System

Abstract

A system is provided.SOLUTION: A system includes means for collecting reviews and purchase histories of users, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, means for extracting features of the generated clean text data, means for classifying reviews and comments into clusters based on the extracted features, means for calculating features of a newly posted review and classifying the newly posted review into a closest cluster, means for evaluating similarity of the newly posted review to other reviews in the cluster to generate optimal word-of-mouth information, and means for displaying the generated word-of-mouth information to the user.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

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

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

[0004] Reviews and ratings on traditional shopping sites and content distribution services are biased by personal preferences and preconceptions, making them less reliable information for users. Furthermore, when there are many reviews, it can be difficult for users to find one that matches their preferences, reducing satisfaction with the purchasing experience. Rating bias is particularly pronounced for entertainment content such as manga, anime, and movies. There is a need for a system that can solve these issues and provide optimal word-of-mouth information for each user. [Means for solving the problem]

[0005] This invention provides a means for collecting user reviews and purchase histories and generating clean text data by removing unnecessary information from the collected data. It also provides a means for extracting features from the generated clean text data and classifying reviews and comments into clusters based on the features. It also provides a means for calculating the features of newly posted reviews and classifying them into existing clusters. It then provides a means for evaluating the similarity between the newly posted review and other reviews in the cluster to generate optimal word-of-mouth information. In this case, cosine similarity is used to evaluate the similarity, and the importance of words in the review is calculated using a TF-IDF vectorizer. Finally, it provides a means for displaying the generated word-of-mouth information to users, allowing them to easily find products and content that suit their preferences. This increases the reliability of reviews and improves user satisfaction with their purchasing experience.

[0006] "User reviews" are ratings and comments made by users about products or content on shopping sites or content distribution services.

[0007] "Purchase history" is a record of products a user has purchased or services they have used in the past.

[0008] "Clean text data" is organized text data from which unnecessary information has been removed and necessary words and phrases have been extracted.

[0009] A "feature" is a numerical representation of important attributes or elements of words or phrases in reviews or comments.

[0010] A "cluster" is a collection of similar reviews or comments grouped based on features.

[0011] "Cosine similarity" is a method for evaluating the similarity between two vectors in a vector space using the cosine of the angle between them.

[0012] "TF-IDF Vectorizer" is a method of vectorizing text data by calculating word weights based on term frequency and inverse document frequency.

[0013] "Word-of-mouth information" refers to evaluations and recommendations that are useful to users, based on user reviews and comments.

[0014] A "system" is a set of devices and programs that are composed of a combination of hardware and software and that realizes a specific function. [Brief explanation of the drawings]

[0015] [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

[0016] 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.

[0017] First, the terms used in the following description will be explained.

[0018] 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).

[0019] 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.

[0020] 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.

[0021] 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.

[0022] 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."

[0023] [First embodiment]

[0024] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0025] 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.

[0026] 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).

[0027] 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.

[0028] 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.

[0029] 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.

[0030] 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.

[0031] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] 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."

[0036] MODE FOR CARRYING OUT THE INVENTION

[0037] System Overview

[0038] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. This system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users.

[0039] System configuration

[0040] The system functions through collaboration between servers, terminals, and users as follows:

[0041] 1. The server collects user reviews and purchase history and stores them in a database.

[0042] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[0043] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[0044] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0045] 3. The server extracts features from the generated clean text data. The features represent the content of the text as numerical values.

[0046] Example: The keyword "easy to use" is assigned a high value.

[0047] 4. The server uses the KMeans algorithm to classify reviews and comments into clusters based on the extracted features.

[0048] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0049] 5. When a user posts a new review, the server calculates the review's features and classifies them into existing clusters.

[0050] Example: If user A posts "This new manga is awesome," the review will be classified into the appropriate cluster.

[0051] 6. The server calculates the cosine similarity between the new review and the reviews of other users in the cluster, and selects the most similar review.

[0052] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0053] 7. The server generates optimal review information and sends it to the device.

[0054] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0055] Example

[0056] For example, if a user wants to see reviews of new anime on a shopping site:

[0057] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0058] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features. Keywords such as "moving story" are assigned a high weight.

[0059] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0060] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0061] This system improves the reliability of reviews and allows users to have a more satisfying purchasing experience.

[0062] The processing flow will be explained below.

[0063] Step 1:

[0064] The server collects user reviews and purchase history and stores them in a database, including information such as review content, rating score, and purchase date and time.

[0065] Example: If user A reviews a product saying, "This product is very easy to use," the review and purchase history are saved on the server.

[0066] Step 2:

[0067] The server removes unnecessary information (e.g., symbols and meaningless words) from the collected data and generates clean text data.

[0068] Example: Remove unnecessary information and extract the important keyword "easy to use" as clean text data.

[0069] Step 3:

[0070] The server extracts features from the generated clean text data using a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer.

[0071] Example: The TF-IDF value of the keyword "easy to use" is calculated.

[0072] Step 4:

[0073] The server converts reviews and comments into numerical vectors based on the TF-IDF values.

[0074] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[0075] Step 5:

[0076] The server uses the KMeans algorithm to classify reviews and comments into clusters.

[0077] For example, reviews with high similarity such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0078] Step 6:

[0079] A user posts a new review.

[0080] Example: User A posts a review saying, "This new manga is great."

[0081] Step 7:

[0082] The server calculates the features of new reviews and classifies them into existing clusters.

[0083] Example: A review that says "It's the best" will be classified into cluster 2.

[0084] Step 8:

[0085] The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity.

[0086] Example: The similarity is evaluated by comparing with reviews in Cluster 2 such as "A very enjoyable work" and "Highly praised."

[0087] Step 9:

[0088] The server generates optimal reviews by selecting the most similar reviews within the cluster and summarizing them as reviews.

[0089] For example, reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0090] Step 10:

[0091] The server transmits the generated review information to the user's terminal.

[0092] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0093] Step 11:

[0094] Users use the displayed reviews to purchase or watch products or content.

[0095] Example: User A purchases a new manga based on reviews and has a highly satisfying purchasing experience.

[0096] This series of processes allows users to easily find products and content that suit their preferences based on reliable review information.

[0097] Example 1

[0098] 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."

[0099] Conventional shopping sites and content distribution services lack systems that can effectively analyze user-submitted reviews and purchase histories to provide appropriate word-of-mouth information tailored to users' preferences. As a result, users have difficulty finding products and content that match their preferences, and reviews are often unreliable, resulting in an unsatisfactory purchasing experience. A solution to this problem is needed.

[0100] 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.

[0101] In this invention, the server includes means for collecting user ratings and transaction histories, means for removing unnecessary information from the collected ratings and transaction histories to generate clean text data, and means for extracting features of the generated clean text data, thereby enabling users to easily find products and content that suit their preferences.

[0102] "User" refers to a person who uses the service or system to make reviews or purchases.

[0103] "Ratings" refer to opinions and impressions posted by users about products or content.

[0104] "Transaction history" refers to records of purchases, browsing, etc. made by a user using a service or system.

[0105] "Removing unnecessary information" refers to the process of removing unimportant information or noise from collected data.

[0106] "Clean text data" refers to data in a text format suitable for analysis after unnecessary information has been removed.

[0107] "Features" refer to important attributes or elements extracted from text data expressed as numerical values.

[0108] A "cluster" refers to a grouping of ratings or comments with similar characteristics.

[0109] The "KMeans algorithm" refers to a type of machine learning algorithm for classifying data into a specified number of clusters.

[0110] "Cosine similarity" refers to a mathematical index for calculating the similarity between two pieces of text data.

[0111] "Word-of-mouth information" refers to a collection of useful information about products and content provided to users.

[0112] "TF-IDF vectorizer" refers to a statistical method for calculating the importance of words in a text.

[0113] "Database" refers to a structured collection of information for storing and managing collected data.

[0114] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. Specifically, it collects user ratings and transaction histories, analyzes this data, and generates appropriate word-of-mouth information to provide to users.

[0115] Hardware and software used

[0116] 1. Server:

[0117] Hardware used: database servers, storage devices, processors (CPUs), memory, dedicated servers for clustering, and network equipment.

[0118] Software used: MySQL or PostgreSQL (database management systems), Python (programming language), NLTK and SpaCy (natural language processing toolkits), Scikit-learn (machine learning library), Django (web framework).

[0119] 2. Terminal:

[0120] The hardware used: the user's PC, smartphone, or tablet.

[0121] Software used: Web browser, mobile application (interface for displaying reviews).

[0122] 3. User:

[0123] Hardware used: PC, smartphone, tablet.

[0124] Software used: Web browser, mobile application.

[0125] Specific examples of processing

[0126] For example, if a user wants to see reviews of new anime on a shopping site, the following steps would occur:

[0127] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0128] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features, assigning a high weight to keywords such as "moving story."

[0129] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0130] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0131] Prompt Sentence Examples

[0132] "User A has posted a review of a new manga. Please explain in detail the process for analyzing that review, classifying it into existing clusters, and selecting the reviews with the most similarities."

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

[0134] Step 1:

[0135] Users post reviews.

[0136] Input: Review text data

[0137] Output: The submitted review is sent to the server.

[0138] The user enters a comment in the review input field and clicks the post button.

[0139] Example: You write a review saying, "This product is very easy to use."

[0140] Step 2:

[0141] The server collects reviews and purchase history and stores them in a database.

[0142] Input: submitted reviews, related purchase history

[0143] Output: Reviews and purchase history stored in a database

[0144] The server collects user-submitted reviews and purchase history in real time and stores them in a database.

[0145] Example: User A's reviews and associated purchase history are stored in a database.

[0146] Step 3:

[0147] The server removes unnecessary information from the collected reviews and generates clean text data.

[0148] Input: Reviews stored in the database

[0149] Output: Clean text data with unnecessary information removed

[0150] The server uses natural language processing techniques (NLTK, SpaCy, etc.) to remove unnecessary information from the text (e.g., HTML tags and special characters).

[0151] Example: Extracting the important keyword "easy to use" from reviews.

[0152] Step 4:

[0153] The server extracts features using clean text data.

[0154] Input: clean text data

[0155] Output: Features of text data (data expressed as numbers)

[0156] The server uses natural language processing and machine learning techniques (such as Scikit-learn) to extract features that represent the content of the text as numbers.

[0157] Example: The keyword "easy to use" is assigned a high value.

[0158] Step 5:

[0159] The server uses the KMeans algorithm to classify reviews into clusters based on their features.

[0160] Input: Features of text data

[0161] Output: Reviews sorted into clusters

[0162] The server runs the KMeans algorithm to classify reviews and comments into clusters based on their features.

[0163] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0164] Step 6:

[0165] When a user submits a new review, the server calculates its features and classifies it into an existing cluster.

[0166] Input: New review text data

[0167] Output: New reviews are classified into the appropriate cluster

[0168] A user enters a new review and clicks the submit button. The server calculates the features of the new review and classifies it into an existing cluster.

[0169] Example: User A's review "This new manga is great" will be classified into the appropriate cluster.

[0170] Step 7:

[0171] The server calculates the cosine similarity and selects the most similar reviews.

[0172] Input: Reviews in cluster, features of new reviews

[0173] Output: Most similar reviews

[0174] The server calculates the cosine similarity of the new review with other reviews in the cluster and selects the most similar review.

[0175] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0176] Step 8:

[0177] The server generates optimal word-of-mouth information and sends it to the device.

[0178] Input: Data of the most similar reviews

[0179] Output: Reviews displayed on the user's device

[0180] The server generates optimal word-of-mouth information based on the selected reviews and sends it to the user's device.

[0181] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0182] (Application example 1)

[0183] 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."

[0184] It is not easy for users to find products and content that suit their preferences on shopping sites and content distribution services. It is also difficult to quickly obtain useful word-of-mouth information from the vast amount of review information. Furthermore, existing systems for increasing user satisfaction have room for improvement, as they lack relevant review information when posting a new review. There is a need to provide a means to solve these problems and provide users with the most appropriate word-of-mouth information.

[0185] 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.

[0186] In this invention, the server includes: means for collecting user reviews and purchase histories; means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data; means for extracting features from the generated clean text data; means for classifying reviews and comments into clusters based on the extracted features; means for calculating features of newly posted reviews and classifying them into the closest cluster; means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information; means for displaying the generated word-of-mouth information to the user; and means for recommending optimal reviews generated from the user's review history and purchase history when the user posts a new review. This allows users to quickly find products and content that suit their preferences and provides a more satisfying purchasing experience. Furthermore, related review information is presented when posting a new review, which also contributes to improving the user's review posting experience.

[0187] "User" refers to a person who uses a shopping site or content distribution service.

[0188] A "review" is text information in which a user writes their own opinion or evaluation of a product or content.

[0189] "Purchase history" is data that includes a list of products that a user has purchased and information about those products.

[0190] "Clean text data" refers to data from which unnecessary information has been removed and only important information has been retained.

[0191] A "feature amount" is a numerical representation of the content of text data, and is a parameter that indicates the characteristics and trends of the data.

[0192] A "cluster" is a group of data that have similar characteristics.

[0193] "Word-of-mouth information" is information that provides optimal reviews and ratings based on user reviews and purchase history.

[0194] "Cosine similarity" is an index that calculates the similarity between two vectors, and is primarily used to evaluate the similarity of text data.

[0195] "TF-IDF vectorizer" is a method for quantifying important words from text data and calculating their importance.

[0196] "Recommendation" is the act of recommending the most suitable products or content based on a user's past actions and data.

[0197] The "best review" is the review that is determined to be the most relevant by evaluating the similarity between the reviews of users in the cluster and the newly posted review.

[0198] System Overview

[0199] This invention is a system that allows users to easily find products and content that suit their preferences. The system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users. Implementing the system requires the following steps:

[0200] System configuration

[0201] Hardware

[0202] This system uses hardware including a server and a user terminal (e.g., a smartphone). The server is used to collect, process, generate, and distribute data, and the user terminal displays the generated word-of-mouth information to the user.

[0203] software

[0204] The following software is used to realize this system.

[0205] requests library: Connects to external APIs.

[0206] scikit-learn library: Natural language processing, clustering, and similarity calculations.

[0207] TfidfVectorizer: Quantifies important words from text data and calculates their importance.

[0208] Processing flow

[0209] 1. Data Collection

[0210] The server collects user reviews and purchase history, for example, using an API to retrieve review history and purchase history based on user ID.

[0211] 2. Data cleansing

[0212] Unnecessary information is removed from the collected data to generate clean text data, using natural language processing techniques such as TfidfVectorizer.

[0213] 3. Feature Extraction and Clustering

[0214] Features of the clean text data are extracted using TfidfVectorizer, and reviews and comments are classified into clusters based on the extracted features using the KMeans clustering algorithm.

[0215] 4. Similarity Assessment

[0216] When a new review is submitted, the features of the review are calculated and it is classified into an existing cluster.The similarity of the new review to other reviews in the cluster is then evaluated using cosine similarity.

[0217] 5. Generating and displaying reviews

[0218] The system generates optimal word-of-mouth information using the most similar reviews and displays it on the user's device. When a user submits a new review, the system recommends the most appropriate review based on the history information it has built.

[0219] Specific examples

[0220] For example, User A posts a review saying, "This new device is very easy to use!" When this review is entered into the system, the server provides the most similar review information from past reviews. This allows User A to obtain review information for other devices that suit his or her preferences.

[0221] Example prompts to input to a generative AI model:

[0222] Get the review and purchase history of user ID: 12345 from the API and output the existing review that is most similar to the new review, "This new device is so easy to use!"

[0223] In this way, the present invention provides a specific method for users to have a highly satisfying purchasing experience.

[0224] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0225] Step 1:

[0226] The server collects user reviews and purchase history. Specifically, the server uses an API to retrieve relevant review and purchase history data based on the user ID. The input is the user ID, and the output is the user's review and purchase history data. The collected data is used as input for the next processing step.

[0227] Step 2:

[0228] The server cleanses the collected reviews and purchase history data to generate clean text data. Specifically, the server uses natural language processing techniques to remove unnecessary information and identify key words and phrases. The input is the collected review and purchase history data, and the output is clean text data. This cleansed data is used in the next step of feature extraction.

[0229] Step 3:

[0230] The server extracts features from the cleansed text data. Specifically, the server quantifies the features of the text data using TfidfVectorizer, and then extracts important product features based on the quantified features. The input is the clean text data, and the output is a feature vector. The extracted features are used in the next clustering step.

[0231] Step 4:

[0232] The server classifies reviews and comments into clusters based on their features. Specifically, the server uses the KMeans clustering algorithm to group reviews with high similarity. The input is a feature vector, and the output is reviews classified into clusters. The classified reviews are used in the next similarity evaluation step.

[0233] Step 5:

[0234] The server calculates the features of newly posted reviews and classifies them into the closest cluster. When a new review is posted, the server extracts the features of the review using TfidfVectorizer and compares them with existing clusters to classify it into the most appropriate cluster. The input is the newly posted review, and the output is the review classified into the closest cluster.

[0235] Step 6:

[0236] The server evaluates the similarity of the newly submitted review to other reviews in the cluster. Specifically, the server calculates the similarity between each review using cosine similarity and determines the most similar review. The input is the newly submitted review and other reviews in the cluster, and the output is the most similar review.

[0237] Step 7:

[0238] The server generates optimal review information and displays it on the user's device. The review information generated based on the new review is delivered to the user's device and visualized. The input is the most similar review and related review information, and the output is the optimal review information displayed on the user's device.

[0239] Step 8:

[0240] When a user submits a new review, the server recommends the most appropriate review based on their review history and purchase history. The server presents the most appropriate review information to the user based on past data, improving the user's review submission experience. The input is the user's review history and purchase history, and the output is the recommended optimal review.

[0241] 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.

[0242] MODE FOR CARRYING OUT THE INVENTION

[0243] System Overview

[0244] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects user reviews and purchase histories, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[0245] System configuration

[0246] The system functions through collaboration between the server, terminals, and users as follows:

[0247] 1. The server collects user reviews and purchase history and stores them in a database. Specific information includes review content, rating score, purchase date and time, etc.

[0248] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[0249] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[0250] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0251] 3. The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of the reviews and comments.

[0252] Example: A positive emotion is perceived from the review "Easy to use."

[0253] 4. The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[0254] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[0255] 5. The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as the basis for clustering.

[0256] Example: Similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0257] 6. When a user posts a new review, the server calculates the review's features and sentiment and classifies it into an existing cluster.

[0258] Example: If user A posts a review saying "This new manga is awesome," the review will be classified into the appropriate cluster.

[0259] 7. The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity and taking into account sentiment information.

[0260] For example, the review is evaluated for its similarity to other positive reviews in the cluster, such as "highly enjoyable" or "rave reviews."

[0261] 8. The server generates optimal reviews and compiles them as reviews. The information is generated based on the positive emotions obtained by the emotion engine.

[0262] Example: Positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0263] 9. The server sends the generated review information to the user's device.

[0264] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0265] 10. Users use the displayed reviews to purchase or watch products or content.

[0266] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[0267] This system improves the reliability of reviews, allowing users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive sentiment.

[0268] The processing flow will be explained below.

[0269] Step 1:

[0270] The server collects user reviews and purchase history and stores them in a database. Collected information includes review content, rating score, purchase date and time, etc.

[0271] Example: When user A posts a review saying "This product is very easy to use," the review and purchase history are saved on the server.

[0272] Step 2:

[0273] The server uses natural language processing technology to remove unnecessary information (e.g., symbols and meaningless words) from the collected reviews and purchase history to generate clean text data.

[0274] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0275] Step 3:

[0276] The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of reviews and comments.

[0277] Example: A positive emotion is recognized from the review "Easy to use."

[0278] Step 4:

[0279] The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[0280] Example: The TF-IDF value of "easy to use" is represented as a numeric vector.

[0281] Step 5:

[0282] The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as a basis for clustering.

[0283] For example, highly similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0284] Step 6:

[0285] A user posts a new review.

[0286] Example: User A posts a review saying, "This new manga is great."

[0287] Step 7:

[0288] The server calculates the features and sentiment of new reviews, quantifies them using a TF-IDF vectorizer, and classifies them into existing clusters.

[0289] For example, a review that says "It's the best" will be classified into cluster 2, recognizing a positive sentiment.

[0290] Step 8:

[0291] The server evaluates the similarity of a new review to other reviews in the cluster using cosine similarity, taking sentiment information into account.

[0292] Example: The similarity with other positive reviews in Cluster 2, such as "A very enjoyable work" and "Highly praised," is evaluated.

[0293] Step 9:

[0294] The server generates optimal reviews and summarizes them based on the positive emotions obtained from the emotion engine.

[0295] For example, positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0296] Step 10:

[0297] The server transmits the generated review information to the user's terminal.

[0298] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0299] Step 11:

[0300] Users use the displayed reviews to purchase or watch products or content.

[0301] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[0302] This series of processes allows users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive emotions.

[0303] Example 2

[0304] 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."

[0305] The review functions on existing shopping sites and content distribution services have made it difficult for users to find products and content that match their preferences. Furthermore, there is a lack of technology that can properly recognize the sentiment behind reviews and comments and provide them as organized word-of-mouth information. This makes it time-consuming for users to select products, resulting in lower satisfaction.

[0306] 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 user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, means for extracting features from the generated clean text data, means for recognizing emotions from the clean text data, means for converting the features into numerical vectors, means for classifying reviews and comments into clusters based on the extracted features and emotions, means for calculating features and emotions of a newly posted review and classifying it into the closest cluster, means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information, and means for displaying the generated word-of-mouth information to the user. This allows users to more easily find products and content that suit their preferences.

[0307] "User" refers to the end user who uses the System to make reviews and purchases.

[0308] "Review" refers to an evaluation or comment made by a user on a product or content.

[0309] "Purchase history" refers to a record of products and content purchased by a user in the past.

[0310] "Server" refers to a computer system that receives input from users and processes, stores, and analyzes data.

[0311] "Database" refers to a system for systematically storing and managing collected data.

[0312] "Removing unnecessary information" refers to the process of removing noise and unnecessary elements from collected data to make it clean.

[0313] "Clean text data" refers to text data after unnecessary information has been removed.

[0314] "Features" refer to important elements or attributes extracted from text data.

[0315] "Emotion recognition" refers to the process of identifying the emotional tone of review and comment text.

[0316] A "numeric vector" refers to text data that has been digitized and expressed in vector format.

[0317] "Clustering" refers to the process of grouping reviews or comments that are highly similar.

[0318] A "cluster" refers to a group of similar data points.

[0319] "Evaluating similarity" refers to the process of measuring the similarity between reviews and comments and using it for classification and ranking.

[0320] "Best review information" refers to the review content that is judged to be most useful to users.

[0321] "Cosine similarity" refers to a method for calculating similarity based on the angle between two vectors.

[0322] "TF-IDF vectorizer" refers to an algorithm that calculates the importance of words in text and represents them as numerical vectors.

[0323] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects reviews and purchase histories posted by users, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[0324] System configuration

[0325] The system is configured as follows:

[0326] Hardware and Software Use

[0327] 1. Server:

[0328] Data collection and storage: We collect user reviews and purchase history and store them in a database (e.g., MySQL or PostgreSQL).

[0329] Data cleaning: Using an NLP library (e.g., SpaCy or NLTK), unnecessary information is removed from the collected reviews to generate clean text data.

[0330] Feature extraction and quantification: Features are extracted and converted into numerical vectors using a TF-IDF vectorizer.

[0331] Sentiment Recognition: Recognize the sentiment of reviews and comments using sentiment analysis libraries (e.g., VADER or TextBlob).

[0332] Clustering: Using the KMeans algorithm, reviews and comments are classified into clusters based on features and sentiment information.

[0333] Similarity evaluation: The similarity between the reviews in the cluster and the newly posted review is evaluated using cosine similarity.

[0334] Word-of-mouth information generation: Generate optimal word-of-mouth information based on the similarity and sentiment information of reviews within a cluster.

[0335] Information transmission: Send the most appropriate review information to the user's device.

[0336] 2. Terminal:

[0337] Display information: Display the review information sent from the server to the user.

[0338] 3. User:

[0339] Review submission: Visit a shopping site or content distribution service and enter a review of a product or content.

[0340] Information reference: The displayed reviews are used as a reference to purchase or watch a product or content.

[0341] Specific examples

[0342] For example, if a user posts a review stating, "This product is very easy to use," the review is collected by the server. The server extracts the key keyword "easy to use" from the collected reviews and recognizes positive sentiment through sentiment analysis. It then calculates the TF-IDF value of "easy to use" using a TF-IDF vectorizer and converts it into a numerical vector. Based on this information, the KMeans algorithm classifies the review into a cluster as a positive review. The new review is then added to the cluster, and its similarity with other reviews is evaluated using cosine similarity to generate optimal review information. Finally, this review information is sent to the user's device, where the user can use the displayed review information to purchase a new product.

[0343] Prompt Sentence Examples

[0344] Below are some specific examples of prompt sentences to input into the generative AI model.

[0345] Example prompt sentence:

[0346] "Please explain the system that recognizes the sentiment of reviews posted by users and generates optimal user reviews. Please provide details of what hardware or software the system uses and what data processing and calculations it performs. Please also include specific examples."

[0347] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0348] Step 1:

[0349] A user posts a review. The user accesses a shopping site or content distribution service, enters a review of the product or content, and presses the submit button. The input includes the user's text review and rating score, which are then sent to the system.

[0350] Step 2:

[0351] The server collects the posted reviews and stores them in a database. The input is the review and rating score submitted by the user, which is received and stored in a database (e.g., MySQL or PostgreSQL). Specifically, if user A posts a review such as "This product is very easy to use," it is stored on the server.

[0352] Step 3:

[0353] The server removes unnecessary information from the collected reviews and generates clean text data. The input is reviews stored in a database, which are then processed using a natural language processing library (e.g., SpaCy or NLTK) to remove HTML tags and special symbols. The output is clean text data such as "easy to use." Specifically, the system extracts the important keyword "easy to use" from User A's review.

[0354] Step 4:

[0355] The server extracts features from the clean text data. The input is the clean text data, which is then vectorized using techniques such as Bag-of-Words and TF-IDF. The output is a numerical vector indicating the importance of each word. Specifically, the server calculates the frequency of occurrence of the phrase "easy to use" and vectorizes it.

[0356] Step 5:

[0357] The server uses an emotion engine to recognize the sentiment of reviews and comments. The input is vectorized text data, which is then used to recognize the sentiment using a sentiment analysis library (e.g., VADER or TextBlob). The output is the emotional tone (positive, negative, neutral). Specifically, a positive sentiment is recognized from the review "Easy to use."

[0358] Step 6:

[0359] The server converts the features into a numerical vector. The input is clean text data with recognized emotions, which is converted into a numerical vector using a TF-IDF vectorizer. The output is a feature expressed as a numerical vector. Specifically, the TF-IDF value of "easy to use" is calculated and converted into a numerical vector.

[0360] Step 7:

[0361] The server uses the KMeans algorithm to classify reviews and comments into clusters. The input is a numerical vector of features containing emotional information, which are then classified using KMeans clustering. The output is the formation of clusters. Specifically, positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0362] Step 8:

[0363] A user posts a new review. The user writes a new review, such as "This new manga is great," and presses the submit button. The input includes the user's new text review and rating score, which are then sent to the system.

[0364] Step 9:

[0365] The server calculates the features and emotions of new reviews and classifies them into existing clusters. The input is the text of the new review and its rating score, which are extracted using the process described above to extract features and perform emotion recognition. After that, they are converted into numerical vectors and classified into existing clusters. The output is that the new review is added to the appropriate cluster. Specifically, the new review is classified into an existing cluster with the same emotion.

[0366] Step 10:

[0367] The server evaluates the similarity within the cluster and generates the optimal review information. The input is the features and emotional information added to the cluster, and the similarity of the reviews is evaluated using cosine similarity. The output is the optimal review information. Specifically, reviews of "Very enjoyable work" and "Highly praised" in the same cluster are evaluated as having high similarity and are combined into new review information.

[0368] Step 11:

[0369] The server sends the generated review information to the user's device. The input is the generated review information, which is sent to the user's device. The output is the review information being displayed on the user's device. Specifically, information such as "Very enjoyable work" and "Highly praised" is displayed on the user's smartphone.

[0370] Step 12:

[0371] Users use the displayed reviews to purchase or watch products or content. The input is the displayed reviews, and they use this information to purchase products or watch content. Specifically, a user sees a review that says "this is a very enjoyable work," purchases a new manga, and has a highly satisfying purchasing experience.

[0372] (Application example 2)

[0373] 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."

[0374] Conventional recommendation systems based on reviews and purchase history have difficulty presenting optimal review information for each user, and have been unable to sufficiently increase user satisfaction. Furthermore, they lack the ability to recognize the detailed sentiment of reviews, making it impossible to provide personalized recommendations based on the user's emotions. In particular, there has been a lack of methods to provide a more intuitive and engaging user experience by utilizing head-mounted displays (HMDs).

[0375] 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.

[0376] In this invention, the server includes means for collecting user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, and means for extracting features from the generated clean text data. This makes it possible to perform clustering based on the review features, classify newly posted reviews into appropriate clusters, and display a recommendation list on the head-mounted display that also takes into account the user's emotional data.

[0377] "User reviews and purchase history" refers to the impressions and ratings of products posted by users, as well as the history of products they have purchased.

[0378] "Means for removing unnecessary information and generating clean text data" refers to a processing device or algorithm that removes noise and unnecessary parts from the collected reviews and purchase history and formats it into an analyzable format.

[0379] "Means for extracting features from the generated clean text data" refers to technology for numerically expressing important words and phrases contained in the text data.

[0380] "Means for classifying reviews and comments into clusters" refers to an algorithm for grouping similar reviews and comments based on their features.

[0381] "Method of classifying into the closest cluster" refers to the method used to assign newly submitted reviews to existing clusters.

[0382] "Means for evaluating the similarity of reviews and generating optimal word-of-mouth information" refers to technology that analyzes the similarity in content and sentiment between reviews and generates review information that is useful to users.

[0383] "Means for displaying the generated review information to the user" refers to equipment or software that visually displays the generated review information on the user's device.

[0384] "Means for recommending content using emotional data" refers to a system that recommends content that matches a user's preferences based on the results of an analysis of the user's emotions.

[0385] "Means for generating a recommendation list to be displayed on a head-mounted display" refers to technology that generates and visually presents a list of recommended content to be displayed on a user's visual device.

[0386] This invention is a system that analyzes user reviews and purchase history to generate and display a personalized recommendation list. How this system is implemented will be explained below.

[0387] First, the server collects user reviews and purchase history, including data from shopping sites and content distribution services, such as review content, rating scores, and purchase dates and times.

[0388] The server then removes unnecessary information from the collected reviews and purchase history to generate clean text data. This process includes text normalization and noise reduction, and uses natural language processing techniques. Specifically, it applies a natural language processing library called Spacy.

[0389] The server then extracts features from the generated clean text data, which include important words and phrases and quantify them using a TF-IDF vectorizer.

[0390] The KMeans algorithm is applied to cluster reviews and comments based on the extracted features. A newly submitted review is also classified into the closest cluster based on its features. This classification evaluates the similarity of the newly submitted review to other reviews in the cluster.

[0391] The server uses cosine similarity to evaluate the similarity and also considers the sentiment of the reviews using an emotion engine, resulting in optimal reviews based on positive sentiment and generated in a way that is most appropriate for each user.

[0392] Furthermore, the server utilizes the user's review history and emotional data to recommend personalized content. This information is generated as a recommendation list displayed on a head-mounted display (HMD), which provides users with intuitive and engaging visual information.

[0393] As a concrete example, if users post reviews on a movie website and collect sentiment data, the following prompts can be fed into a generative AI model for sentiment analysis:

[0394] This movie was very moving and I want to watch it again.

[0395] Use TextBlob to perform sentiment analysis as follows:

[0396] python

[0397] textblob.TextBlob("This movie moved me deeply. I want to watch it again.").sentiment.polarity

[0398] This generates a list of new movie recommendations based on the user's emotional data and displays them to the user through the HMD, allowing users to easily find personalized content based on their preferences and emotions.

[0399] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0400] Step 1:

[0401] The server collects user reviews and purchase history from shopping sites and content distribution services. Specific data includes review content, rating scores, purchase dates and times, etc. This data is stored as raw data on the server. The input is the user behavior log, and the output is a dataset of the collected reviews and purchase history.

[0402] Step 2:

[0403] The server removes unnecessary information from the collected reviews and purchase history to generate clean text data. Specifically, it uses natural language processing technology (e.g., Spacy) to normalize and denoise the text. The input is raw data, and the output is clean text data with unnecessary information removed.

[0404] Step 3:

[0405] The server extracts features from the generated clean text data. It uses a TF-IDF vectorizer to quantify important words and phrases contained in the text data. The input is the clean text data, and the output is feature data in the form of a numerical vector.

[0406] Step 4:

[0407] The server classifies reviews and comments into clusters based on the extracted features. It uses the KMeans algorithm to group similar reviews and comments into clusters. The input is the feature data, and the output is a list of reviews and comments classified into clusters.

[0408] Step 5:

[0409] The server calculates the features of a newly submitted review and classifies it into the closest cluster. This process uses the TF-IDF vectorizer, as in the previous step. The input is the text data of the new review, and the output is the cluster classification result.

[0410] Step 6:

[0411] The server evaluates the similarity of a newly posted review with other reviews in the cluster. It uses cosine similarity and an emotion engine to analyze the similarity of content and emotions between reviews. The input is the feature data of reviews in the cluster and the feature data of the new review, and the output is the similarity evaluation result.

[0412] Step 7:

[0413] The server generates optimal word-of-mouth information. Based on the sentiment analysis results and similarity evaluation results, it generates word-of-mouth information that is most suitable for the user. The input is the similarity evaluation results and sentiment analysis data, and the output is the generated word-of-mouth information.

[0414] Step 8:

[0415] The server displays the generated word-of-mouth information to the user. Specifically, it generates and visually presents a recommendation list to be displayed on a head-mounted display (HMD). The input is the generated word-of-mouth information, and the output is the recommendation list displayed on the HMD.

[0416] Step 9:

[0417] Users can view the recommendation list through the HMD and select content they are interested in. This process allows users to easily find personalized content based on their preferences and emotions. The input is the recommendation list displayed on the HMD, and the output is the user's selected content.

[0418] 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.

[0419] 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.

[0420] 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.

[0421] [Second embodiment]

[0422] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0423] 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.

[0424] 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).

[0425] 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.

[0426] 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.

[0427] 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).

[0428] 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.

[0429] 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.

[0430] 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.

[0431] 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.

[0432] 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.

[0433] 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."

[0434] MODE FOR CARRYING OUT THE INVENTION

[0435] System Overview

[0436] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. This system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users.

[0437] System configuration

[0438] The system functions through collaboration between servers, terminals, and users as follows:

[0439] 1. The server collects user reviews and purchase history and stores them in a database.

[0440] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[0441] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[0442] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0443] 3. The server extracts features from the generated clean text data. The features represent the content of the text as numerical values.

[0444] Example: The keyword "easy to use" is assigned a high value.

[0445] 4. The server uses the KMeans algorithm to classify reviews and comments into clusters based on the extracted features.

[0446] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0447] 5. When a user posts a new review, the server calculates the review's features and classifies them into existing clusters.

[0448] Example: If user A posts "This new manga is awesome," the review will be classified into the appropriate cluster.

[0449] 6. The server calculates the cosine similarity between the new review and the reviews of other users in the cluster, and selects the most similar review.

[0450] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0451] 7. The server generates optimal review information and sends it to the device.

[0452] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0453] Example

[0454] For example, if a user wants to see reviews of new anime on a shopping site:

[0455] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0456] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features. Keywords such as "moving story" are assigned a high weight.

[0457] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0458] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0459] This system improves the reliability of reviews and allows users to have a more satisfying purchasing experience.

[0460] The processing flow will be explained below.

[0461] Step 1:

[0462] The server collects user reviews and purchase history and stores them in a database, including information such as review content, rating score, and purchase date and time.

[0463] Example: If user A reviews a product saying, "This product is very easy to use," the review and purchase history are saved on the server.

[0464] Step 2:

[0465] The server removes unnecessary information (e.g., symbols and meaningless words) from the collected data and generates clean text data.

[0466] Example: Remove unnecessary information and extract the important keyword "easy to use" as clean text data.

[0467] Step 3:

[0468] The server extracts features from the generated clean text data using a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer.

[0469] Example: The TF-IDF value of the keyword "easy to use" is calculated.

[0470] Step 4:

[0471] The server converts reviews and comments into numerical vectors based on the TF-IDF values.

[0472] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[0473] Step 5:

[0474] The server uses the KMeans algorithm to classify reviews and comments into clusters.

[0475] For example, reviews with high similarity such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0476] Step 6:

[0477] A user posts a new review.

[0478] Example: User A posts a review saying, "This new manga is great."

[0479] Step 7:

[0480] The server calculates the features of new reviews and classifies them into existing clusters.

[0481] Example: A review that says "It's the best" will be classified into cluster 2.

[0482] Step 8:

[0483] The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity.

[0484] Example: The similarity is evaluated by comparing with reviews in Cluster 2 such as "A very enjoyable work" and "Highly praised."

[0485] Step 9:

[0486] The server generates optimal reviews by selecting the most similar reviews within the cluster and summarizing them as reviews.

[0487] For example, reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0488] Step 10:

[0489] The server transmits the generated review information to the user's terminal.

[0490] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0491] Step 11:

[0492] Users use the displayed reviews to purchase or watch products or content.

[0493] Example: User A purchases a new manga based on reviews and has a highly satisfying purchasing experience.

[0494] This series of processes allows users to easily find products and content that suit their preferences based on reliable review information.

[0495] Example 1

[0496] 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."

[0497] Conventional shopping sites and content distribution services lack systems that can effectively analyze user-submitted reviews and purchase histories to provide appropriate word-of-mouth information tailored to users' preferences. As a result, users have difficulty finding products and content that match their preferences, and reviews are often unreliable, resulting in an unsatisfactory purchasing experience. A solution to this problem is needed.

[0498] 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.

[0499] In this invention, the server includes means for collecting user ratings and transaction histories, means for removing unnecessary information from the collected ratings and transaction histories to generate clean text data, and means for extracting features of the generated clean text data, thereby enabling users to easily find products and content that suit their preferences.

[0500] "User" refers to a person who uses the service or system to make reviews or purchases.

[0501] "Ratings" refer to opinions and impressions posted by users about products or content.

[0502] "Transaction history" refers to records of purchases, browsing, etc. made by a user using a service or system.

[0503] "Removing unnecessary information" refers to the process of removing unimportant information or noise from collected data.

[0504] "Clean text data" refers to data in a text format suitable for analysis after unnecessary information has been removed.

[0505] "Features" refer to important attributes or elements extracted from text data expressed as numerical values.

[0506] A "cluster" refers to a grouping of ratings or comments with similar characteristics.

[0507] The "KMeans algorithm" refers to a type of machine learning algorithm for classifying data into a specified number of clusters.

[0508] "Cosine similarity" refers to a mathematical index for calculating the similarity between two pieces of text data.

[0509] "Word-of-mouth information" refers to a collection of useful information about products and content provided to users.

[0510] "TF-IDF vectorizer" refers to a statistical method for calculating the importance of words in a text.

[0511] "Database" refers to a structured collection of information for storing and managing collected data.

[0512] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. Specifically, it collects user ratings and transaction histories, analyzes this data, and generates appropriate word-of-mouth information to provide to users.

[0513] Hardware and software used

[0514] 1. Server:

[0515] Hardware used: database servers, storage devices, processors (CPUs), memory, dedicated servers for clustering, and network equipment.

[0516] Software used: MySQL or PostgreSQL (database management systems), Python (programming language), NLTK and SpaCy (natural language processing toolkits), Scikit-learn (machine learning library), Django (web framework).

[0517] 2. Terminal:

[0518] The hardware used: the user's PC, smartphone, or tablet.

[0519] Software used: Web browser, mobile application (interface for displaying reviews).

[0520] 3. User:

[0521] Hardware used: PC, smartphone, tablet.

[0522] Software used: Web browser, mobile application.

[0523] Specific examples of processing

[0524] For example, if a user wants to see reviews of new anime on a shopping site, the following steps would occur:

[0525] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0526] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features, assigning a high weight to keywords such as "moving story."

[0527] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0528] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0529] Prompt Sentence Examples

[0530] "User A has posted a review of a new manga. Please explain in detail the process for analyzing that review, classifying it into existing clusters, and selecting the reviews with the most similarities."

[0531] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0532] Step 1:

[0533] Users post reviews.

[0534] Input: Review text data

[0535] Output: The submitted review is sent to the server.

[0536] The user enters a comment in the review input field and clicks the post button.

[0537] Example: You write a review saying, "This product is very easy to use."

[0538] Step 2:

[0539] The server collects reviews and purchase history and stores them in a database.

[0540] Input: submitted reviews, related purchase history

[0541] Output: Reviews and purchase history stored in a database

[0542] The server collects user-submitted reviews and purchase history in real time and stores them in a database.

[0543] Example: User A's reviews and associated purchase history are stored in a database.

[0544] Step 3:

[0545] The server removes unnecessary information from the collected reviews and generates clean text data.

[0546] Input: Reviews stored in the database

[0547] Output: Clean text data with unnecessary information removed

[0548] The server uses natural language processing techniques (NLTK, SpaCy, etc.) to remove unnecessary information from the text (e.g., HTML tags and special characters).

[0549] Example: Extracting the important keyword "easy to use" from reviews.

[0550] Step 4:

[0551] The server extracts features using clean text data.

[0552] Input: clean text data

[0553] Output: Features of text data (data expressed as numbers)

[0554] The server uses natural language processing and machine learning techniques (such as Scikit-learn) to extract features that represent the content of the text as numbers.

[0555] Example: The keyword "easy to use" is assigned a high value.

[0556] Step 5:

[0557] The server uses the KMeans algorithm to classify reviews into clusters based on their features.

[0558] Input: Features of text data

[0559] Output: Reviews sorted into clusters

[0560] The server runs the KMeans algorithm to classify reviews and comments into clusters based on their features.

[0561] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0562] Step 6:

[0563] When a user submits a new review, the server calculates its features and classifies it into an existing cluster.

[0564] Input: New review text data

[0565] Output: New reviews are classified into the appropriate cluster

[0566] A user enters a new review and clicks the submit button. The server calculates the features of the new review and classifies it into an existing cluster.

[0567] Example: User A's review "This new manga is great" will be classified into the appropriate cluster.

[0568] Step 7:

[0569] The server calculates the cosine similarity and selects the most similar reviews.

[0570] Input: Reviews in cluster, features of new reviews

[0571] Output: Most similar reviews

[0572] The server calculates the cosine similarity of the new review with other reviews in the cluster and selects the most similar review.

[0573] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0574] Step 8:

[0575] The server generates optimal word-of-mouth information and sends it to the device.

[0576] Input: Data of the most similar reviews

[0577] Output: Reviews displayed on the user's device

[0578] The server generates optimal word-of-mouth information based on the selected reviews and sends it to the user's device.

[0579] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0580] (Application example 1)

[0581] 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."

[0582] It is not easy for users to find products and content that suit their preferences on shopping sites and content distribution services. It is also difficult to quickly obtain useful word-of-mouth information from the vast amount of review information. Furthermore, existing systems for increasing user satisfaction have room for improvement, as they lack relevant review information when posting a new review. There is a need to provide a means to solve these problems and provide users with the most appropriate word-of-mouth information.

[0583] 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.

[0584] In this invention, the server includes: means for collecting user reviews and purchase histories; means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data; means for extracting features from the generated clean text data; means for classifying reviews and comments into clusters based on the extracted features; means for calculating features of newly posted reviews and classifying them into the closest cluster; means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information; means for displaying the generated word-of-mouth information to the user; and means for recommending optimal reviews generated from the user's review history and purchase history when the user posts a new review. This allows users to quickly find products and content that suit their preferences and provides a more satisfying purchasing experience. Furthermore, related review information is presented when posting a new review, which also contributes to improving the user's review posting experience.

[0585] "User" refers to a person who uses a shopping site or content distribution service.

[0586] A "review" is text information in which a user writes their own opinion or evaluation of a product or content.

[0587] "Purchase history" is data that includes a list of products that a user has purchased and information about those products.

[0588] "Clean text data" refers to data from which unnecessary information has been removed and only important information has been retained.

[0589] A "feature amount" is a numerical representation of the content of text data, and is a parameter that indicates the characteristics and trends of the data.

[0590] A "cluster" is a group of data that have similar characteristics.

[0591] "Word-of-mouth information" is information that provides optimal reviews and ratings based on user reviews and purchase history.

[0592] "Cosine similarity" is an index that calculates the similarity between two vectors, and is primarily used to evaluate the similarity of text data.

[0593] "TF-IDF vectorizer" is a method for quantifying important words from text data and calculating their importance.

[0594] "Recommendation" is the act of recommending the most suitable products or content based on a user's past actions and data.

[0595] The "best review" is the review that is determined to be the most relevant by evaluating the similarity between the reviews of users in the cluster and the newly posted review.

[0596] System Overview

[0597] This invention is a system that allows users to easily find products and content that suit their preferences. The system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users. Implementing the system requires the following steps:

[0598] System configuration

[0599] Hardware

[0600] This system uses hardware including a server and a user terminal (e.g., a smartphone). The server is used to collect, process, generate, and distribute data, and the user terminal displays the generated word-of-mouth information to the user.

[0601] software

[0602] The following software is used to realize this system.

[0603] requests library: Connects to external APIs.

[0604] scikit-learn library: Natural language processing, clustering, and similarity calculations.

[0605] TfidfVectorizer: Quantifies important words from text data and calculates their importance.

[0606] Processing flow

[0607] 1. Data Collection

[0608] The server collects user reviews and purchase history, for example, using an API to retrieve review history and purchase history based on user ID.

[0609] 2. Data cleansing

[0610] Unnecessary information is removed from the collected data to generate clean text data, using natural language processing techniques such as TfidfVectorizer.

[0611] 3. Feature Extraction and Clustering

[0612] Features of the clean text data are extracted using TfidfVectorizer, and reviews and comments are classified into clusters based on the extracted features using the KMeans clustering algorithm.

[0613] 4. Similarity Assessment

[0614] When a new review is submitted, the features of the review are calculated and it is classified into an existing cluster.The similarity of the new review to other reviews in the cluster is then evaluated using cosine similarity.

[0615] 5. Generating and displaying reviews

[0616] The system generates optimal word-of-mouth information using the most similar reviews and displays it on the user's device. When a user submits a new review, the system recommends the most appropriate review based on the history information it has built.

[0617] Specific examples

[0618] For example, User A posts a review saying, "This new device is very easy to use!" When this review is entered into the system, the server provides the most similar review information from past reviews. This allows User A to obtain review information for other devices that suit his or her preferences.

[0619] Example prompts to input to a generative AI model:

[0620] Get the review and purchase history of user ID: 12345 from the API and output the existing review that is most similar to the new review, "This new device is so easy to use!"

[0621] In this way, the present invention provides a specific method for users to have a highly satisfying purchasing experience.

[0622] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0623] Step 1:

[0624] The server collects user reviews and purchase history. Specifically, the server uses an API to retrieve relevant review and purchase history data based on the user ID. The input is the user ID, and the output is the user's review and purchase history data. The collected data is used as input for the next processing step.

[0625] Step 2:

[0626] The server cleanses the collected reviews and purchase history data to generate clean text data. Specifically, the server uses natural language processing techniques to remove unnecessary information and identify key words and phrases. The input is the collected review and purchase history data, and the output is clean text data. This cleansed data is used in the next step of feature extraction.

[0627] Step 3:

[0628] The server extracts features from the cleansed text data. Specifically, the server quantifies the features of the text data using TfidfVectorizer, and then extracts important product features based on the quantified features. The input is the clean text data, and the output is a feature vector. The extracted features are used in the next clustering step.

[0629] Step 4:

[0630] The server classifies reviews and comments into clusters based on their features. Specifically, the server uses the KMeans clustering algorithm to group reviews with high similarity. The input is a feature vector, and the output is reviews classified into clusters. The classified reviews are used in the next similarity evaluation step.

[0631] Step 5:

[0632] The server calculates the features of newly posted reviews and classifies them into the closest cluster. When a new review is posted, the server extracts the features of the review using TfidfVectorizer and compares them with existing clusters to classify it into the most appropriate cluster. The input is the newly posted review, and the output is the review classified into the closest cluster.

[0633] Step 6:

[0634] The server evaluates the similarity of the newly submitted review to other reviews in the cluster. Specifically, the server calculates the similarity between each review using cosine similarity and determines the most similar review. The input is the newly submitted review and other reviews in the cluster, and the output is the most similar review.

[0635] Step 7:

[0636] The server generates optimal review information and displays it on the user's device. The review information generated based on the new review is delivered to the user's device and visualized. The input is the most similar review and related review information, and the output is the optimal review information displayed on the user's device.

[0637] Step 8:

[0638] When a user submits a new review, the server recommends the most appropriate review based on their review history and purchase history. The server presents the most appropriate review information to the user based on past data, improving the user's review submission experience. The input is the user's review history and purchase history, and the output is the recommended optimal review.

[0639] 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.

[0640] MODE FOR CARRYING OUT THE INVENTION

[0641] System Overview

[0642] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects user reviews and purchase histories, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[0643] System configuration

[0644] The system functions through collaboration between the server, terminals, and users as follows:

[0645] 1. The server collects user reviews and purchase history and stores them in a database. Specific information includes review content, rating score, purchase date and time, etc.

[0646] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[0647] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[0648] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0649] 3. The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of the reviews and comments.

[0650] Example: A positive emotion is perceived from the review "Easy to use."

[0651] 4. The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[0652] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[0653] 5. The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as the basis for clustering.

[0654] Example: Similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0655] 6. When a user posts a new review, the server calculates the review's features and sentiment and classifies it into an existing cluster.

[0656] Example: If user A posts a review saying "This new manga is awesome," the review will be classified into the appropriate cluster.

[0657] 7. The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity and taking into account sentiment information.

[0658] For example, the review is evaluated for its similarity to other positive reviews in the cluster, such as "highly enjoyable" or "rave reviews."

[0659] 8. The server generates optimal reviews and compiles them as reviews. The information is generated based on the positive emotions obtained by the emotion engine.

[0660] Example: Positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0661] 9. The server sends the generated review information to the user's device.

[0662] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0663] 10. Users use the displayed reviews to purchase or watch products or content.

[0664] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[0665] This system improves the reliability of reviews, allowing users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive sentiment.

[0666] The processing flow will be explained below.

[0667] Step 1:

[0668] The server collects user reviews and purchase history and stores them in a database. Collected information includes review content, rating score, purchase date and time, etc.

[0669] Example: When user A posts a review saying "This product is very easy to use," the review and purchase history are saved on the server.

[0670] Step 2:

[0671] The server uses natural language processing technology to remove unnecessary information (e.g., symbols and meaningless words) from the collected reviews and purchase history to generate clean text data.

[0672] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0673] Step 3:

[0674] The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of reviews and comments.

[0675] Example: A positive emotion is recognized from the review "Easy to use."

[0676] Step 4:

[0677] The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[0678] Example: The TF-IDF value of "easy to use" is represented as a numeric vector.

[0679] Step 5:

[0680] The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as a basis for clustering.

[0681] For example, highly similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0682] Step 6:

[0683] A user posts a new review.

[0684] Example: User A posts a review saying, "This new manga is great."

[0685] Step 7:

[0686] The server calculates the features and sentiment of new reviews, quantifies them using a TF-IDF vectorizer, and classifies them into existing clusters.

[0687] For example, a review that says "It's the best" will be classified into cluster 2, recognizing a positive sentiment.

[0688] Step 8:

[0689] The server evaluates the similarity of a new review to other reviews in the cluster using cosine similarity, taking sentiment information into account.

[0690] Example: The similarity with other positive reviews in Cluster 2, such as "A very enjoyable work" and "Highly praised," is evaluated.

[0691] Step 9:

[0692] The server generates optimal reviews and summarizes them based on the positive emotions obtained from the emotion engine.

[0693] For example, positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0694] Step 10:

[0695] The server transmits the generated review information to the user's terminal.

[0696] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0697] Step 11:

[0698] Users use the displayed reviews to purchase or watch products or content.

[0699] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[0700] This series of processes allows users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive emotions.

[0701] Example 2

[0702] 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."

[0703] The review functions on existing shopping sites and content distribution services have made it difficult for users to find products and content that match their preferences. Furthermore, there is a lack of technology that can properly recognize the sentiment behind reviews and comments and provide them as organized word-of-mouth information. This makes it time-consuming for users to select products, resulting in lower satisfaction.

[0704] 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 user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, means for extracting features from the generated clean text data, means for recognizing emotions from the clean text data, means for converting the features into numerical vectors, means for classifying reviews and comments into clusters based on the extracted features and emotions, means for calculating features and emotions of a newly posted review and classifying it into the closest cluster, means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information, and means for displaying the generated word-of-mouth information to the user. This allows users to more easily find products and content that suit their preferences.

[0705] "User" refers to the end user who uses the System to make reviews and purchases.

[0706] "Review" refers to an evaluation or comment made by a user on a product or content.

[0707] "Purchase history" refers to a record of products and content purchased by a user in the past.

[0708] "Server" refers to a computer system that receives input from users and processes, stores, and analyzes data.

[0709] "Database" refers to a system for systematically storing and managing collected data.

[0710] "Removing unnecessary information" refers to the process of removing noise and unnecessary elements from collected data to make it clean.

[0711] "Clean text data" refers to text data after unnecessary information has been removed.

[0712] "Features" refer to important elements or attributes extracted from text data.

[0713] "Emotion recognition" refers to the process of identifying the emotional tone of review and comment text.

[0714] A "numeric vector" refers to text data that has been digitized and expressed in vector format.

[0715] "Clustering" refers to the process of grouping reviews or comments that are highly similar.

[0716] A "cluster" refers to a group of similar data points.

[0717] "Evaluating similarity" refers to the process of measuring the similarity between reviews and comments and using it for classification and ranking.

[0718] "Best review information" refers to the review content that is judged to be most useful to users.

[0719] "Cosine similarity" refers to a method for calculating similarity based on the angle between two vectors.

[0720] "TF-IDF vectorizer" refers to an algorithm that calculates the importance of words in text and represents them as numerical vectors.

[0721] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects reviews and purchase histories posted by users, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[0722] System configuration

[0723] The system is configured as follows:

[0724] Hardware and Software Use

[0725] 1. Server:

[0726] Data collection and storage: We collect user reviews and purchase history and store them in a database (e.g., MySQL or PostgreSQL).

[0727] Data cleaning: Using an NLP library (e.g., SpaCy or NLTK), unnecessary information is removed from the collected reviews to generate clean text data.

[0728] Feature extraction and quantification: Features are extracted and converted into numerical vectors using a TF-IDF vectorizer.

[0729] Sentiment Recognition: Recognize the sentiment of reviews and comments using sentiment analysis libraries (e.g., VADER or TextBlob).

[0730] Clustering: Using the KMeans algorithm, reviews and comments are classified into clusters based on features and sentiment information.

[0731] Similarity evaluation: The similarity between the reviews in the cluster and the newly posted review is evaluated using cosine similarity.

[0732] Word-of-mouth information generation: Generate optimal word-of-mouth information based on the similarity and sentiment information of reviews within a cluster.

[0733] Information transmission: Send the most appropriate review information to the user's device.

[0734] 2. Terminal:

[0735] Display information: Display the review information sent from the server to the user.

[0736] 3. User:

[0737] Review submission: Visit a shopping site or content distribution service and enter a review of a product or content.

[0738] Information reference: The displayed reviews are used as a reference to purchase or watch a product or content.

[0739] Specific examples

[0740] For example, if a user posts a review stating, "This product is very easy to use," the review is collected by the server. The server extracts the key keyword "easy to use" from the collected reviews and recognizes positive sentiment through sentiment analysis. It then calculates the TF-IDF value of "easy to use" using a TF-IDF vectorizer and converts it into a numerical vector. Based on this information, the KMeans algorithm classifies the review into a cluster as a positive review. The new review is then added to the cluster, and its similarity with other reviews is evaluated using cosine similarity to generate optimal review information. Finally, this review information is sent to the user's device, where the user can use the displayed review information to purchase a new product.

[0741] Prompt Sentence Examples

[0742] Below are some specific examples of prompt sentences to input into the generative AI model.

[0743] Example prompt sentence:

[0744] "Please explain the system that recognizes the sentiment of reviews posted by users and generates optimal user reviews. Please provide details of what hardware or software the system uses and what data processing and calculations it performs. Please also include specific examples."

[0745] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0746] Step 1:

[0747] A user posts a review. The user accesses a shopping site or content distribution service, enters a review of the product or content, and presses the submit button. The input includes the user's text review and rating score, which are then sent to the system.

[0748] Step 2:

[0749] The server collects the posted reviews and stores them in a database. The input is the review and rating score submitted by the user, which is received and stored in a database (e.g., MySQL or PostgreSQL). Specifically, if user A posts a review such as "This product is very easy to use," it is stored on the server.

[0750] Step 3:

[0751] The server removes unnecessary information from the collected reviews and generates clean text data. The input is reviews stored in a database, which are then processed using a natural language processing library (e.g., SpaCy or NLTK) to remove HTML tags and special symbols. The output is clean text data such as "easy to use." Specifically, the system extracts the important keyword "easy to use" from User A's review.

[0752] Step 4:

[0753] The server extracts features from the clean text data. The input is the clean text data, which is then vectorized using techniques such as Bag-of-Words and TF-IDF. The output is a numerical vector indicating the importance of each word. Specifically, the server calculates the frequency of occurrence of the phrase "easy to use" and vectorizes it.

[0754] Step 5:

[0755] The server uses an emotion engine to recognize the sentiment of reviews and comments. The input is vectorized text data, which is then used to recognize the sentiment using a sentiment analysis library (e.g., VADER or TextBlob). The output is the emotional tone (positive, negative, neutral). Specifically, a positive sentiment is recognized from the review "Easy to use."

[0756] Step 6:

[0757] The server converts the features into a numerical vector. The input is clean text data with recognized emotions, which is converted into a numerical vector using a TF-IDF vectorizer. The output is a feature expressed as a numerical vector. Specifically, the TF-IDF value of "easy to use" is calculated and converted into a numerical vector.

[0758] Step 7:

[0759] The server uses the KMeans algorithm to classify reviews and comments into clusters. The input is a numerical vector of features containing emotional information, which are then classified using KMeans clustering. The output is the formation of clusters. Specifically, positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0760] Step 8:

[0761] A user posts a new review. The user writes a new review, such as "This new manga is great," and presses the submit button. The input includes the user's new text review and rating score, which are then sent to the system.

[0762] Step 9:

[0763] The server calculates the features and emotions of new reviews and classifies them into existing clusters. The input is the text of the new review and its rating score, which are extracted using the process described above to extract features and perform emotion recognition. After that, they are converted into numerical vectors and classified into existing clusters. The output is that the new review is added to the appropriate cluster. Specifically, the new review is classified into an existing cluster with the same emotion.

[0764] Step 10:

[0765] The server evaluates the similarity within the cluster and generates the optimal review information. The input is the features and emotional information added to the cluster, and the similarity of the reviews is evaluated using cosine similarity. The output is the optimal review information. Specifically, reviews of "Very enjoyable work" and "Highly praised" in the same cluster are evaluated as having high similarity and are combined into new review information.

[0766] Step 11:

[0767] The server sends the generated review information to the user's device. The input is the generated review information, which is sent to the user's device. The output is the review information being displayed on the user's device. Specifically, information such as "Very enjoyable work" and "Highly praised" is displayed on the user's smartphone.

[0768] Step 12:

[0769] Users use the displayed reviews to purchase or watch products or content. The input is the displayed reviews, and they use this information to purchase products or watch content. Specifically, a user sees a review that says "this is a very enjoyable work," purchases a new manga, and has a highly satisfying purchasing experience.

[0770] (Application example 2)

[0771] 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."

[0772] Conventional recommendation systems based on reviews and purchase history have difficulty presenting optimal review information for each user, and have been unable to sufficiently increase user satisfaction. Furthermore, they lack the ability to recognize the detailed sentiment of reviews, making it impossible to provide personalized recommendations based on the user's emotions. In particular, there has been a lack of methods to provide a more intuitive and engaging user experience by utilizing head-mounted displays (HMDs).

[0773] 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.

[0774] In this invention, the server includes means for collecting user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, and means for extracting features from the generated clean text data. This makes it possible to perform clustering based on the review features, classify newly posted reviews into appropriate clusters, and display a recommendation list on the head-mounted display that also takes into account the user's emotional data.

[0775] "User reviews and purchase history" refers to the impressions and ratings of products posted by users, as well as the history of products they have purchased.

[0776] "Means for removing unnecessary information and generating clean text data" refers to a processing device or algorithm that removes noise and unnecessary parts from the collected reviews and purchase history and formats it into an analyzable format.

[0777] "Means for extracting features from the generated clean text data" refers to technology for numerically expressing important words and phrases contained in the text data.

[0778] "Means for classifying reviews and comments into clusters" refers to an algorithm for grouping similar reviews and comments based on their features.

[0779] "Method of classifying into the closest cluster" refers to the method used to assign newly submitted reviews to existing clusters.

[0780] "Means for evaluating the similarity of reviews and generating optimal word-of-mouth information" refers to technology that analyzes the similarity in content and sentiment between reviews and generates review information that is useful to users.

[0781] "Means for displaying the generated review information to the user" refers to equipment or software that visually displays the generated review information on the user's device.

[0782] "Means for recommending content using emotional data" refers to a system that recommends content that matches a user's preferences based on the results of an analysis of the user's emotions.

[0783] "Means for generating a recommendation list to be displayed on a head-mounted display" refers to technology that generates and visually presents a list of recommended content to be displayed on a user's visual device.

[0784] This invention is a system that analyzes user reviews and purchase history to generate and display a personalized recommendation list. How this system is implemented will be explained below.

[0785] First, the server collects user reviews and purchase history, including data from shopping sites and content distribution services, such as review content, rating scores, and purchase dates and times.

[0786] The server then removes unnecessary information from the collected reviews and purchase history to generate clean text data. This process includes text normalization and noise reduction, and uses natural language processing techniques. Specifically, it applies a natural language processing library called Spacy.

[0787] The server then extracts features from the generated clean text data, which include important words and phrases and quantify them using a TF-IDF vectorizer.

[0788] The KMeans algorithm is applied to cluster reviews and comments based on the extracted features. A newly submitted review is also classified into the closest cluster based on its features. This classification evaluates the similarity of the newly submitted review to other reviews in the cluster.

[0789] The server uses cosine similarity to evaluate the similarity and also considers the sentiment of the reviews using an emotion engine, resulting in optimal reviews based on positive sentiment and generated in a way that is most appropriate for each user.

[0790] Furthermore, the server utilizes the user's review history and emotional data to recommend personalized content. This information is generated as a recommendation list displayed on a head-mounted display (HMD), which provides users with intuitive and engaging visual information.

[0791] As a concrete example, if users post reviews on a movie website and collect sentiment data, the following prompts can be fed into a generative AI model for sentiment analysis:

[0792] This movie was very moving and I want to watch it again.

[0793] Use TextBlob to perform sentiment analysis as follows:

[0794] python

[0795] textblob.TextBlob("This movie moved me deeply. I want to watch it again.").sentiment.polarity

[0796] This generates a list of new movie recommendations based on the user's emotional data and displays them to the user through the HMD, allowing users to easily find personalized content based on their preferences and emotions.

[0797] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0798] Step 1:

[0799] The server collects user reviews and purchase history from shopping sites and content distribution services. Specific data includes review content, rating scores, purchase dates and times, etc. This data is stored as raw data on the server. The input is the user behavior log, and the output is a dataset of the collected reviews and purchase history.

[0800] Step 2:

[0801] The server removes unnecessary information from the collected reviews and purchase history to generate clean text data. Specifically, it uses natural language processing technology (e.g., Spacy) to normalize and denoise the text. The input is raw data, and the output is clean text data with unnecessary information removed.

[0802] Step 3:

[0803] The server extracts features from the generated clean text data. It uses a TF-IDF vectorizer to quantify important words and phrases contained in the text data. The input is the clean text data, and the output is feature data in the form of a numerical vector.

[0804] Step 4:

[0805] The server classifies reviews and comments into clusters based on the extracted features. It uses the KMeans algorithm to group similar reviews and comments into clusters. The input is the feature data, and the output is a list of reviews and comments classified into clusters.

[0806] Step 5:

[0807] The server calculates the features of a newly submitted review and classifies it into the closest cluster. This process uses the TF-IDF vectorizer, as in the previous step. The input is the text data of the new review, and the output is the cluster classification result.

[0808] Step 6:

[0809] The server evaluates the similarity of a newly posted review with other reviews in the cluster. It uses cosine similarity and an emotion engine to analyze the similarity of content and emotions between reviews. The input is the feature data of reviews in the cluster and the feature data of the new review, and the output is the similarity evaluation result.

[0810] Step 7:

[0811] The server generates optimal word-of-mouth information. Based on the sentiment analysis results and similarity evaluation results, it generates word-of-mouth information that is most suitable for the user. The input is the similarity evaluation results and sentiment analysis data, and the output is the generated word-of-mouth information.

[0812] Step 8:

[0813] The server displays the generated word-of-mouth information to the user. Specifically, it generates and visually presents a recommendation list to be displayed on a head-mounted display (HMD). The input is the generated word-of-mouth information, and the output is the recommendation list displayed on the HMD.

[0814] Step 9:

[0815] Users can view the recommendation list through the HMD and select content they are interested in. This process allows users to easily find personalized content based on their preferences and emotions. The input is the recommendation list displayed on the HMD, and the output is the user's selected content.

[0816] 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.

[0817] 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.

[0818] 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.

[0819] [Third embodiment]

[0820] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0821] 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.

[0822] 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).

[0823] 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.

[0824] 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.

[0825] 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).

[0826] 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.

[0827] 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.

[0828] 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.

[0829] 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.

[0830] 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.

[0831] 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."

[0832] MODE FOR CARRYING OUT THE INVENTION

[0833] System Overview

[0834] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. This system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users.

[0835] System configuration

[0836] The system functions through collaboration between servers, terminals, and users as follows:

[0837] 1. The server collects user reviews and purchase history and stores them in a database.

[0838] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[0839] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[0840] Example: Important keywords such as "easy to use" are extracted from User A's review.

[0841] 3. The server extracts features from the generated clean text data. The features represent the content of the text as numerical values.

[0842] Example: The keyword "easy to use" is assigned a high value.

[0843] 4. The server uses the KMeans algorithm to classify reviews and comments into clusters based on the extracted features.

[0844] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0845] 5. When a user posts a new review, the server calculates the review's features and classifies them into existing clusters.

[0846] Example: If user A posts "This new manga is awesome," the review will be classified into the appropriate cluster.

[0847] 6. The server calculates the cosine similarity between the new review and the reviews of other users in the cluster, and selects the most similar review.

[0848] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0849] 7. The server generates optimal review information and sends it to the device.

[0850] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0851] Example

[0852] For example, if a user wants to see reviews of new anime on a shopping site:

[0853] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0854] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features. Keywords such as "moving story" are assigned a high weight.

[0855] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0856] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0857] This system improves the reliability of reviews and allows users to have a more satisfying purchasing experience.

[0858] The processing flow will be explained below.

[0859] Step 1:

[0860] The server collects user reviews and purchase history and stores them in a database, including information such as review content, rating score, and purchase date and time.

[0861] Example: If user A reviews a product saying, "This product is very easy to use," the review and purchase history are saved on the server.

[0862] Step 2:

[0863] The server removes unnecessary information (e.g., symbols and meaningless words) from the collected data and generates clean text data.

[0864] Example: Remove unnecessary information and extract the important keyword "easy to use" as clean text data.

[0865] Step 3:

[0866] The server extracts features from the generated clean text data using a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer.

[0867] Example: The TF-IDF value of the keyword "easy to use" is calculated.

[0868] Step 4:

[0869] The server converts reviews and comments into numerical vectors based on the TF-IDF values.

[0870] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[0871] Step 5:

[0872] The server uses the KMeans algorithm to classify reviews and comments into clusters.

[0873] For example, reviews with high similarity such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0874] Step 6:

[0875] A user posts a new review.

[0876] Example: User A posts a review saying, "This new manga is great."

[0877] Step 7:

[0878] The server calculates the features of new reviews and classifies them into existing clusters.

[0879] Example: A review that says "It's the best" will be classified into cluster 2.

[0880] Step 8:

[0881] The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity.

[0882] Example: The similarity is evaluated by comparing with reviews in Cluster 2 such as "A very enjoyable work" and "Highly praised."

[0883] Step 9:

[0884] The server generates optimal reviews by selecting the most similar reviews within the cluster and summarizing them as reviews.

[0885] For example, reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[0886] Step 10:

[0887] The server transmits the generated review information to the user's terminal.

[0888] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[0889] Step 11:

[0890] Users use the displayed reviews to purchase or watch products or content.

[0891] Example: User A purchases a new manga based on reviews and has a highly satisfying purchasing experience.

[0892] This series of processes allows users to easily find products and content that suit their preferences based on reliable review information.

[0893] Example 1

[0894] 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."

[0895] Conventional shopping sites and content distribution services lack systems that can effectively analyze user-submitted reviews and purchase histories to provide appropriate word-of-mouth information tailored to users' preferences. As a result, users have difficulty finding products and content that match their preferences, and reviews are often unreliable, resulting in an unsatisfactory purchasing experience. A solution to this problem is needed.

[0896] 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.

[0897] In this invention, the server includes means for collecting user ratings and transaction histories, means for removing unnecessary information from the collected ratings and transaction histories to generate clean text data, and means for extracting features of the generated clean text data, thereby enabling users to easily find products and content that suit their preferences.

[0898] "User" refers to a person who uses the service or system to make reviews or purchases.

[0899] "Ratings" refer to opinions and impressions posted by users about products or content.

[0900] "Transaction history" refers to records of purchases, browsing, etc. made by a user using a service or system.

[0901] "Removing unnecessary information" refers to the process of removing unimportant information or noise from collected data.

[0902] "Clean text data" refers to data in a text format suitable for analysis after unnecessary information has been removed.

[0903] "Features" refer to important attributes or elements extracted from text data expressed as numerical values.

[0904] A "cluster" refers to a grouping of ratings or comments with similar characteristics.

[0905] The "KMeans algorithm" refers to a type of machine learning algorithm for classifying data into a specified number of clusters.

[0906] "Cosine similarity" refers to a mathematical index for calculating the similarity between two pieces of text data.

[0907] "Word-of-mouth information" refers to a collection of useful information about products and content provided to users.

[0908] "TF-IDF vectorizer" refers to a statistical method for calculating the importance of words in a text.

[0909] "Database" refers to a structured collection of information for storing and managing collected data.

[0910] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. Specifically, it collects user ratings and transaction histories, analyzes this data, and generates appropriate word-of-mouth information to provide to users.

[0911] Hardware and software used

[0912] 1. Server:

[0913] Hardware used: database servers, storage devices, processors (CPUs), memory, dedicated servers for clustering, and network equipment.

[0914] Software used: MySQL or PostgreSQL (database management systems), Python (programming language), NLTK and SpaCy (natural language processing toolkits), Scikit-learn (machine learning library), Django (web framework).

[0915] 2. Terminal:

[0916] The hardware used: the user's PC, smartphone, or tablet.

[0917] Software used: Web browser, mobile application (interface for displaying reviews).

[0918] 3. User:

[0919] Hardware used: PC, smartphone, tablet.

[0920] Software used: Web browser, mobile application.

[0921] Specific examples of processing

[0922] For example, if a user wants to see reviews of new anime on a shopping site, the following steps would occur:

[0923] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[0924] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features, assigning a high weight to keywords such as "moving story."

[0925] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[0926] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[0927] Prompt Sentence Examples

[0928] "User A has posted a review of a new manga. Please explain in detail the process for analyzing that review, classifying it into existing clusters, and selecting the reviews with the most similarities."

[0929] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0930] Step 1:

[0931] Users post reviews.

[0932] Input: Review text data

[0933] Output: The submitted review is sent to the server.

[0934] The user enters a comment in the review input field and clicks the post button.

[0935] Example: You write a review saying, "This product is very easy to use."

[0936] Step 2:

[0937] The server collects reviews and purchase history and stores them in a database.

[0938] Input: submitted reviews, related purchase history

[0939] Output: Reviews and purchase history stored in a database

[0940] The server collects user-submitted reviews and purchase history in real time and stores them in a database.

[0941] Example: User A's reviews and associated purchase history are stored in a database.

[0942] Step 3:

[0943] The server removes unnecessary information from the collected reviews and generates clean text data.

[0944] Input: Reviews stored in the database

[0945] Output: Clean text data with unnecessary information removed

[0946] The server uses natural language processing techniques (NLTK, SpaCy, etc.) to remove unnecessary information from the text (e.g., HTML tags and special characters).

[0947] Example: Extracting the important keyword "easy to use" from reviews.

[0948] Step 4:

[0949] The server extracts features using clean text data.

[0950] Input: clean text data

[0951] Output: Features of text data (data expressed as numbers)

[0952] The server uses natural language processing and machine learning techniques (such as Scikit-learn) to extract features that represent the content of the text as numbers.

[0953] Example: The keyword "easy to use" is assigned a high value.

[0954] Step 5:

[0955] The server uses the KMeans algorithm to classify reviews into clusters based on their features.

[0956] Input: Features of text data

[0957] Output: Reviews sorted into clusters

[0958] The server runs the KMeans algorithm to classify reviews and comments into clusters based on their features.

[0959] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[0960] Step 6:

[0961] When a user submits a new review, the server calculates its features and classifies it into an existing cluster.

[0962] Input: New review text data

[0963] Output: New reviews are classified into the appropriate cluster

[0964] A user enters a new review and clicks the submit button. The server calculates the features of the new review and classifies it into an existing cluster.

[0965] Example: User A's review "This new manga is great" will be classified into the appropriate cluster.

[0966] Step 7:

[0967] The server calculates the cosine similarity and selects the most similar reviews.

[0968] Input: Reviews in cluster, features of new reviews

[0969] Output: Most similar reviews

[0970] The server calculates the cosine similarity of the new review with other reviews in the cluster and selects the most similar review.

[0971] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[0972] Step 8:

[0973] The server generates optimal word-of-mouth information and sends it to the device.

[0974] Input: Data of the most similar reviews

[0975] Output: Reviews displayed on the user's device

[0976] The server generates optimal word-of-mouth information based on the selected reviews and sends it to the user's device.

[0977] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[0978] (Application example 1)

[0979] 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."

[0980] It is not easy for users to find products and content that suit their preferences on shopping sites and content distribution services. It is also difficult to quickly obtain useful word-of-mouth information from the vast amount of review information. Furthermore, existing systems for increasing user satisfaction have room for improvement, as they lack relevant review information when posting a new review. There is a need to provide a means to solve these problems and provide users with the most appropriate word-of-mouth information.

[0981] 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.

[0982] In this invention, the server includes: means for collecting user reviews and purchase histories; means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data; means for extracting features from the generated clean text data; means for classifying reviews and comments into clusters based on the extracted features; means for calculating features of newly posted reviews and classifying them into the closest cluster; means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information; means for displaying the generated word-of-mouth information to the user; and means for recommending optimal reviews generated from the user's review history and purchase history when the user posts a new review. This allows users to quickly find products and content that suit their preferences and provides a more satisfying purchasing experience. Furthermore, related review information is presented when posting a new review, which also contributes to improving the user's review posting experience.

[0983] "User" refers to a person who uses a shopping site or content distribution service.

[0984] A "review" is text information in which a user writes their own opinion or evaluation of a product or content.

[0985] "Purchase history" is data that includes a list of products that a user has purchased and information about those products.

[0986] "Clean text data" refers to data from which unnecessary information has been removed and only important information has been retained.

[0987] A "feature amount" is a numerical representation of the content of text data, and is a parameter that indicates the characteristics and trends of the data.

[0988] A "cluster" is a group of data that have similar characteristics.

[0989] "Word-of-mouth information" is information that provides optimal reviews and ratings based on user reviews and purchase history.

[0990] "Cosine similarity" is an index that calculates the similarity between two vectors, and is primarily used to evaluate the similarity of text data.

[0991] "TF-IDF vectorizer" is a method for quantifying important words from text data and calculating their importance.

[0992] "Recommendation" is the act of recommending the most suitable products or content based on a user's past actions and data.

[0993] The "best review" is the review that is determined to be the most relevant by evaluating the similarity between the reviews of users in the cluster and the newly posted review.

[0994] System Overview

[0995] This invention is a system that allows users to easily find products and content that suit their preferences. The system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users. Implementing the system requires the following steps:

[0996] System configuration

[0997] Hardware

[0998] This system uses hardware including a server and a user terminal (e.g., a smartphone). The server is used to collect, process, generate, and distribute data, and the user terminal displays the generated word-of-mouth information to the user.

[0999] software

[1000] The following software is used to realize this system.

[1001] requests library: Connects to external APIs.

[1002] scikit-learn library: Natural language processing, clustering, and similarity calculations.

[1003] TfidfVectorizer: Quantifies important words from text data and calculates their importance.

[1004] Processing flow

[1005] 1. Data Collection

[1006] The server collects user reviews and purchase history, for example, using an API to retrieve review history and purchase history based on user ID.

[1007] 2. Data cleansing

[1008] Unnecessary information is removed from the collected data to generate clean text data, using natural language processing techniques such as TfidfVectorizer.

[1009] 3. Feature Extraction and Clustering

[1010] Features of the clean text data are extracted using TfidfVectorizer, and reviews and comments are classified into clusters based on the extracted features using the KMeans clustering algorithm.

[1011] 4. Similarity Assessment

[1012] When a new review is submitted, the features of the review are calculated and it is classified into an existing cluster.The similarity of the new review to other reviews in the cluster is then evaluated using cosine similarity.

[1013] 5. Generating and displaying reviews

[1014] The system generates optimal word-of-mouth information using the most similar reviews and displays it on the user's device. When a user submits a new review, the system recommends the most appropriate review based on the history information it has built.

[1015] Specific examples

[1016] For example, User A posts a review saying, "This new device is very easy to use!" When this review is entered into the system, the server provides the most similar review information from past reviews. This allows User A to obtain review information for other devices that suit his or her preferences.

[1017] Example prompts to input to a generative AI model:

[1018] Get the review and purchase history of user ID: 12345 from the API and output the existing review that is most similar to the new review, "This new device is so easy to use!"

[1019] In this way, the present invention provides a specific method for users to have a highly satisfying purchasing experience.

[1020] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1021] Step 1:

[1022] The server collects user reviews and purchase history. Specifically, the server uses an API to retrieve relevant review and purchase history data based on the user ID. The input is the user ID, and the output is the user's review and purchase history data. The collected data is used as input for the next processing step.

[1023] Step 2:

[1024] The server cleanses the collected reviews and purchase history data to generate clean text data. Specifically, the server uses natural language processing techniques to remove unnecessary information and identify key words and phrases. The input is the collected review and purchase history data, and the output is clean text data. This cleansed data is used in the next step of feature extraction.

[1025] Step 3:

[1026] The server extracts features from the cleansed text data. Specifically, the server quantifies the features of the text data using TfidfVectorizer, and then extracts important product features based on the quantified features. The input is the clean text data, and the output is a feature vector. The extracted features are used in the next clustering step.

[1027] Step 4:

[1028] The server classifies reviews and comments into clusters based on their features. Specifically, the server uses the KMeans clustering algorithm to group reviews with high similarity. The input is a feature vector, and the output is reviews classified into clusters. The classified reviews are used in the next similarity evaluation step.

[1029] Step 5:

[1030] The server calculates the features of newly posted reviews and classifies them into the closest cluster. When a new review is posted, the server extracts the features of the review using TfidfVectorizer and compares them with existing clusters to classify it into the most appropriate cluster. The input is the newly posted review, and the output is the review classified into the closest cluster.

[1031] Step 6:

[1032] The server evaluates the similarity of the newly submitted review to other reviews in the cluster. Specifically, the server calculates the similarity between each review using cosine similarity and determines the most similar review. The input is the newly submitted review and other reviews in the cluster, and the output is the most similar review.

[1033] Step 7:

[1034] The server generates optimal review information and displays it on the user's device. The review information generated based on the new review is delivered to the user's device and visualized. The input is the most similar review and related review information, and the output is the optimal review information displayed on the user's device.

[1035] Step 8:

[1036] When a user submits a new review, the server recommends the most appropriate review based on their review history and purchase history. The server presents the most appropriate review information to the user based on past data, improving the user's review submission experience. The input is the user's review history and purchase history, and the output is the recommended optimal review.

[1037] 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.

[1038] MODE FOR CARRYING OUT THE INVENTION

[1039] System Overview

[1040] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects user reviews and purchase histories, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[1041] System configuration

[1042] The system functions through collaboration between the server, terminals, and users as follows:

[1043] 1. The server collects user reviews and purchase history and stores them in a database. Specific information includes review content, rating score, purchase date and time, etc.

[1044] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[1045] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[1046] Example: Important keywords such as "easy to use" are extracted from User A's review.

[1047] 3. The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of the reviews and comments.

[1048] Example: A positive emotion is perceived from the review "Easy to use."

[1049] 4. The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[1050] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[1051] 5. The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as the basis for clustering.

[1052] Example: Similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1053] 6. When a user posts a new review, the server calculates the review's features and sentiment and classifies it into an existing cluster.

[1054] Example: If user A posts a review saying "This new manga is awesome," the review will be classified into the appropriate cluster.

[1055] 7. The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity and taking into account sentiment information.

[1056] For example, the review is evaluated for its similarity to other positive reviews in the cluster, such as "highly enjoyable" or "rave reviews."

[1057] 8. The server generates optimal reviews and compiles them as reviews. The information is generated based on the positive emotions obtained by the emotion engine.

[1058] Example: Positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[1059] 9. The server sends the generated review information to the user's device.

[1060] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[1061] 10. Users use the displayed reviews to purchase or watch products or content.

[1062] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[1063] This system improves the reliability of reviews, allowing users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive sentiment.

[1064] The processing flow will be explained below.

[1065] Step 1:

[1066] The server collects user reviews and purchase history and stores them in a database. Collected information includes review content, rating score, purchase date and time, etc.

[1067] Example: When user A posts a review saying "This product is very easy to use," the review and purchase history are saved on the server.

[1068] Step 2:

[1069] The server uses natural language processing technology to remove unnecessary information (e.g., symbols and meaningless words) from the collected reviews and purchase history to generate clean text data.

[1070] Example: Important keywords such as "easy to use" are extracted from User A's review.

[1071] Step 3:

[1072] The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of reviews and comments.

[1073] Example: A positive emotion is recognized from the review "Easy to use."

[1074] Step 4:

[1075] The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[1076] Example: The TF-IDF value of "easy to use" is represented as a numeric vector.

[1077] Step 5:

[1078] The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as a basis for clustering.

[1079] For example, highly similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1080] Step 6:

[1081] A user posts a new review.

[1082] Example: User A posts a review saying, "This new manga is great."

[1083] Step 7:

[1084] The server calculates the features and sentiment of new reviews, quantifies them using a TF-IDF vectorizer, and classifies them into existing clusters.

[1085] For example, a review that says "It's the best" will be classified into cluster 2, recognizing a positive sentiment.

[1086] Step 8:

[1087] The server evaluates the similarity of a new review to other reviews in the cluster using cosine similarity, taking sentiment information into account.

[1088] Example: The similarity with other positive reviews in Cluster 2, such as "A very enjoyable work" and "Highly praised," is evaluated.

[1089] Step 9:

[1090] The server generates optimal reviews and summarizes them based on the positive emotions obtained from the emotion engine.

[1091] For example, positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[1092] Step 10:

[1093] The server transmits the generated review information to the user's terminal.

[1094] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[1095] Step 11:

[1096] Users use the displayed reviews to purchase or watch products or content.

[1097] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[1098] This series of processes allows users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive emotions.

[1099] Example 2

[1100] 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."

[1101] The review functions on existing shopping sites and content distribution services have made it difficult for users to find products and content that match their preferences. Furthermore, there is a lack of technology that can properly recognize the sentiment behind reviews and comments and provide them as organized word-of-mouth information. This makes it time-consuming for users to select products, resulting in lower satisfaction.

[1102] 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 user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, means for extracting features from the generated clean text data, means for recognizing emotions from the clean text data, means for converting the features into numerical vectors, means for classifying reviews and comments into clusters based on the extracted features and emotions, means for calculating features and emotions of a newly posted review and classifying it into the closest cluster, means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information, and means for displaying the generated word-of-mouth information to the user. This allows users to more easily find products and content that suit their preferences.

[1103] "User" refers to the end user who uses the System to make reviews and purchases.

[1104] "Review" refers to an evaluation or comment made by a user on a product or content.

[1105] "Purchase history" refers to a record of products and content purchased by a user in the past.

[1106] "Server" refers to a computer system that receives input from users and processes, stores, and analyzes data.

[1107] "Database" refers to a system for systematically storing and managing collected data.

[1108] "Removing unnecessary information" refers to the process of removing noise and unnecessary elements from collected data to make it clean.

[1109] "Clean text data" refers to text data after unnecessary information has been removed.

[1110] "Features" refer to important elements or attributes extracted from text data.

[1111] "Emotion recognition" refers to the process of identifying the emotional tone of review and comment text.

[1112] A "numeric vector" refers to text data that has been digitized and expressed in vector format.

[1113] "Clustering" refers to the process of grouping reviews or comments that are highly similar.

[1114] A "cluster" refers to a group of similar data points.

[1115] "Evaluating similarity" refers to the process of measuring the similarity between reviews and comments and using it for classification and ranking.

[1116] "Best review information" refers to the review content that is judged to be most useful to users.

[1117] "Cosine similarity" refers to a method for calculating similarity based on the angle between two vectors.

[1118] "TF-IDF vectorizer" refers to an algorithm that calculates the importance of words in text and represents them as numerical vectors.

[1119] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects reviews and purchase histories posted by users, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[1120] System configuration

[1121] The system is configured as follows:

[1122] Hardware and Software Use

[1123] 1. Server:

[1124] Data collection and storage: We collect user reviews and purchase history and store them in a database (e.g., MySQL or PostgreSQL).

[1125] Data cleaning: Using an NLP library (e.g., SpaCy or NLTK), unnecessary information is removed from the collected reviews to generate clean text data.

[1126] Feature extraction and quantification: Features are extracted and converted into numerical vectors using a TF-IDF vectorizer.

[1127] Sentiment Recognition: Recognize the sentiment of reviews and comments using sentiment analysis libraries (e.g., VADER or TextBlob).

[1128] Clustering: Using the KMeans algorithm, reviews and comments are classified into clusters based on features and sentiment information.

[1129] Similarity evaluation: The similarity between the reviews in the cluster and the newly posted review is evaluated using cosine similarity.

[1130] Word-of-mouth information generation: Generate optimal word-of-mouth information based on the similarity and sentiment information of reviews within a cluster.

[1131] Information transmission: Send the most appropriate review information to the user's device.

[1132] 2. Terminal:

[1133] Display information: Display the review information sent from the server to the user.

[1134] 3. User:

[1135] Review submission: Visit a shopping site or content distribution service and enter a review of a product or content.

[1136] Information reference: The displayed reviews are used as a reference to purchase or watch a product or content.

[1137] Specific examples

[1138] For example, if a user posts a review stating, "This product is very easy to use," the review is collected by the server. The server extracts the key keyword "easy to use" from the collected reviews and recognizes positive sentiment through sentiment analysis. It then calculates the TF-IDF value of "easy to use" using a TF-IDF vectorizer and converts it into a numerical vector. Based on this information, the KMeans algorithm classifies the review into a cluster as a positive review. The new review is then added to the cluster, and its similarity with other reviews is evaluated using cosine similarity to generate optimal review information. Finally, this review information is sent to the user's device, where the user can use the displayed review information to purchase a new product.

[1139] Prompt Sentence Examples

[1140] Below are some specific examples of prompt sentences to input into the generative AI model.

[1141] Example prompt sentence:

[1142] "Please explain the system that recognizes the sentiment of reviews posted by users and generates optimal user reviews. Please provide details of what hardware or software the system uses and what data processing and calculations it performs. Please also include specific examples."

[1143] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1144] Step 1:

[1145] A user posts a review. The user accesses a shopping site or content distribution service, enters a review of the product or content, and presses the submit button. The input includes the user's text review and rating score, which are then sent to the system.

[1146] Step 2:

[1147] The server collects the posted reviews and stores them in a database. The input is the review and rating score submitted by the user, which is received and stored in a database (e.g., MySQL or PostgreSQL). Specifically, if user A posts a review such as "This product is very easy to use," it is stored on the server.

[1148] Step 3:

[1149] The server removes unnecessary information from the collected reviews and generates clean text data. The input is reviews stored in a database, which are then processed using a natural language processing library (e.g., SpaCy or NLTK) to remove HTML tags and special symbols. The output is clean text data such as "easy to use." Specifically, the system extracts the important keyword "easy to use" from User A's review.

[1150] Step 4:

[1151] The server extracts features from the clean text data. The input is the clean text data, which is then vectorized using techniques such as Bag-of-Words and TF-IDF. The output is a numerical vector indicating the importance of each word. Specifically, the server calculates the frequency of occurrence of the phrase "easy to use" and vectorizes it.

[1152] Step 5:

[1153] The server uses an emotion engine to recognize the sentiment of reviews and comments. The input is vectorized text data, which is then used to recognize the sentiment using a sentiment analysis library (e.g., VADER or TextBlob). The output is the emotional tone (positive, negative, neutral). Specifically, a positive sentiment is recognized from the review "Easy to use."

[1154] Step 6:

[1155] The server converts the features into a numerical vector. The input is clean text data with recognized emotions, which is converted into a numerical vector using a TF-IDF vectorizer. The output is a feature expressed as a numerical vector. Specifically, the TF-IDF value of "easy to use" is calculated and converted into a numerical vector.

[1156] Step 7:

[1157] The server uses the KMeans algorithm to classify reviews and comments into clusters. The input is a numerical vector of features containing emotional information, which are then classified using KMeans clustering. The output is the formation of clusters. Specifically, positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1158] Step 8:

[1159] A user posts a new review. The user writes a new review, such as "This new manga is great," and presses the submit button. The input includes the user's new text review and rating score, which are then sent to the system.

[1160] Step 9:

[1161] The server calculates the features and emotions of new reviews and classifies them into existing clusters. The input is the text of the new review and its rating score, which are extracted using the process described above to extract features and perform emotion recognition. After that, they are converted into numerical vectors and classified into existing clusters. The output is that the new review is added to the appropriate cluster. Specifically, the new review is classified into an existing cluster with the same emotion.

[1162] Step 10:

[1163] The server evaluates the similarity within the cluster and generates the optimal review information. The input is the features and emotional information added to the cluster, and the similarity of the reviews is evaluated using cosine similarity. The output is the optimal review information. Specifically, reviews of "Very enjoyable work" and "Highly praised" in the same cluster are evaluated as having high similarity and are combined into new review information.

[1164] Step 11:

[1165] The server sends the generated review information to the user's device. The input is the generated review information, which is sent to the user's device. The output is the review information being displayed on the user's device. Specifically, information such as "Very enjoyable work" and "Highly praised" is displayed on the user's smartphone.

[1166] Step 12:

[1167] Users use the displayed reviews to purchase or watch products or content. The input is the displayed reviews, and they use this information to purchase products or watch content. Specifically, a user sees a review that says "this is a very enjoyable work," purchases a new manga, and has a highly satisfying purchasing experience.

[1168] (Application example 2)

[1169] 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."

[1170] Conventional recommendation systems based on reviews and purchase history have difficulty presenting optimal review information for each user, and have been unable to sufficiently increase user satisfaction. Furthermore, they lack the ability to recognize the detailed sentiment of reviews, making it impossible to provide personalized recommendations based on the user's emotions. In particular, there has been a lack of methods to provide a more intuitive and engaging user experience by utilizing head-mounted displays (HMDs).

[1171] 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.

[1172] In this invention, the server includes means for collecting user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, and means for extracting features from the generated clean text data. This makes it possible to perform clustering based on the review features, classify newly posted reviews into appropriate clusters, and display a recommendation list on the head-mounted display that also takes into account the user's emotional data.

[1173] "User reviews and purchase history" refers to the impressions and ratings of products posted by users, as well as the history of products they have purchased.

[1174] "Means for removing unnecessary information and generating clean text data" refers to a processing device or algorithm that removes noise and unnecessary parts from the collected reviews and purchase history and formats it into an analyzable format.

[1175] "Means for extracting features from the generated clean text data" refers to technology for numerically expressing important words and phrases contained in the text data.

[1176] "Means for classifying reviews and comments into clusters" refers to an algorithm for grouping similar reviews and comments based on their features.

[1177] "Method of classifying into the closest cluster" refers to the method used to assign newly submitted reviews to existing clusters.

[1178] "Means for evaluating the similarity of reviews and generating optimal word-of-mouth information" refers to technology that analyzes the similarity in content and sentiment between reviews and generates review information that is useful to users.

[1179] "Means for displaying the generated review information to the user" refers to equipment or software that visually displays the generated review information on the user's device.

[1180] "Means for recommending content using emotional data" refers to a system that recommends content that matches a user's preferences based on the results of an analysis of the user's emotions.

[1181] "Means for generating a recommendation list to be displayed on a head-mounted display" refers to technology that generates and visually presents a list of recommended content to be displayed on a user's visual device.

[1182] This invention is a system that analyzes user reviews and purchase history to generate and display a personalized recommendation list. How this system is implemented will be explained below.

[1183] First, the server collects user reviews and purchase history, including data from shopping sites and content distribution services, such as review content, rating scores, and purchase dates and times.

[1184] The server then removes unnecessary information from the collected reviews and purchase history to generate clean text data. This process includes text normalization and noise reduction, and uses natural language processing techniques. Specifically, it applies a natural language processing library called Spacy.

[1185] The server then extracts features from the generated clean text data, which include important words and phrases and quantify them using a TF-IDF vectorizer.

[1186] The KMeans algorithm is applied to cluster reviews and comments based on the extracted features. A newly submitted review is also classified into the closest cluster based on its features. This classification evaluates the similarity of the newly submitted review to other reviews in the cluster.

[1187] The server uses cosine similarity to evaluate the similarity and also considers the sentiment of the reviews using an emotion engine, resulting in optimal reviews based on positive sentiment and generated in a way that is most appropriate for each user.

[1188] Furthermore, the server utilizes the user's review history and emotional data to recommend personalized content. This information is generated as a recommendation list displayed on a head-mounted display (HMD), which provides users with intuitive and engaging visual information.

[1189] As a concrete example, if users post reviews on a movie website and collect sentiment data, the following prompts can be fed into a generative AI model for sentiment analysis:

[1190] This movie was very moving and I want to watch it again.

[1191] Use TextBlob to perform sentiment analysis as follows:

[1192] python

[1193] textblob.TextBlob("This movie moved me deeply. I want to watch it again.").sentiment.polarity

[1194] This generates a list of new movie recommendations based on the user's emotional data and displays them to the user through the HMD, allowing users to easily find personalized content based on their preferences and emotions.

[1195] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1196] Step 1:

[1197] The server collects user reviews and purchase history from shopping sites and content distribution services. Specific data includes review content, rating scores, purchase dates and times, etc. This data is stored as raw data on the server. The input is the user behavior log, and the output is a dataset of the collected reviews and purchase history.

[1198] Step 2:

[1199] The server removes unnecessary information from the collected reviews and purchase history to generate clean text data. Specifically, it uses natural language processing technology (e.g., Spacy) to normalize and denoise the text. The input is raw data, and the output is clean text data with unnecessary information removed.

[1200] Step 3:

[1201] The server extracts features from the generated clean text data. It uses a TF-IDF vectorizer to quantify important words and phrases contained in the text data. The input is the clean text data, and the output is feature data in the form of a numerical vector.

[1202] Step 4:

[1203] The server classifies reviews and comments into clusters based on the extracted features. It uses the KMeans algorithm to group similar reviews and comments into clusters. The input is the feature data, and the output is a list of reviews and comments classified into clusters.

[1204] Step 5:

[1205] The server calculates the features of a newly submitted review and classifies it into the closest cluster. This process uses the TF-IDF vectorizer, as in the previous step. The input is the text data of the new review, and the output is the cluster classification result.

[1206] Step 6:

[1207] The server evaluates the similarity of a newly posted review with other reviews in the cluster. It uses cosine similarity and an emotion engine to analyze the similarity of content and emotions between reviews. The input is the feature data of reviews in the cluster and the feature data of the new review, and the output is the similarity evaluation result.

[1208] Step 7:

[1209] The server generates optimal word-of-mouth information. Based on the sentiment analysis results and similarity evaluation results, it generates word-of-mouth information that is most suitable for the user. The input is the similarity evaluation results and sentiment analysis data, and the output is the generated word-of-mouth information.

[1210] Step 8:

[1211] The server displays the generated word-of-mouth information to the user. Specifically, it generates and visually presents a recommendation list to be displayed on a head-mounted display (HMD). The input is the generated word-of-mouth information, and the output is the recommendation list displayed on the HMD.

[1212] Step 9:

[1213] Users can view the recommendation list through the HMD and select content they are interested in. This process allows users to easily find personalized content based on their preferences and emotions. The input is the recommendation list displayed on the HMD, and the output is the user's selected content.

[1214] 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.

[1215] 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.

[1216] 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.

[1217] [Fourth embodiment]

[1218] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1219] 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.

[1220] 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).

[1221] 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.

[1222] 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.

[1223] 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).

[1224] 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.

[1225] 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.

[1226] 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.

[1227] 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.

[1228] 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.

[1229] 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.

[1230] 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."

[1231] MODE FOR CARRYING OUT THE INVENTION

[1232] System Overview

[1233] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. This system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users.

[1234] System configuration

[1235] The system functions through collaboration between servers, terminals, and users as follows:

[1236] 1. The server collects user reviews and purchase history and stores them in a database.

[1237] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[1238] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[1239] Example: Important keywords such as "easy to use" are extracted from User A's review.

[1240] 3. The server extracts features from the generated clean text data. The features represent the content of the text as numerical values.

[1241] Example: The keyword "easy to use" is assigned a high value.

[1242] 4. The server uses the KMeans algorithm to classify reviews and comments into clusters based on the extracted features.

[1243] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1244] 5. When a user posts a new review, the server calculates the review's features and classifies them into existing clusters.

[1245] Example: If user A posts "This new manga is awesome," the review will be classified into the appropriate cluster.

[1246] 6. The server calculates the cosine similarity between the new review and the reviews of other users in the cluster, and selects the most similar review.

[1247] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[1248] 7. The server generates optimal review information and sends it to the device.

[1249] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[1250] Example

[1251] For example, if a user wants to see reviews of new anime on a shopping site:

[1252] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[1253] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features. Keywords such as "moving story" are assigned a high weight.

[1254] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[1255] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[1256] This system improves the reliability of reviews and allows users to have a more satisfying purchasing experience.

[1257] The processing flow will be explained below.

[1258] Step 1:

[1259] The server collects user reviews and purchase history and stores them in a database, including information such as review content, rating score, and purchase date and time.

[1260] Example: If user A reviews a product saying, "This product is very easy to use," the review and purchase history are saved on the server.

[1261] Step 2:

[1262] The server removes unnecessary information (e.g., symbols and meaningless words) from the collected data and generates clean text data.

[1263] Example: Remove unnecessary information and extract the important keyword "easy to use" as clean text data.

[1264] Step 3:

[1265] The server extracts features from the generated clean text data using a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer.

[1266] Example: The TF-IDF value of the keyword "easy to use" is calculated.

[1267] Step 4:

[1268] The server converts reviews and comments into numerical vectors based on the TF-IDF values.

[1269] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[1270] Step 5:

[1271] The server uses the KMeans algorithm to classify reviews and comments into clusters.

[1272] For example, reviews with high similarity such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1273] Step 6:

[1274] A user posts a new review.

[1275] Example: User A posts a review saying, "This new manga is great."

[1276] Step 7:

[1277] The server calculates the features of new reviews and classifies them into existing clusters.

[1278] Example: A review that says "It's the best" will be classified into cluster 2.

[1279] Step 8:

[1280] The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity.

[1281] Example: The similarity is evaluated by comparing with reviews in Cluster 2 such as "A very enjoyable work" and "Highly praised."

[1282] Step 9:

[1283] The server generates optimal reviews by selecting the most similar reviews within the cluster and summarizing them as reviews.

[1284] For example, reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[1285] Step 10:

[1286] The server transmits the generated review information to the user's terminal.

[1287] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[1288] Step 11:

[1289] Users use the displayed reviews to purchase or watch products or content.

[1290] Example: User A purchases a new manga based on reviews and has a highly satisfying purchasing experience.

[1291] This series of processes allows users to easily find products and content that suit their preferences based on reliable review information.

[1292] Example 1

[1293] 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."

[1294] Conventional shopping sites and content distribution services lack systems that can effectively analyze user-submitted reviews and purchase histories to provide appropriate word-of-mouth information tailored to users' preferences. As a result, users have difficulty finding products and content that match their preferences, and reviews are often unreliable, resulting in an unsatisfactory purchasing experience. A solution to this problem is needed.

[1295] 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.

[1296] In this invention, the server includes means for collecting user ratings and transaction histories, means for removing unnecessary information from the collected ratings and transaction histories to generate clean text data, and means for extracting features of the generated clean text data, thereby enabling users to easily find products and content that suit their preferences.

[1297] "User" refers to a person who uses the service or system to make reviews or purchases.

[1298] "Ratings" refer to opinions and impressions posted by users about products or content.

[1299] "Transaction history" refers to records of purchases, browsing, etc. made by a user using a service or system.

[1300] "Removing unnecessary information" refers to the process of removing unimportant information or noise from collected data.

[1301] "Clean text data" refers to data in a text format suitable for analysis after unnecessary information has been removed.

[1302] "Features" refer to important attributes or elements extracted from text data expressed as numerical values.

[1303] A "cluster" refers to a grouping of ratings or comments with similar characteristics.

[1304] The "KMeans algorithm" refers to a type of machine learning algorithm for classifying data into a specified number of clusters.

[1305] "Cosine similarity" refers to a mathematical index for calculating the similarity between two pieces of text data.

[1306] "Word-of-mouth information" refers to a collection of useful information about products and content provided to users.

[1307] "TF-IDF vectorizer" refers to a statistical method for calculating the importance of words in a text.

[1308] "Database" refers to a structured collection of information for storing and managing collected data.

[1309] This invention is a system that enhances the review function of shopping sites and content distribution services, allowing users to easily find products and content that suit their preferences. Specifically, it collects user ratings and transaction histories, analyzes this data, and generates appropriate word-of-mouth information to provide to users.

[1310] Hardware and software used

[1311] 1. Server:

[1312] Hardware used: database servers, storage devices, processors (CPUs), memory, dedicated servers for clustering, and network equipment.

[1313] Software used: MySQL or PostgreSQL (database management systems), Python (programming language), NLTK and SpaCy (natural language processing toolkits), Scikit-learn (machine learning library), Django (web framework).

[1314] 2. Terminal:

[1315] The hardware used: the user's PC, smartphone, or tablet.

[1316] Software used: Web browser, mobile application (interface for displaying reviews).

[1317] 3. User:

[1318] Hardware used: PC, smartphone, tablet.

[1319] Software used: Web browser, mobile application.

[1320] Specific examples of processing

[1321] For example, if a user wants to see reviews of new anime on a shopping site, the following steps would occur:

[1322] 1. User B has previously posted a review saying, "I like the touching story." Based on this, the server accumulates that data.

[1323] 2. When User B posts a review of a new anime, the server analyzes the review and calculates its features, assigning a high weight to keywords such as "moving story."

[1324] 3. When the server classifies a new review into an existing cluster, it selects reviews in the same cluster that are highly similar to the "moving and brought me to tears" category.

[1325] 4. By generating word-of-mouth information and sending it to the device, User B can easily find information that matches their preferences.

[1326] Prompt Sentence Examples

[1327] "User A has posted a review of a new manga. Please explain in detail the process for analyzing that review, classifying it into existing clusters, and selecting the reviews with the most similarities."

[1328] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1329] Step 1:

[1330] Users post reviews.

[1331] Input: Review text data

[1332] Output: The submitted review is sent to the server.

[1333] The user enters a comment in the review input field and clicks the post button.

[1334] Example: You write a review saying, "This product is very easy to use."

[1335] Step 2:

[1336] The server collects reviews and purchase history and stores them in a database.

[1337] Input: submitted reviews, related purchase history

[1338] Output: Reviews and purchase history stored in a database

[1339] The server collects user-submitted reviews and purchase history in real time and stores them in a database.

[1340] Example: User A's reviews and associated purchase history are stored in a database.

[1341] Step 3:

[1342] The server removes unnecessary information from the collected reviews and generates clean text data.

[1343] Input: Reviews stored in the database

[1344] Output: Clean text data with unnecessary information removed

[1345] The server uses natural language processing techniques (NLTK, SpaCy, etc.) to remove unnecessary information from the text (e.g., HTML tags and special characters).

[1346] Example: Extracting the important keyword "easy to use" from reviews.

[1347] Step 4:

[1348] The server extracts features using clean text data.

[1349] Input: clean text data

[1350] Output: Features of text data (data expressed as numbers)

[1351] The server uses natural language processing and machine learning techniques (such as Scikit-learn) to extract features that represent the content of the text as numbers.

[1352] Example: The keyword "easy to use" is assigned a high value.

[1353] Step 5:

[1354] The server uses the KMeans algorithm to classify reviews into clusters based on their features.

[1355] Input: Features of text data

[1356] Output: Reviews sorted into clusters

[1357] The server runs the KMeans algorithm to classify reviews and comments into clusters based on their features.

[1358] For example, highly similar reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1359] Step 6:

[1360] When a user submits a new review, the server calculates its features and classifies it into an existing cluster.

[1361] Input: New review text data

[1362] Output: New reviews are classified into the appropriate cluster

[1363] A user enters a new review and clicks the submit button. The server calculates the features of the new review and classifies it into an existing cluster.

[1364] Example: User A's review "This new manga is great" will be classified into the appropriate cluster.

[1365] Step 7:

[1366] The server calculates the cosine similarity and selects the most similar reviews.

[1367] Input: Reviews in cluster, features of new reviews

[1368] Output: Most similar reviews

[1369] The server calculates the cosine similarity of the new review with other reviews in the cluster and selects the most similar review.

[1370] Example: Reviews within a cluster such as "Very enjoyable" or "A masterpiece" are selected.

[1371] Step 8:

[1372] The server generates optimal word-of-mouth information and sends it to the device.

[1373] Input: Data of the most similar reviews

[1374] Output: Reviews displayed on the user's device

[1375] The server generates optimal word-of-mouth information based on the selected reviews and sends it to the user's device.

[1376] Example: User A's device will display review information such as "A very enjoyable work" and "A masterpiece."

[1377] (Application example 1)

[1378] 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."

[1379] It is not easy for users to find products and content that suit their preferences on shopping sites and content distribution services. It is also difficult to quickly obtain useful word-of-mouth information from the vast amount of review information. Furthermore, existing systems for increasing user satisfaction have room for improvement, as they lack relevant review information when posting a new review. There is a need to provide a means to solve these problems and provide users with the most appropriate word-of-mouth information.

[1380] 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.

[1381] In this invention, the server includes: means for collecting user reviews and purchase histories; means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data; means for extracting features from the generated clean text data; means for classifying reviews and comments into clusters based on the extracted features; means for calculating features of newly posted reviews and classifying them into the closest cluster; means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information; means for displaying the generated word-of-mouth information to the user; and means for recommending optimal reviews generated from the user's review history and purchase history when the user posts a new review. This allows users to quickly find products and content that suit their preferences and provides a more satisfying purchasing experience. Furthermore, related review information is presented when posting a new review, which also contributes to improving the user's review posting experience.

[1382] "User" refers to a person who uses a shopping site or content distribution service.

[1383] A "review" is text information in which a user writes their own opinion or evaluation of a product or content.

[1384] "Purchase history" is data that includes a list of products that a user has purchased and information about those products.

[1385] "Clean text data" refers to data from which unnecessary information has been removed and only important information has been retained.

[1386] A "feature amount" is a numerical representation of the content of text data, and is a parameter that indicates the characteristics and trends of the data.

[1387] A "cluster" is a group of data that have similar characteristics.

[1388] "Word-of-mouth information" is information that provides optimal reviews and ratings based on user reviews and purchase history.

[1389] "Cosine similarity" is an index that calculates the similarity between two vectors, and is primarily used to evaluate the similarity of text data.

[1390] "TF-IDF vectorizer" is a method for quantifying important words from text data and calculating their importance.

[1391] "Recommendation" is the act of recommending the most suitable products or content based on a user's past actions and data.

[1392] The "best review" is the review that is determined to be the most relevant by evaluating the similarity between the reviews of users in the cluster and the newly posted review.

[1393] System Overview

[1394] This invention is a system that allows users to easily find products and content that suit their preferences. The system generates optimal word-of-mouth information based on users' reviews and purchase history and provides it to users. Implementing the system requires the following steps:

[1395] System configuration

[1396] Hardware

[1397] This system uses hardware including a server and a user terminal (e.g., a smartphone). The server is used to collect, process, generate, and distribute data, and the user terminal displays the generated word-of-mouth information to the user.

[1398] software

[1399] The following software is used to realize this system.

[1400] requests library: Connects to external APIs.

[1401] scikit-learn library: Natural language processing, clustering, and similarity calculations.

[1402] TfidfVectorizer: Quantifies important words from text data and calculates their importance.

[1403] Processing flow

[1404] 1. Data Collection

[1405] The server collects user reviews and purchase history, for example, using an API to retrieve review history and purchase history based on user ID.

[1406] 2. Data cleansing

[1407] Unnecessary information is removed from the collected data to generate clean text data, using natural language processing techniques such as TfidfVectorizer.

[1408] 3. Feature Extraction and Clustering

[1409] Features of the clean text data are extracted using TfidfVectorizer, and reviews and comments are classified into clusters based on the extracted features using the KMeans clustering algorithm.

[1410] 4. Similarity Assessment

[1411] When a new review is submitted, the features of the review are calculated and it is classified into an existing cluster.The similarity of the new review to other reviews in the cluster is then evaluated using cosine similarity.

[1412] 5. Generating and displaying reviews

[1413] The system generates optimal word-of-mouth information using the most similar reviews and displays it on the user's device. When a user submits a new review, the system recommends the most appropriate review based on the history information it has built.

[1414] Specific examples

[1415] For example, User A posts a review saying, "This new device is very easy to use!" When this review is entered into the system, the server provides the most similar review information from past reviews. This allows User A to obtain review information for other devices that suit his or her preferences.

[1416] Example prompts to input to a generative AI model:

[1417] Get the review and purchase history of user ID: 12345 from the API and output the existing review that is most similar to the new review, "This new device is so easy to use!"

[1418] In this way, the present invention provides a specific method for users to have a highly satisfying purchasing experience.

[1419] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1420] Step 1:

[1421] The server collects user reviews and purchase history. Specifically, the server uses an API to retrieve relevant review and purchase history data based on the user ID. The input is the user ID, and the output is the user's review and purchase history data. The collected data is used as input for the next processing step.

[1422] Step 2:

[1423] The server cleanses the collected reviews and purchase history data to generate clean text data. Specifically, the server uses natural language processing techniques to remove unnecessary information and identify key words and phrases. The input is the collected review and purchase history data, and the output is clean text data. This cleansed data is used in the next step of feature extraction.

[1424] Step 3:

[1425] The server extracts features from the cleansed text data. Specifically, the server quantifies the features of the text data using TfidfVectorizer, and then extracts important product features based on the quantified features. The input is the clean text data, and the output is a feature vector. The extracted features are used in the next clustering step.

[1426] Step 4:

[1427] The server classifies reviews and comments into clusters based on their features. Specifically, the server uses the KMeans clustering algorithm to group reviews with high similarity. The input is a feature vector, and the output is reviews classified into clusters. The classified reviews are used in the next similarity evaluation step.

[1428] Step 5:

[1429] The server calculates the features of newly posted reviews and classifies them into the closest cluster. When a new review is posted, the server extracts the features of the review using TfidfVectorizer and compares them with existing clusters to classify it into the most appropriate cluster. The input is the newly posted review, and the output is the review classified into the closest cluster.

[1430] Step 6:

[1431] The server evaluates the similarity of the newly submitted review to other reviews in the cluster. Specifically, the server calculates the similarity between each review using cosine similarity and determines the most similar review. The input is the newly submitted review and other reviews in the cluster, and the output is the most similar review.

[1432] Step 7:

[1433] The server generates optimal review information and displays it on the user's device. The review information generated based on the new review is delivered to the user's device and visualized. The input is the most similar review and related review information, and the output is the optimal review information displayed on the user's device.

[1434] Step 8:

[1435] When a user submits a new review, the server recommends the most appropriate review based on their review history and purchase history. The server presents the most appropriate review information to the user based on past data, improving the user's review submission experience. The input is the user's review history and purchase history, and the output is the recommended optimal review.

[1436] 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.

[1437] MODE FOR CARRYING OUT THE INVENTION

[1438] System Overview

[1439] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects user reviews and purchase histories, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[1440] System configuration

[1441] The system functions through collaboration between the server, terminals, and users as follows:

[1442] 1. The server collects user reviews and purchase history and stores them in a database. Specific information includes review content, rating score, purchase date and time, etc.

[1443] Example: If user A posts a review such as "This product is very easy to use," it will be stored on the server.

[1444] 2. The server uses natural language processing technology to remove unnecessary information from the collected reviews and comments and generate clean text data.

[1445] Example: Important keywords such as "easy to use" are extracted from User A's review.

[1446] 3. The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of the reviews and comments.

[1447] Example: A positive emotion is perceived from the review "Easy to use."

[1448] 4. The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[1449] Example: The TF-IDF value of "easy to use" is represented as part of a numeric vector.

[1450] 5. The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as the basis for clustering.

[1451] Example: Similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1452] 6. When a user posts a new review, the server calculates the review's features and sentiment and classifies it into an existing cluster.

[1453] Example: If user A posts a review saying "This new manga is awesome," the review will be classified into the appropriate cluster.

[1454] 7. The server evaluates the similarity of the new review to other reviews in the cluster using cosine similarity and taking into account sentiment information.

[1455] For example, the review is evaluated for its similarity to other positive reviews in the cluster, such as "highly enjoyable" or "rave reviews."

[1456] 8. The server generates optimal reviews and compiles them as reviews. The information is generated based on the positive emotions obtained by the emotion engine.

[1457] Example: Positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[1458] 9. The server sends the generated review information to the user's device.

[1459] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[1460] 10. Users use the displayed reviews to purchase or watch products or content.

[1461] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[1462] This system improves the reliability of reviews, allowing users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive sentiment.

[1463] The processing flow will be explained below.

[1464] Step 1:

[1465] The server collects user reviews and purchase history and stores them in a database. Collected information includes review content, rating score, purchase date and time, etc.

[1466] Example: When user A posts a review saying "This product is very easy to use," the review and purchase history are saved on the server.

[1467] Step 2:

[1468] The server uses natural language processing technology to remove unnecessary information (e.g., symbols and meaningless words) from the collected reviews and purchase history to generate clean text data.

[1469] Example: Important keywords such as "easy to use" are extracted from User A's review.

[1470] Step 3:

[1471] The server extracts features from the generated clean text data and then uses an emotion engine to recognize the emotions of reviews and comments.

[1472] Example: A positive emotion is recognized from the review "Easy to use."

[1473] Step 4:

[1474] The server uses a TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to quantify the features and convert them into numerical vectors.

[1475] Example: The TF-IDF value of "easy to use" is represented as a numeric vector.

[1476] Step 5:

[1477] The server uses the KMeans algorithm to classify reviews and comments into clusters, using the emotional information recognized by the emotion engine as a basis for clustering.

[1478] For example, highly similar positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1479] Step 6:

[1480] A user posts a new review.

[1481] Example: User A posts a review saying, "This new manga is great."

[1482] Step 7:

[1483] The server calculates the features and sentiment of new reviews, quantifies them using a TF-IDF vectorizer, and classifies them into existing clusters.

[1484] For example, a review that says "It's the best" will be classified into cluster 2, recognizing a positive sentiment.

[1485] Step 8:

[1486] The server evaluates the similarity of a new review to other reviews in the cluster using cosine similarity, taking sentiment information into account.

[1487] Example: The similarity with other positive reviews in Cluster 2, such as "A very enjoyable work" and "Highly praised," is evaluated.

[1488] Step 9:

[1489] The server generates optimal reviews and summarizes them based on the positive emotions obtained from the emotion engine.

[1490] For example, positive reviews such as "A very enjoyable work" and "Highly praised" are generated as word-of-mouth information.

[1491] Step 10:

[1492] The server transmits the generated review information to the user's terminal.

[1493] Example: User A's device will display information such as "A very enjoyable work" and "Highly praised."

[1494] Step 11:

[1495] Users use the displayed reviews to purchase or watch products or content.

[1496] Example: User A purchases a new manga based on recommended reviews and has a highly satisfying purchasing experience.

[1497] This series of processes allows users to easily find products and content that suit their preferences based on reliable word-of-mouth information based on positive emotions.

[1498] Example 2

[1499] 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."

[1500] The review functions on existing shopping sites and content distribution services have made it difficult for users to find products and content that match their preferences. Furthermore, there is a lack of technology that can properly recognize the sentiment behind reviews and comments and provide them as organized word-of-mouth information. This makes it time-consuming for users to select products, resulting in lower satisfaction.

[1501] 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 user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, means for extracting features from the generated clean text data, means for recognizing emotions from the clean text data, means for converting the features into numerical vectors, means for classifying reviews and comments into clusters based on the extracted features and emotions, means for calculating features and emotions of a newly posted review and classifying it into the closest cluster, means for evaluating the similarity of the newly posted review with other reviews in the cluster and generating optimal word-of-mouth information, and means for displaying the generated word-of-mouth information to the user. This allows users to more easily find products and content that suit their preferences.

[1502] "User" refers to the end user who uses the System to make reviews and purchases.

[1503] "Review" refers to an evaluation or comment made by a user on a product or content.

[1504] "Purchase history" refers to a record of products and content purchased by a user in the past.

[1505] "Server" refers to a computer system that receives input from users and processes, stores, and analyzes data.

[1506] "Database" refers to a system for systematically storing and managing collected data.

[1507] "Removing unnecessary information" refers to the process of removing noise and unnecessary elements from collected data to make it clean.

[1508] "Clean text data" refers to text data after unnecessary information has been removed.

[1509] "Features" refer to important elements or attributes extracted from text data.

[1510] "Emotion recognition" refers to the process of identifying the emotional tone of review and comment text.

[1511] A "numeric vector" refers to text data that has been digitized and expressed in vector format.

[1512] "Clustering" refers to the process of grouping reviews or comments that are highly similar.

[1513] A "cluster" refers to a group of similar data points.

[1514] "Evaluating similarity" refers to the process of measuring the similarity between reviews and comments and using it for classification and ranking.

[1515] "Best review information" refers to the review content that is judged to be most useful to users.

[1516] "Cosine similarity" refers to a method for calculating similarity based on the angle between two vectors.

[1517] "TF-IDF vectorizer" refers to an algorithm that calculates the importance of words in text and represents them as numerical vectors.

[1518] This invention is a system that enhances the review function of shopping sites and content distribution services. This system collects reviews and purchase histories posted by users, recognizes the emotions in the reviews and comments using an emotion engine, and generates and displays word-of-mouth information that is optimal for each user, allowing users to easily find products and content that suit their preferences.

[1519] System configuration

[1520] The system is configured as follows:

[1521] Hardware and Software Use

[1522] 1. Server:

[1523] Data collection and storage: We collect user reviews and purchase history and store them in a database (e.g., MySQL or PostgreSQL).

[1524] Data cleaning: Using an NLP library (e.g., SpaCy or NLTK), unnecessary information is removed from the collected reviews to generate clean text data.

[1525] Feature extraction and quantification: Features are extracted and converted into numerical vectors using a TF-IDF vectorizer.

[1526] Sentiment Recognition: Recognize the sentiment of reviews and comments using sentiment analysis libraries (e.g., VADER or TextBlob).

[1527] Clustering: Using the KMeans algorithm, reviews and comments are classified into clusters based on features and sentiment information.

[1528] Similarity evaluation: The similarity between the reviews in the cluster and the newly posted review is evaluated using cosine similarity.

[1529] Word-of-mouth information generation: Generate optimal word-of-mouth information based on the similarity and sentiment information of reviews within a cluster.

[1530] Information transmission: Send the most appropriate review information to the user's device.

[1531] 2. Terminal:

[1532] Display information: Display the review information sent from the server to the user.

[1533] 3. User:

[1534] Review submission: Visit a shopping site or content distribution service and enter a review of a product or content.

[1535] Information reference: The displayed reviews are used as a reference to purchase or watch a product or content.

[1536] Specific examples

[1537] For example, if a user posts a review stating, "This product is very easy to use," the review is collected by the server. The server extracts the key keyword "easy to use" from the collected reviews and recognizes positive sentiment through sentiment analysis. It then calculates the TF-IDF value of "easy to use" using a TF-IDF vectorizer and converts it into a numerical vector. Based on this information, the KMeans algorithm classifies the review into a cluster as a positive review. The new review is then added to the cluster, and its similarity with other reviews is evaluated using cosine similarity to generate optimal review information. Finally, this review information is sent to the user's device, where the user can use the displayed review information to purchase a new product.

[1538] Prompt Sentence Examples

[1539] Below are some specific examples of prompt sentences to input into the generative AI model.

[1540] Example prompt sentence:

[1541] "Please explain the system that recognizes the sentiment of reviews posted by users and generates optimal user reviews. Please provide details of what hardware or software the system uses and what data processing and calculations it performs. Please also include specific examples."

[1542] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1543] Step 1:

[1544] A user posts a review. The user accesses a shopping site or content distribution service, enters a review of the product or content, and presses the submit button. The input includes the user's text review and rating score, which are then sent to the system.

[1545] Step 2:

[1546] The server collects the posted reviews and stores them in a database. The input is the review and rating score submitted by the user, which is received and stored in a database (e.g., MySQL or PostgreSQL). Specifically, if user A posts a review such as "This product is very easy to use," it is stored on the server.

[1547] Step 3:

[1548] The server removes unnecessary information from the collected reviews and generates clean text data. The input is reviews stored in a database, which are then processed using a natural language processing library (e.g., SpaCy or NLTK) to remove HTML tags and special symbols. The output is clean text data such as "easy to use." Specifically, the system extracts the important keyword "easy to use" from User A's review.

[1549] Step 4:

[1550] The server extracts features from the clean text data. The input is the clean text data, which is then vectorized using techniques such as Bag-of-Words and TF-IDF. The output is a numerical vector indicating the importance of each word. Specifically, the server calculates the frequency of occurrence of the phrase "easy to use" and vectorizes it.

[1551] Step 5:

[1552] The server uses an emotion engine to recognize the sentiment of reviews and comments. The input is vectorized text data, which is then used to recognize the sentiment using a sentiment analysis library (e.g., VADER or TextBlob). The output is the emotional tone (positive, negative, neutral). Specifically, a positive sentiment is recognized from the review "Easy to use."

[1553] Step 6:

[1554] The server converts the features into a numerical vector. The input is clean text data with recognized emotions, which is converted into a numerical vector using a TF-IDF vectorizer. The output is a feature expressed as a numerical vector. Specifically, the TF-IDF value of "easy to use" is calculated and converted into a numerical vector.

[1555] Step 7:

[1556] The server uses the KMeans algorithm to classify reviews and comments into clusters. The input is a numerical vector of features containing emotional information, which are then classified using KMeans clustering. The output is the formation of clusters. Specifically, positive reviews such as "convenient and easy to use" and "very easy to use" are classified into the same cluster.

[1557] Step 8:

[1558] A user posts a new review. The user writes a new review, such as "This new manga is great," and presses the submit button. The input includes the user's new text review and rating score, which are then sent to the system.

[1559] Step 9:

[1560] The server calculates the features and emotions of new reviews and classifies them into existing clusters. The input is the text of the new review and its rating score, which are extracted using the process described above to extract features and perform emotion recognition. After that, they are converted into numerical vectors and classified into existing clusters. The output is that the new review is added to the appropriate cluster. Specifically, the new review is classified into an existing cluster with the same emotion.

[1561] Step 10:

[1562] The server evaluates the similarity within the cluster and generates the optimal review information. The input is the features and emotional information added to the cluster, and the similarity of the reviews is evaluated using cosine similarity. The output is the optimal review information. Specifically, reviews of "Very enjoyable work" and "Highly praised" in the same cluster are evaluated as having high similarity and are combined into new review information.

[1563] Step 11:

[1564] The server sends the generated review information to the user's device. The input is the generated review information, which is sent to the user's device. The output is the review information being displayed on the user's device. Specifically, information such as "Very enjoyable work" and "Highly praised" is displayed on the user's smartphone.

[1565] Step 12:

[1566] Users use the displayed reviews to purchase or watch products or content. The input is the displayed reviews, and they use this information to purchase products or watch content. Specifically, a user sees a review that says "this is a very enjoyable work," purchases a new manga, and has a highly satisfying purchasing experience.

[1567] (Application example 2)

[1568] 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."

[1569] Conventional recommendation systems based on reviews and purchase history have difficulty presenting optimal review information for each user, and have been unable to sufficiently increase user satisfaction. Furthermore, they lack the ability to recognize the detailed sentiment of reviews, making it impossible to provide personalized recommendations based on the user's emotions. In particular, there has been a lack of methods to provide a more intuitive and engaging user experience by utilizing head-mounted displays (HMDs).

[1570] 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.

[1571] In this invention, the server includes means for collecting user reviews and purchase histories, means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data, and means for extracting features from the generated clean text data. This makes it possible to perform clustering based on the review features, classify newly posted reviews into appropriate clusters, and display a recommendation list on the head-mounted display that also takes into account the user's emotional data.

[1572] "User reviews and purchase history" refers to the impressions and ratings of products posted by users, as well as the history of products they have purchased.

[1573] "Means for removing unnecessary information and generating clean text data" refers to a processing device or algorithm that removes noise and unnecessary parts from the collected reviews and purchase history and formats it into an analyzable format.

[1574] "Means for extracting features from the generated clean text data" refers to technology for numerically expressing important words and phrases contained in the text data.

[1575] "Means for classifying reviews and comments into clusters" refers to an algorithm for grouping similar reviews and comments based on their features.

[1576] "Method of classifying into the closest cluster" refers to the method used to assign newly submitted reviews to existing clusters.

[1577] "Means for evaluating the similarity of reviews and generating optimal word-of-mouth information" refers to technology that analyzes the similarity in content and sentiment between reviews and generates review information that is useful to users.

[1578] "Means for displaying the generated review information to the user" refers to equipment or software that visually displays the generated review information on the user's device.

[1579] "Means for recommending content using emotional data" refers to a system that recommends content that matches a user's preferences based on the results of an analysis of the user's emotions.

[1580] "Means for generating a recommendation list to be displayed on a head-mounted display" refers to technology that generates and visually presents a list of recommended content to be displayed on a user's visual device.

[1581] This invention is a system that analyzes user reviews and purchase history to generate and display a personalized recommendation list. How this system is implemented will be explained below.

[1582] First, the server collects user reviews and purchase history, including data from shopping sites and content distribution services, such as review content, rating scores, and purchase dates and times.

[1583] The server then removes unnecessary information from the collected reviews and purchase history to generate clean text data. This process includes text normalization and noise reduction, and uses natural language processing techniques. Specifically, it applies a natural language processing library called Spacy.

[1584] The server then extracts features from the generated clean text data, which include important words and phrases and quantify them using a TF-IDF vectorizer.

[1585] The KMeans algorithm is applied to cluster reviews and comments based on the extracted features. A newly submitted review is also classified into the closest cluster based on its features. This classification evaluates the similarity of the newly submitted review to other reviews in the cluster.

[1586] The server uses cosine similarity to evaluate the similarity and also considers the sentiment of the reviews using an emotion engine, resulting in optimal reviews based on positive sentiment and generated in a way that is most appropriate for each user.

[1587] Furthermore, the server utilizes the user's review history and emotional data to recommend personalized content. This information is generated as a recommendation list displayed on a head-mounted display (HMD), which provides users with intuitive and engaging visual information.

[1588] As a concrete example, if users post reviews on a movie website and collect sentiment data, the following prompts can be fed into a generative AI model for sentiment analysis:

[1589] This movie was very moving and I want to watch it again.

[1590] Use TextBlob to perform sentiment analysis as follows:

[1591] python

[1592] textblob.TextBlob("This movie moved me deeply. I want to watch it again.").sentiment.polarity

[1593] This generates a list of new movie recommendations based on the user's emotional data and displays them to the user through the HMD, allowing users to easily find personalized content based on their preferences and emotions.

[1594] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1595] Step 1:

[1596] The server collects user reviews and purchase history from shopping sites and content distribution services. Specific data includes review content, rating scores, purchase dates and times, etc. This data is stored as raw data on the server. The input is the user behavior log, and the output is a dataset of the collected reviews and purchase history.

[1597] Step 2:

[1598] The server removes unnecessary information from the collected reviews and purchase history to generate clean text data. Specifically, it uses natural language processing technology (e.g., Spacy) to normalize and denoise the text. The input is raw data, and the output is clean text data with unnecessary information removed.

[1599] Step 3:

[1600] The server extracts features from the generated clean text data. It uses a TF-IDF vectorizer to quantify important words and phrases contained in the text data. The input is the clean text data, and the output is feature data in the form of a numerical vector.

[1601] Step 4:

[1602] The server classifies reviews and comments into clusters based on the extracted features. It uses the KMeans algorithm to group similar reviews and comments into clusters. The input is the feature data, and the output is a list of reviews and comments classified into clusters.

[1603] Step 5:

[1604] The server calculates the features of a newly submitted review and classifies it into the closest cluster. This process uses the TF-IDF vectorizer, as in the previous step. The input is the text data of the new review, and the output is the cluster classification result.

[1605] Step 6:

[1606] The server evaluates the similarity of a newly posted review with other reviews in the cluster. It uses cosine similarity and an emotion engine to analyze the similarity of content and emotions between reviews. The input is the feature data of reviews in the cluster and the feature data of the new review, and the output is the similarity evaluation result.

[1607] Step 7:

[1608] The server generates optimal word-of-mouth information. Based on the sentiment analysis results and similarity evaluation results, it generates word-of-mouth information that is most suitable for the user. The input is the similarity evaluation results and sentiment analysis data, and the output is the generated word-of-mouth information.

[1609] Step 8:

[1610] The server displays the generated word-of-mouth information to the user. Specifically, it generates and visually presents a recommendation list to be displayed on a head-mounted display (HMD). The input is the generated word-of-mouth information, and the output is the recommendation list displayed on the HMD.

[1611] Step 9:

[1612] Users can view the recommendation list through the HMD and select content they are interested in. This process allows users to easily find personalized content based on their preferences and emotions. The input is the recommendation list displayed on the HMD, and the output is the user's selected content.

[1613] 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.

[1614] 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.

[1615] 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.

[1616] 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.

[1617] 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.

[1618] 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.

[1619] 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).

[1620] 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.

[1621] 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."

[1622] 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.

[1623] 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).

[1624] 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.

[1625] 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.

[1626] 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.

[1627] 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.

[1628] 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.

[1629] 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.

[1630] 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.

[1631] 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.

[1632] 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.

[1633] 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.

[1634] The following is further disclosed regarding the above embodiment.

[1635] (Claim 1)

[1636] a means of collecting user reviews and purchase history;

[1637] A means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data;

[1638] A means for extracting features from the generated clean text data;

[1639] a means for classifying reviews and comments into clusters based on the extracted features;

[1640] A method to calculate the features of newly posted reviews and classify them into the closest cluster.

[1641] A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal word-of-mouth information;

[1642] The system includes a means for displaying the generated review information to a user.

[1643] (Claim 2)

[1644] 10. The system of claim 1, further comprising means for evaluating the similarity of the newly posted review with other reviews in the cluster using cosine similarity.

[1645] (Claim 3)

[1646] 2. The system according to claim 1, further comprising: means for calculating importance of words in reviews using a TF-IDF vectorizer when generating word-of-mouth information.

[1647] "Example 1"

[1648] (Claim 1)

[1649] means of collecting user ratings and transaction history;

[1650] A means for removing unnecessary information from the collected evaluations and transaction history to generate clean text data;

[1651] A means for extracting features from the generated clean text data;

[1652] means for classifying the ratings and comments into clusters based on the extracted features;

[1653] A means to calculate the features of newly submitted ratings and classify them into the closest cluster;

[1654] A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal reviews;

[1655] a means for displaying the generated review information to users;

[1656] means for storing the collected and analyzed data in a database;

[1657] A means of using the KMeans algorithm for clustering;

[1658] The system includes a means for calculating cosine similarity and selecting reviews with high similarity.

[1659] (Claim 2)

[1660] 2. The system of claim 1, further comprising means for evaluating the similarity of the newly posted rating with other ratings in the cluster using cosine similarity.

[1661] (Claim 3)

[1662] 2. The system according to claim 1, further comprising means for calculating the importance of evaluation words using a TF-IDF vectorizer when generating word-of-mouth information.

[1663] "Application Example 1"

[1664] (Claim 1)

[1665] a means of collecting user reviews and purchase history;

[1666] A means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data;

[1667] A means for extracting features from the generated clean text data;

[1668] a means for classifying reviews and comments into clusters based on the extracted features;

[1669] A method to calculate the features of newly posted reviews and classify them into the closest cluster.

[1670] A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal word-of-mouth information;

[1671] a means for displaying the generated review information to users;

[1672] A system including a means for recommending optimal reviews generated from a user's review history and purchase history when the user posts a new review.

[1673] (Claim 2)

[1674] 2. The system according to claim 1, further comprising means for evaluating the similarity of a newly posted review with other reviews in the cluster using cosine similarity, and displaying related content.

[1675] (Claim 3)

[1676] The system according to claim 1, further comprising means for automatically calculating the importance of words in a review using a TF-IDF vectorizer when a new review is posted, and recommending related reviews.

[1677] "Example 2: Combining Emotion Engines"

[1678] (Claim 1)

[1679] a means of collecting user reviews and purchase history;

[1680] A means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data;

[1681] A means for extracting features from the generated clean text data;

[1682] A means of recognizing sentiment in clean text data;

[1683] A means for converting the feature quantity into a numerical vector;

[1684] a means for classifying the reviews and comments into clusters based on the extracted features and sentiments;

[1685] A method to calculate the features and sentiment of newly posted reviews and classify them into the closest cluster;

[1686] A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal word-of-mouth information;

[1687] The system includes a means for displaying the generated review information to a user.

[1688] (Claim 2)

[1689] 10. The system of claim 1, further comprising means for evaluating the similarity of the newly posted review with other reviews in the cluster using cosine similarity.

[1690] (Claim 3)

[1691] 2. The system according to claim 1, further comprising: means for calculating importance of words in reviews using a TF-IDF vectorizer when generating word-of-mouth information.

[1692] "Application example 2 when combining emotion engines"

[1693] (Claim 1)

[1694] a means of collecting user reviews and purchase history;

[1695] A means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data;

[1696] A means for extracting features from the generated clean text data;

[1697] a means for classifying reviews and comments into clusters based on the extracted features;

[1698] A method to calculate the features of newly posted reviews and classify them into the closest cluster.

[1699] A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal word-of-mouth information;

[1700] a means for displaying the generated review information to users;

[1701] A means of recommending content using user review history and sentiment data;

[1702] A method for generating a recommendation list to be displayed on a head-mounted display

[1703] A system including:

[1704] (Claim 2)

[1705] 10. The system of claim 1, further comprising means for evaluating the similarity of the newly posted review with other reviews in the cluster using cosine similarity.

[1706] (Claim 3)

[1707] 2. The system according to claim 1, further comprising: means for calculating importance of words in reviews using a TF-IDF vectorizer when generating word-of-mouth information. [Explanation of symbols]

[1708] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. A means of collecting user reviews and purchase history; A means for removing unnecessary information from the collected reviews and purchase histories to generate clean text data; A means for extracting features from the generated clean text data; a means for classifying reviews and comments into clusters based on the extracted features; A method to calculate the features of newly posted reviews and classify them into the closest cluster. A method for evaluating the similarity of a newly posted review with other reviews in the cluster and generating optimal word-of-mouth information; The system includes a means for displaying the generated review information to a user.

2. The system of claim 1 , further comprising means for evaluating the similarity of the newly posted review with other reviews in the cluster using cosine similarity.

3. The system according to claim 1, further comprising means for calculating importance of words in reviews using a TF-IDF vectorizer when generating word-of-mouth information.

Images