Comment presentation device and comment presentation method
Through the comment prompt device, using prediction models and language models to generate comments such as those written by the user themselves, solving the problem that the prior art cannot generate comments for things that have not been commented, and providing users with more accurate references before online purchases or service appointments .
Patent Information
- Application Number
- CN202410707299.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-06-27
- Filing Date
- 2024-06-03
- Publication Date
- 2024-12-27
AI Technical Summary
The prior art cannot generate comments such as those written by the user for things that have not been commented on, making it difficult for users to find comments with similar experiences when purchasing online or making appointments for services.
Through the comment prompt device, a prediction model and a language model are used to generate comments such as those written by the user themselves, and comments on things that the user has not experienced based on the user's specification information and hobby information.
It enables users to provide comments that are written by themselves, helping users make smarter decisions before purchasing or making a service appointment.
Smart Images
Figure CN119205226A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a review prompting device and a review prompting method, and more particularly to a technology for generating reviews using a prediction model. Background Art
[0002] When purchasing products or booking services online, many people refer to reviews posted by others. Here, there is a technology for automatically generating reviews by predicting reviews related to items (for example, see Patent Document 1 and Non-Patent Document 1).
[0003] [Prior art literature]
[0004] [Patent Document]
[0005] [Patent Document 1] Japanese Patent No. 7133689
[0006] [Non-patent literature]
[0007] [Non-Patent Document 1] Akihiro Kokubo and Kazunari Sugiyama, “Characteristics of Comments in Recommendation Systems and a Model for Descriptive Text Generation Using PPLM,” 14th Data Engineering and Information Management Forum (DEIM 2022), March 2022, pp. C24-6:1 to C24-6:8 Summary of the invention
[0008] [Problems to be solved by the invention]
[0009] In Patent Document 1 and Non-Patent Document 1, the deep learning model must have learning data that has learned the comments and evaluations that the user himself has made on certain things in the past. Therefore, it is impossible to automatically generate comments for users who have not made comments or evaluations on other things in the past.
[0010] The present disclosure has been made in view of such a situation, and an object of the present invention is to provide a technology capable of presenting a user with a review as if written by the user himself or herself about something that the user has not experienced.
[0011] [Technical means to solve the problem]
[0012] In order to solve the above-mentioned problem, a comment prompting device according to an embodiment of the present invention includes: a specification storage unit for storing specification information related to each of a plurality of review objects; an object prompting unit for prompting a user at least one of the plurality of review objects as a candidate for discussing the enjoyment of the review object; an information storage unit for storing information related to the user; a comment generation unit for generating comments as if made by the user using a prediction model based on the specification information of the prompted review object and the information related to the user; and a comment output unit for prompting the generated comments to the user.
[0013] Another embodiment of the present disclosure is a comment prompting method. The method includes the following process: prompting a user at least one of a plurality of comment objects as a candidate for discussing the enjoyment of the comment object; reading out specification information related to the prompted comment object; reading out information related to the user; using a prediction model to generate a comment as if made by the user based on the specification information of the prompted comment object and the information related to the user; and prompting the generated comment to the user.
[0014] [Effects of the Invention]
[0015] According to the present disclosure, a technique can be provided that can present a user with a review about something that the user has not experienced, as if the review was written by the user himself. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 This is a diagram showing the basic structure of the review prompt system.
[0017] Figure 2 This is a functional block diagram showing each structure of the review prompt system.
[0018] Figure 3 This is a diagram showing the basic process of generating comments.
[0019] Figure 4 This is a diagram showing an example of a simple test screen for asking questions when the appearance is used as an evaluation point of view.
[0020] Figure 5 This is a diagram showing an example of a simple test screen for asking questions when the function is used as an evaluation point of view.
[0021] Figure 6 This is a diagram showing an example of a simple test screen that presents a question for understanding the user's expression tendency.
[0022] Figure 7 This is a diagram showing an example of a screen for presenting review target candidates, comments, and predicted evaluation values.
[0023] Figure 8 This is a diagram showing a basic process of generating a comment in the second embodiment.
[0024] Fig. 9 This is a diagram showing a basic process of generating a comment in the third embodiment.
[0025] Fig.10 This is a diagram showing a basic process of generating a comment in the fourth embodiment.
[0026] [Explanation of Symbols]
[0027] 16: Object prompting unit
[0028] 18: Comment output section
[0029] 54: Object prompting unit
[0030] 56: Comment generation department
[0031] 58: Comment output section
[0032] 62: Specifications Storage Department
[0033] 64: Prior information storage unit
[0034] 66: Ask the storage department
[0035] 70: Feedback Acquisition Department DETAILED DESCRIPTION
[0036] Hereinafter, based on the embodiment, Figure 1 In the embodiments and modifications, the same or equivalent components are denoted by the same reference numerals, and duplicate descriptions are omitted as appropriate.
[0037] (First Embodiment)
[0038] When purchasing products or booking services online, many people refer to reviews posted by others who have already enjoyed these products or services. At this time, if there are too many reviews, it may be difficult to decide which review to refer to. In this case, users often search for reviews from people in similar situations to themselves, but they may not be able to find reviews from people in similar situations to themselves.
[0039] Therefore, from the perspective of what kind of comments the user would write if he / she enjoyed the product or service, a temporary comment is generated and presented to the user using the prior information related to the user and the prescribed prediction model and language model. Before enjoying the product or service, that is, before actually purchasing the product or booking the service, the user can see that if it were him / her, he / she might write such a comment and refer to it. The user can decide whether to enjoy the product or service with reference to such a prior comment.
[0040] The review objects are various products or services. For example, in addition to the purchase of products such as shoes and clothes, the application for participation in a marathon competition, the reservation of a hotel, the reservation of a restaurant, etc. are also included in the review objects. In addition, the reservation of services also includes the purchase of tickets such as service admission tickets or seat tickets.
[0041] Figure 1The basic structure of the comment prompting system 100 is shown. The comment prompting system 100 mainly includes a user terminal 10 and a comment prompting server 50. The comment prompting system 100 may further include a language model server 90 having a language model for generating natural text as a comment, or may construct a language model inside the comment prompting server 50. Alternatively, the following specifications may be adopted, namely, a large-scale language model server constructed in the cloud outside the system is used as the language model server 90 via an application programming interface (API). Since learning for constructing a language model from the beginning requires a huge computational cost, a large-scale external language model learned in advance may be used, or a unique language model may be efficiently constructed by fine-tuning such an external language model. Moreover, in such fine-tuning, a data set of past comments accumulated in the domain of the provider of the comment prompting server 50 may be used for re-learning, or a large data set of comments provided by a third party may be used for re-learning. As a result, it is expected that the difference between the generated comments and the actual existing comments will be reduced.
[0042] The review prompt server 50 is a server that generates comments related to the review object based on the prediction model and the language model, and can also have the functions of an online shopping website that sells specific products or an online reservation website that accepts reservations for specific services. The review prompt server 50 is a server computer connected to multiple user terminals 10 via a network 80 including the Internet or a local area network (LAN). The review prompt server 50 can include a single server computer or a combination of multiple server computers. The user terminal 10 is a terminal operated by the user, which can be an information terminal such as a smart phone or a tablet terminal held by the user himself, or a personal computer.
[0043] The user can operate the user terminal 10 to access the online shopping website or online reservation website built on the review prompt server 50, and browse the reviews displayed together with the information of the product or service. In addition, the "review prompt device" mentioned in the claims can refer to the review prompt system 100 as a whole, or to the review prompt server 50. Moreover, in the present embodiment, most of the characteristic functions contained in the "review prompt device" mentioned in the claims are implemented in the form possessed by the review prompt server 50, so the review prompt server 50 is essentially equivalent to the "review prompt device". However, the characteristic functions of the "review prompt device" can also be distributed to the user terminal 10 and the review prompt server 50, and can also be implemented by making the user terminal 10 have most of its forms.
[0044] Figure 2 1 is a functional block diagram showing each structure of the comment prompting system 100. The comment prompting system 100 in this embodiment includes a user terminal 10 and a comment prompting server 50. However, the comment prompting system 100 may be implemented by various hardware structures or software structures. For example, the comment prompting system 100 may include only the comment prompting server 50 or only the user terminal 10.
[0045] Figure 2 In the figure, regarding the user terminal 10 and the comment prompt server 50, the functional blocks realized by the linkage of various hardware structures and software structures are respectively depicted. Therefore, those skilled in the art should understand that these functional blocks can be realized in various forms by hardware only, software only or a combination thereof. The user terminal 10 includes, for example, a combination of hardware such as a microprocessor, a touch panel, a memory, and a communication module. In the user terminal 10, a program with the following functions is installed, or by accessing the website provided by the comment prompt server 50 through a web browser, the following functions are essentially performed through the collaboration of software and hardware. The user terminal 10 includes an operation processing unit 12, a display control unit 14, an object prompt unit 16, a comment output unit 18, and a communication unit 20 as its functions.
[0046] The operation processing unit 12 receives user operations on operation input components such as a touch panel, for example, an operation to select a review object. The display control unit 14 displays information received from the review prompt server 50 on a screen such as a touch panel. The object prompt unit 16 prompts the user with information such as a product or service that is a review object received from the review prompt server 50 by displaying the screen via the display control unit 14. The review output unit 18 outputs the comments generated by the review prompt server 50 in the form of a screen display via the display control unit 14. The object prompt unit 16 and the review output unit 18 may also have the same functions as the object prompt unit 54 and the review output unit 58 in the review prompt server 50 described later. Moreover, the functions of at least one of these "object prompt unit" and "comment output unit" may be realized only on the user terminal 10 side or only on the review prompt server 50 side, or may be realized through the cooperation between the user terminal 10 side and the review prompt server 50 side.
[0047] The review prompt server 50 includes, for example, a combination of hardware such as a microprocessor, a memory, a display, and a communication module. The review prompt server 50 may be a server computer constructed and managed by a shoe manufacturer. Among them, in the review prompt server 50, a program having the following functions is running. The review prompt server 50 includes a communication unit 52, an object prompt unit 54, an evaluation unit 57, a review output unit 58, a data storage unit 60, and a feedback acquisition unit 70 as its functions. The communication unit 20 of the user terminal 10 is connected to the communication unit 52 of the review prompt server 50 via a network 80. The data storage unit 60 includes a specification storage unit 62, a prior information storage unit 64, an inquiry storage unit 66, and a model storage unit 68. The evaluation unit 57 includes an evaluation viewpoint determination unit 53, a preference determination unit 55, a review generation unit 56, and a learning processing unit 59.
[0048] The specification storage unit 62 stores the types of multiple review objects and their respective related specification information. The review object is a product or service that is a candidate for the user to discuss and enjoy. For example, it can be a product such as shoes or clothes, or it can be an event such as a marathon competition or a concert, a service such as an accommodation facility, a restaurant, etc. The specification storage unit 62 classifies and stores these items that may become review objects by category. The specification information is information related to the attributes or specifications of the product or service, such as the quality, price, type, purpose, category, customer base, etc. In the case where the review prompt server 50 also serves as an online shopping website for selling sports shoes, the information used to determine the purpose or type of using the sports shoes can be information related to the product specifications, such as the price, purpose, appearance type, color type, category of sports, corresponding athlete level, etc.
[0049] The object prompting unit 54 sends the data of the online shopping website or the online reservation website to the user terminal 10 via the communication unit 52. The object prompting unit 54 prompts the user at least one of the multiple review objects stored in the specification storage unit 62 as a candidate for discussing the enjoyment of the review object. The review object prompted to the user may be, for example, an object before the user enjoys it, or an object that has not been reviewed regardless of whether it is enjoyed or not, as a candidate for the review object. Whether it is before the user enjoys it can be inferred by referring to the purchase history or order history or the answer obtained from the user as to whether it is before the enjoyment, or the object that the user has enjoyed or the object that has been reviewed can be excluded from the review object. In the case where the review object is a commodity, the object prompting unit 54 electronically displays the commodities sold on the website, such as multiple types of shoes or clothes, so as to prompt multiple commodities as candidates for the user to discuss the enjoyment. When the user indicates the display of details of any commodity, the commodity with the above-mentioned detailed display is selected as the review object.
[0050] Furthermore, when the review object is a service, multiple types of services are electronically introduced on the online reservation website of the service or the online ticket reservation website, thereby prompting multiple services as candidates for the user to discuss and enjoy. When the user instructs to display the details of any service, the service with the details displayed is selected as the review object.
[0051] The prior information storage unit 64 stores prior information acquired and accumulated in advance about the user himself. Here, the prior information generally includes two types of information. As the first information, the prior information is information related to the user himself input in the past or information on the past action history, etc., which has the nature of being general customer information that is not necessarily directly related to or dependent on the product or service selected by the user as the review object. On the other hand, as the second information, the prior information includes the following information, that is, information obtained based on user answers in the form of inquiries in the form of simple tests such as perception surveys conducted in advance on the user in relation to the product or service selected by the user as the review object. The first information and the second information include information that has the nature of being a data set used for machine learning. The first information and the second information are explained below.
[0052] The first information in the prior information includes the user's identification information, attribute information, preference information, product purchase history, service reservation history, product usage history or usage level, user input information, etc. The first information may also include past comments made by the user on other products or services, and may also reflect feedback information described later.
[0053] The user's identification information is, for example, the user's account information or password, information required to log in to the comment prompt server 50, and is generated based on the user's input information during the account registration procedure when the comment prompt server 50 is first used. In addition, other information including the user's attribute information mainly includes: information entered by the user as necessary information during the user's account registration procedure, or information that is not necessarily related to the goods or services selected by the user as the subject of the review and is entered based on a questionnaire for inquiring about the user's interests or concerns. The more the user answers to the large number of items prompted as such a questionnaire, especially the areas or items of concern to themselves, the more it helps to improve the quality or accuracy of the comments in the comment generation described later. By linking the user's identification information with personal information such as the attribute information described below, the user process from the selection of the comment object to the prompt of the comment can be simplified.
[0054] The user's attribute information includes, for example, the user's age, gender, height, weight, place of residence, and language used. In addition, it may also include information such as usage, purpose, customer group, and price level as attributes in the category of a specific product or service. For example, if the product is sports shoes, it may also include information such as foot shape data or athlete level and sports purpose. If the product is clothing, it may also include information such as body shape data or clothing category.
[0055] The preference information includes information such as each user's preference for goods or services, each user's purpose of using goods, each user's purpose of using services, etc. The purchase history of goods or the reservation history of services includes each user's purchase history of goods or the reservation history of services provided by the review presentation server 50 .
[0056] The second information in the prior information includes information on the user's answers to simple tests such as a feeling survey related to the product or service that is the subject of the review. The inquiry storage unit 66 stores simple inquiries such as a feeling survey related to each of the plurality of review subjects, i.e., inquiry information. The subject prompting unit 54 prompts the inquiry information to the user who discusses the enjoyment of the review subject, and obtains the user's answer to the inquiry information as the prior information related to the user.
[0057] The inquiry information is an inquiry for determining which evaluation viewpoint among multiple evaluation viewpoints is used to generate a comment that is in line with or close to the user's preferences. As an evaluation viewpoint, for example, at least one of the criteria such as appearance, function, feeling, number of purchases, service, reaction from a third party, price, etc. is set as the axis of evaluation. The inquiry information corresponding to each evaluation viewpoint is stored in the inquiry storage unit 66 and prompted to the user. The inquiry information may also be an inquiry that allows the user to select which one is in line with the user's preferences based on the prompting of multiple images. Another inquiry information may also be an inquiry that allows the user to select the importance of each item to the user based on the prompting of multiple items. Another inquiry information may also be an inquiry that allows the user to select which one is close to the user's comment in terms of writing style, tone, language, and sentence length based on the prompting of multiple comment examples with different writing styles or tones. The evaluation viewpoint determination unit 53 determines the evaluation viewpoint of the comment based on the user's answer to the inquiry information. It may also be that the more the user answers to the fields or items that the user is concerned about among the large number of items prompted as the inquiry information, the more important the evaluation viewpoint corresponding to the field or item with the more answers is. The shorter the user's answer time to a question is, the clearer the user's preference related to the evaluation point of view is, so the evaluation point of view determination unit 53 may determine that the evaluation point of view is relatively more important. The comment generation unit 56 generates a comment according to the determined evaluation point of view.
[0058] The model storage unit 68 stores in advance a prediction model that predicts the evaluation of the product or service (hereinafter referred to as "predicted evaluation value") based on the specification information of the product or service and the prior information of each user. The prediction model is a pre-learned model for predicting the evaluation value based on the specification information of the product or service and the prior information of each user. The preference determination unit 55 generates a predicted evaluation value using the prediction model based on the specification information of the product or service and the prior information of each user. The predicted evaluation value may include parameters of a plurality of evaluation viewpoints described later, or a weight corresponding to the preference of each user may be given to a plurality of evaluation viewpoints. The preference determination unit 55 may also determine the user's preference as information for input to the prediction model based on the user's prior information. The preference determination unit 55 may also determine the user's preference based on the user's answer results in a plurality of evaluation viewpoints in a simple test such as a perception survey. The preference determination unit 55 may determine that the preference for a field or item is higher when the user answers more to a field or item that the user is particularly concerned about among a large number of items presented as the inquiry information. The preference determination unit 55 may determine that the preference level for the field or item is higher on the premise that the shorter the answer time to one inquiry, the clearer the user's preference related to the evaluation point of view.
[0059] The model storage unit 68 further stores a language model for generating comments based on the specification information of the product or service, the user's prior information, and the predicted evaluation value. The language model may also be a model in which the large-scale language model of the language model server 90 connected via the network 80 is fine-tuned by the learning processing unit 59. The learning processing unit 59 may also perform a process of fine-tuning the language model using a data set of prior information of each user. The learning processing unit 59 may also perform a process of fine-tuning the language model using feedback information described later. The learning processing unit 59 may also perform re-learning on the prediction model using feedback information described later.
[0060] In addition, in the present embodiment, the following example is described, that is, after the preference determination unit 55 generates a predicted evaluation value using the prediction model, the predicted evaluation value and the language model are used to generate a comment. In a modified example, the prediction model itself may be pre-constructed in a form including a language model, and the prediction model may be learned in advance so that the prediction model directly generates a predicted evaluation value and a comment based on the specification information of the product or service and the prior information of the user. For example, in the case where the prior information includes a large amount of a data set of past comments of multiple people, the language model may be fine-tuned using such a large amount of comments, and a prediction model that generates comments based on the specification information and the prior information may be pre-constructed.
[0061] The information input to the prediction model or language model includes not only specification information of the product, attribute information or preference information of the user, and past comments, but also the result of a simple test such as a perception survey conducted on the user when generating the comment, that is, the answer information. The simple test such as the perception survey is configured to easily grasp the information that is important for exploring the user's preferences, that is, the information indicating which of the multiple evaluation viewpoints the user particularly values. That is, in order to grasp the characteristics of the preferences from the state where there is a lack of information related to the user's preferences, multiple simple inquiry items are divided and correspond to multiple evaluation viewpoints. For example, an inquiry that asks the user to select a favorite color tone after presenting multiple images with greatly different color tone tendencies is an inquiry corresponding to the appearance evaluation viewpoint. For example, when the user selects an image with blue as the base tone, the preference determination unit 55 determines that the user's favorite color is blue. For example, for a user who answers many questions related to the color tone, it is determined that the appearance evaluation viewpoint is relatively important to the user. That is, the inquiry information included in the perception survey is configured to correspond to multiple evaluation viewpoints, and the information indicating the user's preferences for each evaluation viewpoint and the information indicating the type of evaluation viewpoint that is particularly important are included in the user's answer information to the inquiry. For example, the evaluation viewpoint determination unit 53 may also infer which evaluation viewpoint the user particularly values based on the items that the user arbitrarily answers or the items that the user answers with emphasis in the multiple inquiries corresponding to the multiple evaluation viewpoints. Alternatively, the evaluation viewpoint determination unit 53 may also infer which evaluation viewpoint the user values based on how short the time required for the user to answer is or how quickly the user answers. Conversely, the evaluation viewpoint determination unit 53 may also infer which evaluation viewpoint the user does not value based on the items that the user's answer rate is low or the items that the user has not answered at all in the multiple inquiries corresponding to the multiple evaluation viewpoints.
[0062] The user's answer history to the inquiry is accumulated in the prior information storage unit 64. The preference determination unit 55 and the evaluation point of view determination unit 53 can reuse the user's answer content contained in the past answer history accumulated in the prior information storage unit 64 and refer to it in determining the user's preference or the evaluation point of view, or add the information of the past inquiry and answer to the user's answer to the new inquiry information and refer to it in determining the user's preference or the evaluation point of view.
[0063] The review output unit 58 sends the review generated by the review generating unit 56 to the user terminal 10, thereby presenting the review to the user. The review output unit 58 may also present the predicted evaluation value generated by the review generating unit 56 to the user in the form of the number of stars or a radar chart, etc., together with the review.
[0064] The feedback acquisition unit 70 acquires feedback information input by the user after the user who was prompted with the comment actually enjoyed the product or service. For example, it can be a numerical value that scores how close the comment prompted to the user is to the feelings of the user who actually enjoyed the product or service, or it can be a comment written by the user based on the actual feelings. In the case where the feedback information is a comment actually written by the user, the comment can also be stored in the prior information storage unit 64 as a new data set used in learning. The comment generation unit 56 can also regenerate the comment based on the feedback information. In addition, the feedback information can also be used as a data set for relearning the prediction model or language model. By using the feedback information for relearning, the feature quantities such as the accuracy improvement of the evaluation viewpoint that is valued, the wording or length of the comment quality, etc. are collected through analysis when the comment is generated next time, so that the comment generation that reflects personal characteristics can be realized. The comment output unit 58 can also send the regenerated comment to the user terminal 10, thereby prompting it to the user again.
[0065] Figure 3 The basic process of generating comments is shown. First, when a user selects a product or service as a review object before enjoying it (symbol 110), a question corresponding to the selected product or service is prompted to the user in the form of a simple test such as a perception survey (symbol 112). When the user answers the question, the information of the answer is included in the prior information (symbol 114). The prior information related to the user is sent to the evaluation unit 57, and based on the prior information, a comment is generated using a prescribed prediction model and language model (symbol 116). The generated comment is output to the user terminal 10 by the comment output unit 58 (symbol 118).
[0066] After enjoying the product or service, the user sends feedback information based on the actual experience (symbol 120). The learning processing unit 59 relearns the prediction model or language model based on the feedback information. The relearned prediction model or language model can also be used to regenerate comments (symbol 116), and the comment output unit 58 re-outputs the regenerated comments to the user terminal 10 (symbol 118).
[0067] Figure 4 An example of a simple test screen for prompting an inquiry when the appearance is set as an evaluation point of view is shown. In the simple test screen 200 of the user terminal 10, a pair of product images with contrasting colors or appearances are prompted, and a query is prompted to ask which appearance is closer to the user's preference. In the present embodiment, as an inquiry when sports shoes are selected as the review object, a pair of shoe images with contrasting colors or appearances are prompted. Such a pair of product images is prompted multiple times, and the more the user answers, the more the tendency of the appearance of the shoes that the user likes can be understood. Such answer information is stored as prior information in the prior information storage unit 64.
[0068] Figure 5 An example of a simple test screen that prompts an inquiry when the function is set as an evaluation point of view. In the simple test screen 200 of the user terminal 10, items related to the function of the shoe, such as "preventing injury", "performance", and "breathability", are prompted, and an inquiry word is prompted to inquire about the importance of each item to the user. The more the user answers the importance of each item, the more the user's tendency to value which function can be understood. This answer information is stored as advance information in the advance information storage unit 64.
[0069] In addition, the object prompting unit 54 may also prompt the user with inquiries related to the user's feelings, expectations, reactions from third parties, price and other evaluation viewpoints. As an inquiry related to the user's feelings, for example, it may be a question that asks the user to evaluate and answer the degree of subjective emotions such as a sense of security or comfort after being prompted with multiple images. As an inquiry related to the user's expectations, for example, it may be a question that asks the user to answer the expectations of the image with a numerical value after being prompted with multiple images. The reaction from a third party refers to an objective evaluation from a third party. For example, as a method of reflecting the opinions of a celebrity (influencer) that one likes or the content of a review that one agrees with, it may be a question that asks the user to select a comment that is close to the user's own comment after being prompted with various comment examples. As an inquiry related to price, for example, it may be a question that asks the user to select the price that the user imagines for the image after being prompted with multiple images, or a question that asks the user to select a price that is close to the user's own monetary values.
[0070] Figure 6 An example of a simple test screen for asking questions for understanding the user's expression tendency is shown. In the simple test screen 200 of the user terminal 10, a plurality of comment examples of different writing styles or intonations are shown, and the user is prompted to select an inquiry phrase that is close to the writing style or intonation that the user likes. The writing style or intonation of the user can be estimated based on the user's selection. Such answer information is stored as advance information in the advance information storage unit 64.
[0071] Furthermore, after the user has answered the choice of the comment example in response to the first question, the user's preferences or important evaluation viewpoints may be estimated based on the content of the answer, and the object presentation unit 54 may determine which option should be presented to the user as the comment example for the second question and thereafter based on the estimation result.
[0072] Figure 7An example of a screen showing candidates for review objects and prompts for comments and predicted evaluation values. In screen 201, three products are presented to the user as selection candidates 210. In the example shown, three types of shoes, "Type A", "Type B", and "Type C", are presented. Furthermore, in order to show the functional differences among the three types of shoes, the specifications of each product stored in the pre-specification storage unit 62 are presented in the form of a radar chart 211. For example, at the beginning, any product becomes a selected state, and corresponding to the product in the selected state, the comments 213 and predicted evaluation values 212 generated by the review generation unit 56 are displayed in the comment column 214. If the user selects another product, the comments 213 and predicted evaluation values 212 corresponding to the product in the selected state are displayed in the comment column 214.
[0073] (Second Embodiment)
[0074] The review prompting system 100 in the second embodiment differs from the review prompting system 100 in the first embodiment in that the function of acquiring feedback information on the generated comments and regenerating the comments is removed. The following description will focus on the differences from the first embodiment and omit the description of the common points.
[0075] Figure 8 The basic process of generating comments in the second embodiment is shown. First, when a user selects a product or service as a review object (symbol 110), a query corresponding to the selected product or service is prompted to the user in the form of a simple test (symbol 112). When the user answers the query, the information of the answer is included in the prior information (symbol 114). The stored prior information related to the user is sent to the comment generation unit 56, and based on the prior information, a comment is generated using a prescribed prediction model and language model (symbol 116). The generated comment is output to the user terminal 10 by the comment output unit 58 (symbol 118).
[0076] In the second embodiment, after enjoying the product or service, the user does not provide feedback to evaluate the review based on the actual experience. However, even in this case, the review as if made by the user can be accurately generated based on the prior information of the implementation of the simple test, and presented to the user before enjoying the product or service.
[0077] (Third Embodiment)
[0078] The difference between the review prompting system 100 in the third embodiment and the review prompting system 100 in the first and second embodiments is that the function of implementing the perception survey for asking questions to the user to obtain answers is removed. The following description will focus on the differences from the first and second embodiments, and the description of the common points will be omitted.
[0079] Fig. 9 The basic process of generating a comment in the third embodiment is shown. First, when a user selects a product or service as a comment object (symbol 110), the stored information related to the user is sent to the comment generation unit 56, and based on the information, a comment is generated using a predetermined prediction model and language model (symbol 116). The generated comment is output to the user terminal 10 by the comment output unit 58 (symbol 118).
[0080] In the third embodiment, after enjoying the product or service, the user does not provide feedback to evaluate the review based on the user's actual experience. Moreover, a simple test is not performed before the review is generated. However, even in this case, the review written by the user can be generated with good accuracy through the pre-stored prior information, and is presented to the user before enjoying the product or service.
[0081] (Fourth Embodiment)
[0082] The difference between the review prompt system 100 in the fourth embodiment and the review prompt system 100 in the first to third embodiments is that only the results of a simple test such as a feeling survey on the review object are used as the prior information, and the review is generated without using prior information other than the results of the simple test. That is, the difference from the review prompt system 100 in the first to third embodiments is that the first information in the prior information is not used. The following description will focus on the differences from the first to third embodiments, and the description of the common points will be omitted.
[0083] Fig.10 The basic process of generating comments in the fourth embodiment is represented. First, when the user selects a product or service as the subject of the comment (symbol 110), the inquiry corresponding to the selected product or service is prompted to the user in the form of a simple test (symbol 112). When the user answers the inquiry, the information of the answer is sent to the comment generation unit 56, and based on the information, the specified prediction model and language model are used to generate comments (symbol 116). The generated comments are output to the user terminal 10 by the comment output unit 58 (symbol 118). In this embodiment, the evaluation point of view is determined and the comments are generated based on the results of simple tests such as perception surveys, rather than being affected by the basic information of the user. In this way, the hobbies or states of the time when the comments are to be referred to can be strongly reflected in the comments.
[0084] (Fifth Embodiment)
[0085] The review prompting system 100 in the fifth embodiment is different from the review prompting system 100 in the first to fourth embodiments in that a plurality of review objects are selected to generate comments. The following description will focus on the differences from the first to fourth embodiments, and description of the common points will be omitted.
[0086] In the fifth embodiment, the user selects multiple review objects. In this case, the result is that multiple items are reflected in the generation of the comment. As an example, imagine a situation where a marathon and accommodation facilities are selected as review objects. In the perception survey, a complex preference inquiry is prompted, such as what are the expected evaluation opinions of the facilities to be accommodated, based on the premise of participating in the marathon, and a comment is generated based on the answers to the inquiry. In addition, a perception survey can be implemented for each of the multiple review objects, and the preference weight of the review object can be determined based on the number of answers or answer tendencies, and the weight can be reflected when the comment is generated.
[0087] The above is an explanation of the embodiment. Those skilled in the art should understand that the present embodiment is for illustration only, and the combination of the above-mentioned components or the above-mentioned processing procedures may have various variations, and such variations are also within the scope. Moreover, if the above-mentioned embodiment is generalized, the following forms are obtained.
[0088] (1) A comment prompting device according to an embodiment of the present disclosure includes:
[0089] A specification storage unit for storing specification information related to each of the plurality of review objects;
[0090] An object prompting unit prompts the user that at least one of the plurality of review objects is a review object before the user enjoys the review;
[0091] An information storage unit stores information related to the user;
[0092] a review generating unit that generates a review similar to that made by the user using a prediction model based on the specification information of the presented review object and information related to the user; and
[0093] The comment output unit presents the generated comment to the user.
[0094] According to the review presenting device of (1) above, it is possible to provide a technique capable of presenting a review as if written by the user himself or herself, regarding something that the user has not experienced.
[0095] (2) The comment prompting device according to (1) may also include:
[0096] a query storage unit for storing query information, wherein the query information includes queries for the perception survey of each of the plurality of review objects;
[0097] The object presenting unit presents the inquiry information to a user who discusses the enjoyment of the review object, and acquires an answer of the user to the inquiry information as information related to the user.
[0098] (3) The comment prompting device according to (2), wherein:
[0099] The evaluation unit determines the user's preference based on the user's answer to the inquiry information.
[0100] (4) The comment prompting device according to (2) or (3), wherein:
[0101] The query information includes a query for determining which evaluation viewpoint among a plurality of evaluation viewpoints is used to generate a review that suits the user's preference.
[0102] The evaluation unit determines an evaluation point of view of the comment based on an answer of the user to the inquiry information, and generates the comment according to the determined evaluation point of view.
[0103] (5) The comment presentation device according to any one of (1) to (4), wherein:
[0104] The information related to the user includes at least one of attribute information, preference information, and past comments of the user.
[0105] (6) The comment prompting device according to any one of (1) to (5) may further include:
[0106] a feedback acquisition unit, which acquires feedback information input by the user after the user who is prompted with the comment actually enjoys the experience;
[0107] The evaluation unit relearns the prediction model for generating the comment based on the feedback information.
[0108] (7) A comment prompting method according to an embodiment of the present disclosure includes the following steps:
[0109] Prompting the user at least one review object among the plurality of review objects that the user has not yet enjoyed;
[0110] Reading out specification information related to the prompted review object;
[0111] reading out information related to the user;
[0112] generating a review as if made by the user using a prediction model based on the specification information of the provided review object and information related to the user; and
[0113] The generated comments are prompted to the user.
Claims
1. A comment prompting device, characterized in that: include: a specification storage unit for storing specification information related to each of the plurality of review objects; An object prompting unit prompts the user that at least one of the plurality of review objects is a review object before the user enjoys the review; An information storage unit, storing information related to the user; An evaluation unit that generates a review similar to that made by the user using a prediction model based on the specification information of the presented review object and information related to the user; as well as The comment output unit presents the generated comment to the user.
2. The comment prompting device according to claim 1, characterized in that: Also includes: a query storage unit for storing query information, wherein the query information includes queries for the perception survey of each of the plurality of review objects; The object presenting unit presents the inquiry information to a user who discusses the enjoyment of the review object, and acquires an answer of the user to the inquiry information as information related to the user.
3. The comment prompting device according to claim 2, characterized in that: The evaluation unit determines the user's preference based on the user's answer to the inquiry information.
4. The comment prompting device according to claim 2, characterized in that: The query information includes a query for determining which evaluation viewpoint among a plurality of evaluation viewpoints is used to generate a review that suits the user's preference. The evaluation unit determines an evaluation point of view of the comment based on an answer of the user to the inquiry information, and generates the comment according to the determined evaluation point of view.
5. The comment prompting device according to claim 1, characterized in that: The information related to the user includes at least one of attribute information, preference information, and past comments of the user.
6. The comment prompting device according to claim 1, characterized in that: Also includes: a feedback acquisition unit, which acquires feedback information input by the user after the user who is prompted with the comment actually enjoys the experience; The evaluation unit relearns the prediction model for generating the comment based on the feedback information.
7. A comment prompt method, characterized in that: The process includes: Prompting the user at least one of the multiple review objects before the user enjoys the review; Reading out specification information related to the prompted review object; reading out information related to the user; generating a review as if made by the user using a prediction model based on the specification information of the review object and information related to the user; as well as The generated comments are prompted to the user.