifies users and administrators of the emotion score, aiming to improve the reliability of reviews. For example, the server may notify users that "this review has been flagged as unnatural" and encourage them to resubmit. It may also notify administrators that "a new unnatural review has been detected. Please check it."

[1586] Example prompts for generative AI models

[1587] "Category this review comment as it relates to a specific product and its sentiment. Example: 'The image quality of this phone is outstanding.'"

[1588] "Please judge whether the review comments for this listing are unnatural. Example: 'Great product!', 'Great product!'"

[1589] "Check if this review comment matches an existing comment or is a copypasta review. Example: 'This product is amazing!'"

[1590] The above is an embodiment of the system of the present invention.

[1591] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1592] Step 1:

[1593] Collect review comments:

[1594] Input: Shopping platform API endpoint URL

[1595] Specific operation: The server periodically sends a request to an API endpoint to retrieve review comments. For example, it accesses the API endpoint "https: / / example.com / api / reviews".

[1596] Data processing: Metadata (user ID, product ID, posting date and time, etc.) is added to the acquired review comments.

[1597] Output: A dataset containing review comments and metadata

[1598] Specific operation: The server stores the acquired dataset in a database (e.g., MySQL or PostgreSQL).

[1599] Step 2:

[1600] Data preprocessing:

[1601] Input: Review comments stored in the database

[1602] What happens: The server cleans the review comments, specifically using a Python NLP library (e.g., NLPK) to remove unnecessary whitespace, special characters, and HTML tags.

[1603] Data processing: Review comments are tokenized into words and sentences.

[1604] Output: Clean, tokenized review comments

[1605] Specific actions: For example, "This product wonderful !" to "This product is great!" and then split into individual words.

[1606] Step 3:

[1607] Product relevance determination:

[1608] Input: Clean and tokenized review comments, product feature wordlist

[1609] Specific operation: The server compares the review comments with the feature word list, which includes keywords based on product attributes such as "image quality," "battery," and "operability."

[1610] Data calculation: Score the frequency of occurrence of keywords in review comments.

[1611] Output: Product relevance score

[1612] Specific operation: For example, a comment such as "The image quality of this smartphone is outstanding" will be given a high score because the word list includes "image quality."

[1613] Step 4:

[1614] Unnatural review verdict:

[1615] Input: All review comments posted within a certain period

[1616] Specific operation: The server analyzes the content patterns of review comments, checking whether reviews with the same content or writing style have been posted consecutively.

[1617] Data arithmetic: Analysis to detect unnatural patterns

[1618] Output: Review comments flagged as unnatural

[1619] What happens: For example, if multiple reviews with the same content, such as "Great product!", are posted within a short period of time, we flag them as unusual.

[1620] Step 5:

[1621] Copypaste review verdict:

[1622] Input: New review comment, Existing review comment

[1623] What happens: The server uses advanced text comparison algorithms (e.g., Python's difflib library) to compare the new comment with existing review comments.

[1624] Data calculation: The process of checking the content of comments for consistency

[1625] Output: Review comments flagged as copypasta reviews

[1626] Specific behavior: For example, if a comment saying "This product is amazing!" already exists, add a flag.

[1627] Step 6:

[1628] Emotion Recognition with Emotion Engine:

[1629] Input: Review comment

[1630] Specific operation: The server inputs the review comments into a sentiment analysis engine (e.g., TextBlob or VADER) and classifies the sentiment.

[1631] Data arithmetic: Calculating sentiment scores

[1632] Output: Review comments with sentiment scores

[1633] Specific behavior: For example, a comment such as "This phone's battery runs out quickly, making it useless" is scored as negative.

[1634] Step 7:

[1635] Feedback and Filtering:

[1636] Input: Review comments (flagged for relevance, unnaturalness, and copy-paste), sentiment score

[1637] Specific behavior: When an attempt is made to post a review comment, the server will only allow it to be posted if it is not related to the product, unnatural, or copied and pasted. If so, an error message will be displayed to the user.

[1638] Data Calculation: Feedback and Submission Judgment

[1639] Output: Notification to users and administrators

[1640] Specific behavior: For example, the server notifies the user that "this review has been flagged as unnatural" and encourages them to resubmit, while simultaneously notifying the administrator that "a new unnatural review has been detected."

[1641] The above are the specific processing steps of the program of this system.

[1642] (Application example 2)

[1643] 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."

[1644] Conventional review comment evaluation systems often lack the functionality to fully guarantee the reliability of reviews. For example, reviews that are unrelated to the product, unnatural reviews, or copy-and-paste reviews are often mixed in, making it difficult for users to obtain accurate information. Furthermore, there is a demand for more accurate product evaluations by analyzing the sentiment of review comments. However, current systems lack sentiment analysis functionality, making it impossible to improve the quality and reliability of reviews.

[1645] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1646] In this invention, the server includes means for collecting review comments, means for preprocessing data of the review comments, means for comparing the review comments with a feature word list to determine product relevance, means for analyzing patterns of the review comments to determine unnatural reviews, means for comparing the review comments to determine comments with identical content as copy-and-paste reviews, means for performing sentiment analysis of the review comments, and means for providing feedback on the review relevance, unnaturalness, copy-and-paste determination results, and sentiment analysis results. This allows users to obtain highly reliable review information and evaluate products more accurately.

[1647] "Review comments" refer to impressions and evaluations written by users about products and services.

[1648] "Means of collection" refers to the process or function of obtaining specific information from the internet or databases.

[1649] "Data preprocessing" refers to the process of analyzing collected data or performing pre-analysis processing to remove unnecessary parts and format the data.

[1650] A "characteristic word list" is a list of important keywords related to a specific product or category.

[1651] "Method for determining product relevance" refers to the method or algorithm used to evaluate how relevant a review comment is to a particular product.

[1652] "Methods for analyzing patterns to identify unnatural reviews" refers to methods for analyzing the content and posting patterns of review comments to detect unnatural content and typical actions.

[1653] A "copy-and-paste review" refers to a review comment with the same content, copied and pasted exactly from another review comment.

[1654] "Sentiment analysis" refers to the technology of analyzing user emotions and intentions from text data and classifying them as positive, negative, or neutral.

[1655] "Feedback" refers to the process of returning analysis and evaluation results to the user or system.

[1656] MODE FOR CARRYING OUT THE INVENTION

[1657] This invention is a system for improving the reliability of review comments on shopping platforms and providing optimal information to users. This system collects review comments, performs data preprocessing, and performs product relevance, unnaturalness, copy-and-paste, and sentiment analysis before providing feedback.

[1658] System Overview

[1659] The server includes the following means:

[1660] 1. How to collect review comments:

[1661] The server periodically retrieves all review comments from the shopping platform via API, and the collected review comments are stored in a database.

[1662] 2. Methods for data preprocessing of review comments:

[1663] The server cleans the collected review comments by removing unnecessary spaces, special characters, and HTML tags, and tokenizing the review comments into words and sentences to make them analyzable.

[1664] 3. A method for determining product relevance by comparing review comments with a feature word list:

[1665] The server compares review comments with a word list based on the characteristics and attributes of each product. For example, for a review about a smartphone, a word list such as "image quality," "battery," and "operability" is prepared. The server scores the number of times words included in the word list appear in the review comments, and if the score is above a certain level, it is determined to be a "review highly relevant to the product."

[1666] 4. Methods for analyzing review comment patterns and identifying unnatural reviews:

[1667] The server analyzes all review comments posted within a certain period of time to check for patterns in content, checking to see if similar reviews or writing styles have been posted in succession within a short period of time, and if an unnatural pattern is detected, flagging the review comment.

[1668] 5. How to compare review comments and identify comments with identical content as copy-paste reviews:

[1669] Every time a new review comment is submitted, the server compares it with all existing review comments, using sophisticated text comparison algorithms to check whether an identical review comment already exists, and if a match is found, flags the review comment as a "copy-and-paste review."

[1670] 6. Methods for sentiment analysis of review comments:

[1671] The server analyzes the review comments using an emotion engine and classifies them into positive, negative, and neutral emotions. Based on the scoring by the emotion engine, the server calculates an emotion score for the review comments and reflects it in the overall rating of the product.

[1672] 7. Feedback on review relevance, unnaturalness, copy-paste determination, and sentiment analysis results:

[1673] When a review comment is attempted to be posted, the server allows it to be posted only if it is not flagged as being related to the product, unnatural, or copied and pasted. If a flag is raised, an error message is displayed to the user explaining the reason for the post. Feedback is also provided to the user and administrator along with the emotion score calculated by the emotion engine, and administrators are notified of the review and prompted to take further action.

[1674] Hardware and software used

[1675] Hardware: Servers, cloud storage devices

[1676] Software: Python, requests, BeautifulSoup, re, nltk, textblob, scikit-learn, emotion engine (TextBlob, etc.)

[1677] Specific examples

[1678] As a concrete example, the procedure for processing a review comment such as "The image quality of this smartphone is outstanding" is shown below. This comment is determined to be highly relevant to the product because it contains the keyword "image quality," and is approved because it does not match unnatural reviews or copy-paste reviews.

[1679] Prompt Sentence Examples

[1680] Example prompts to input to a generative AI model:

[1681] Let's analyze the new review comment, "The image quality of this phone is outstanding." We'll rate it based on the following criteria:

[1682] 1. Product Relevance

[1683] 2. Unnatural patterns

[1684] 3. Copy and paste

[1685] 4. Sentiment analysis

[1686] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1687] Step 1:

[1688] Collecting review comments

[1689] The server periodically calls the shopping platform's API to retrieve review comments for each product. As input, it uses the API endpoint URL and authentication information. As output, it generates JSON data containing the retrieved review comments. This data is stored in the server's database.

[1690] Step 2:

[1691] Data Preprocessing

[1692] The server cleans the collected review comments. As input, it uses the text data of the review comments stored in the database. Specifically, it removes unnecessary whitespace, special characters, and HTML tags, and tokenizes the review comments into words and sentences. As output, clean text data is generated and passed to the next step.

[1693] Step 3:

[1694] Product relevance determination

[1695] The server evaluates the relevance of review comments using a word list based on the features and attributes of each product. As input, it uses the clean review comments and the feature word list. Specifically, it scores how often words included in the feature word list appear in the review comments. As output, it generates a relevance score for each review comment.

[1696] Step 4:

[1697] Identifying unnatural reviews

[1698] The server analyzes all review comments posted within a certain period of time and checks for patterns. As input, it uses text data from review comments over a certain period of time and the current review comment. Specifically, it applies a text pattern matching algorithm to evaluate the similarity. As output, it generates a flag indicating whether the review is unnatural.

[1699] Step 5:

[1700] Judgment on copy-paste reviews

[1701] Each time a new review comment is posted, the server compares it with all existing review comments. As input, it uses data from the new review comment and the existing review comments. Specifically, it uses advanced text comparison algorithms (e.g., cosine similarity) to check whether there are any identical review comments. As output, it generates a flag indicating whether the review is a copy-and-paste review.

[1702] Step 6:

[1703] sentiment analysis

[1704] The server parses the review comments into a sentiment engine. It uses the clean review comments as input. It sends the text to the sentiment engine (e.g., TextBlob) and gets a sentiment score: positive, negative, or neutral. It generates a sentiment score for each review comment as output.

[1705] Step 7:

[1706] Feedback and Filtering

[1707] When an attempt is made to post a review comment, the server provides feedback based on the judgment results obtained in the previous steps. All judgment results (relevance, unnaturalness, copy-paste judgment results, and sentiment score) are used as input. Specific behavior is to allow posting only if no flags are raised, and to display an error message to the user if a flag is raised. More detailed feedback is notified to the administrator. As output, an error message or approval message is displayed to the user, and the review comment is saved in the database.

[1708] 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.

[1709] 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.

[1710] 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.

[1711] 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.

[1712] 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.

[1713] 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.

[1714] 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).

[1715] 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.

[1716] 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."

[1717] 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.

[1718] 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).

[1719] 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.

[1720] 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.

[1721] 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.

[1722] 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.

[1723] 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.

[1724] 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.

[1725] 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.

[1726] 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.

[1727] 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.

[1728] 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.

[1729] The following is further disclosed regarding the above embodiment.

[1730] (Claim 1)

[1731] a means for collecting review comments;

[1732] a means for performing data preprocessing of review comments;

[1733] A means for determining product relevance by comparing the review comments with a feature word list;

[1734] A means for analyzing patterns of review comments and determining unnatural reviews;

[1735] A method for comparing review comments and determining whether comments with the same content are copy-paste reviews,

[1736] A means of providing feedback on the relevance, unnaturalness, and copy-paste judgment results of reviews,

[1737] A system including:

[1738] (Claim 2)

[1739] 2. The system according to claim 1, wherein the characteristic word list for the review comments is set based on product attributes.

[1740] (Claim 3)

[1741] 2. The system according to claim 1, wherein the pattern analysis of review comments is performed based on the degree of consistency of the contents of review comments within a certain period of time.

[1742] "Example 1"

[1743] (Claim 1)

[1744] a means for collecting review comments;

[1745] a means for performing data preprocessing of review comments;

[1746] A means for determining product relevance by comparing the review comments with a feature word list;

[1747] A means for analyzing patterns of review comments and determining unnatural reviews;

[1748] A method for comparing review comments and determining whether comments with the same content are copy-paste reviews,

[1749] A means of providing feedback on the relevance, unnaturalness, and copy-paste judgment results of reviews,

[1750] A system including:

[1751] (Claim 2)

[1752] The system according to claim 1, wherein the list of characteristic words for review comments is set based on product attributes.

[1753] (Claim 3)

[1754] 2. The system according to claim 1, wherein the pattern analysis of review comments is performed based on the degree of consistency of the contents of review comments within a certain period of time.

[1755] "Application Example 1"

[1756] (Claim 1)

[1757] a means for collecting review comments;

[1758] a means for performing data preprocessing of review comments;

[1759] A means for determining product relevance by comparing the review comments with a feature word list;

[1760] A means for analyzing patterns of review comments and determining unnatural reviews;

[1761] A method for comparing review comments and determining whether comments with the same content are copy-paste reviews,

[1762] A means of providing feedback on the relevance, unnaturalness, and copy-paste judgment results of reviews,

[1763] A means to evaluate and provide feedback on the reliability of review comments in real time,

[1764] A way to display reviews in order of trustworthiness,

[1765] A system including:

[1766] (Claim 2)

[1767] 2. The system according to claim 1, wherein the characteristic word list for the review comments is set based on product attributes.

[1768] (Claim 3)

[1769] 2. The system according to claim 1, wherein the pattern analysis of review comments is performed based on the degree of consistency of the contents of review comments within a certain period of time.

[1770] "Example 2: Combining Emotion Engines"

[1771] (Claim 1)

[1772] a means for collecting review comments;

[1773] a means for performing data preprocessing of review comments;

[1774] A means for determining product relevance by comparing the review comments with a feature word list;

[1775] A means for analyzing patterns of review comments and determining unnatural reviews;

[1776] A method for comparing review comments and determining whether comments with the same content are copy-paste reviews,

[1777] A means of analyzing the sentiment of review comments;

[1778] A means of providing feedback on the relevance, unnaturalness, copy-paste judgment results, and sentiment analysis results of reviews;

[1779] If the review comments are unnatural or copied and pasted, there are ways to reject the post,

[1780] A system including:

[1781] (Claim 2)

[1782] 2. The system according to claim 1, wherein the characteristic word list for the review comments is set based on product attributes.

[1783] (Claim 3)

[1784] 2. The system according to claim 1, wherein the pattern analysis of review comments is performed based on the degree of consistency of the contents of review comments within a certain period of time.

[1785] "Application example 2 when combining emotion engines"

[1786] (Claim 1)

[1787] a means for collecting review comments;

[1788] a means for performing data preprocessing of review comments;

[1789] A means for determining product relevance by comparing the review comments with a feature word list;

[1790] A means for analyzing patterns of review comments and determining unnatural reviews;

[1791] A method for comparing review comments and determining whether comments with the same content are copy-paste reviews,

[1792] a means for performing sentiment analysis of review comments;

[1793] A means of providing feedback on the relevance, unnaturalness, copy-paste judgment results, and sentiment analysis results of reviews;

[1794] A system including:

[1795] (Claim 2)

[1796] 2. The system according to claim 1, wherein the characteristic word list for the review comments is set based on product attributes.

[1797] (Claim 3)

[1798] 2. The system according to claim 1, wherein the pattern analysis of review comments is performed based on the degree of consistency of the contents of review comments within a certain period of time. [Explanation of symbols]

[1799] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. a means for collecting review comments; a means for performing data preprocessing of review comments; A means for determining product relevance by comparing the review comments with a feature word list; A means for analyzing patterns of review comments and determining unnatural reviews; A method for comparing review comments and determining whether comments with the same content are copy-paste reviews, A means of providing feedback on the relevance, unnaturalness, and copy-paste judgment results of reviews, A system including:

2. 2. The system according to claim 1, wherein the characteristic word list for the review comments is set based on product attributes.

3. 2. The system according to claim 1, wherein the pattern analysis of the review comments is performed based on the degree of coincidence of the contents of the review comments within a certain period of time.

Images