Server device, reply content generation method, and program

Abstract

We provide technology to assist with responding to customer reviews. [Solution] The server device includes means for outputting a reply generation operation unit that outputs multiple reviews for a transaction target and receives instructions for generating a reply corresponding to each review; means for presenting options for a reply policy for the review corresponding to the reply generation operation unit when the reply generation operation unit receives an instruction; means for acquiring the reply policy selected from the options; means for inputting the selected review and an instruction to a large-scale language model that instructs the model to respond to the review according to the acquired reply policy, and for generating a reply to the review based on the output of the large-scale language model; and means for outputting the generated reply to the review.

Description

Technical Field

[0001] The present invention relates to a server device, a method for generating reply content, and a program.

Background Art

[0002] In media such as web pages that introduce products etc. on an EC (Electronic Commerce) site, it is often provided with a function for users who have purchased a product to write a review of that product. Reviews of products by consumers can enhance the reliability of the products and the advertising effect can be expected. Patent Document 1 discloses a system that collects and analyzes reviews related to a product when a user orders the product etc., and transmits the analyzed data to the provider of the product etc. According to this system, a user who has purchased a product can confirm the reviews posted in the past from their my page and the replies to the reviews. Replies from the store side to consumers' reviews lead to an improvement in the reliability and favorability of the store. However, it is required to reply to reviews quickly and appropriately, and this reply work is a burden for the store side.

Prior Art Documents

Patent Documents

[0003] <00000​​​​​​​​​​​​​​​​​​​​According to one aspect of the present invention, the server device includes means for outputting a plurality of reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews; means for presenting options for a reply policy to the review corresponding to the reply generation operation unit when the reply generation operation unit receives the instructions; means for acquiring the reply policy selected from the options; means for inputting the reviews and an instruction to a large-scale language model to provide a reply to the reviews in accordance with the acquired reply policy, and generating a reply to the reviews based on the output of the large-scale language model; and means for outputting the generated reply to the reviews. The generation means includes means for obtaining the purchase history of the transaction item by the end user who posted the review, and if the transaction item related to the review is included multiple times in the purchase history, the generation means inputs the review and an instruction to the large language model that instructs the model to send a reply including an expression of gratitude for the multiple purchases of the transaction item related to the review, and generates a reply to the review based on the output of the large language model. Furthermore, according to one aspect of the present invention, if the store where the transaction item related to the review was purchased is included in the usage history multiple times, the generation means may input the review and an instruction to the large language model, instructing it to send a reply that includes an expression of gratitude for using the store related to the review multiple times, and generate a reply to the review based on the output of the large language model. Furthermore, according to one aspect of the present invention, if the review history includes multiple reviews by the end user for the store or transaction subject to the review, the generation means may input the review and an instruction to the large language model, instructing it to send a reply that includes an acknowledgment of the end user's multiple reviews, and generate a reply to the review based on the output of the large language model. Furthermore, according to one aspect of the present invention, the generation means may input the review and an instruction to the large-scale language model, which instructs the model to provide a reply including the review and information introducing similar trading partners, and generate a reply to the review based on the output of the large-scale language model. Furthermore, according to one aspect of the present invention, the generation means may input the review and an instruction sentence instructing the large-scale language model to send a reply that includes a description of the content of past reviews, and generate a reply sentence to the review based on the output of the large-scale language model. Furthermore, according to one aspect of the present invention, if the review exceeds a predetermined number of characters, the generation means inputs the review and an instruction to the large-scale language model, instructing it to provide a reply to the review that includes a sentence expressing gratitude for the detailed review, and generates a reply to the review based on the output of the large-scale language model. Furthermore, according to one aspect of the present invention, the server device includes means for outputting a plurality of reviews for a transaction target and outputting a reply text generation operation unit that receives instructions for generating a reply text in association with each of the reviews; presentation means for outputting a reply text generation instruction screen that, when the reply text generation operation unit receives the instruction, presents options for reply policies for the review corresponding to the reply text generation operation unit and receives instructions for generating a reply text based on the reply policy indicated by the selected option; acquisition means for acquiring the reply policy selected from the options; generation means for inputting the review and an instruction to a large-scale language model that instructs the model to respond to the review in accordance with the acquired reply policy, and generating a reply text to the review based on the output of the large-scale language model; and means for outputting the generated reply text to the review, wherein the reply text generation instruction screen may be configured to present a first option indicating a reply policy that automatically generates the reply content and a second option indicating a reply policy that accepts instructions for the reply content, and if the second option is selected, an input field for the instructions content may be further output.

[0006] According to one aspect of the present invention, the reply content generation method involves a computer outputting multiple reviews for a transaction target, outputting a reply text generation operation unit that receives instructions for generating a reply text in association with each of the reviews, presenting a selection of reply policies for the corresponding review to the reply text generation operation unit upon receiving the instructions, obtaining the selected reply policy from the selection, inputting the review and an instruction to a large-scale language model to provide a reply to the review in accordance with the obtained reply policy, generating a reply text to the review based on the output of the large-scale language model, and outputting the generated reply text. Furthermore, when generating a reply to the review, the purchase history of the transaction item by the end user who posted the review is obtained, and if the transaction item related to the review is included multiple times in the purchase history, the large-scale language model is input with the review and an instruction to send a reply that includes an expression of gratitude for the multiple purchases of the transaction item related to the review, and the reply to the review is generated based on the output of the large-scale language model. Furthermore, according to one aspect of the present invention, the reply content generation method is configured such that a computer outputs a plurality of reviews for a transaction target, outputs a reply content generation operation unit that receives instructions for generating a reply text in association with each of the reviews, and when the reply content generation operation unit receives the instructions, it presents options for reply policies for the review corresponding to the reply content generation operation unit, and outputs a reply content generation instruction screen that receives instructions for generating a reply text based on the reply policy indicated by the selected option, obtains the reply policy selected from the options, inputs the review and an instruction text that instructs the large-scale language model to reply to the review in accordance with the obtained reply policy, generates a reply text to the review based on the output of the large-scale language model, outputs the generated reply text, and the reply content generation instruction screen presents a first option indicating a reply policy that automatically generates reply content and a second option indicating a reply policy that accepts instructions for the reply content, and if the second option is selected, it further outputs an input field for the instructions.

[0007] According to one aspect of the present invention, the program provides a computer with means for outputting a reply generation operation unit that outputs a plurality of reviews for a transaction subject and receives instructions for generating a reply corresponding to each of the reviews; means for presenting a selection of reply policies for the review corresponding to the reply generation operation unit when the reply generation operation unit receives the instructions; means for obtaining the reply policy selected from the selection; means for inputting the reviews and instructions to a large language model to provide a reply to the reviews in accordance with the obtained reply policy, and for generating a reply to the reviews based on the output of the large language model; and means for outputting the generated reply. A means of obtaining the purchase history of the transaction item by the end user who posted the aforementioned review, Functions as Furthermore, if the transaction item related to the review is included multiple times in the purchase history, the generating means inputs the review and an instruction to the large language model that instructs it to send a reply including a thank you for purchasing the transaction item related to the review multiple times, and generates a reply to the review based on the output of the large language model. Furthermore, according to one aspect of the present invention, the program may function as a means for outputting a computer that outputs a plurality of reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews; a means for outputting a reply generation instruction screen that, when the reply generation operation unit receives the instruction, presents options for reply policies for the review corresponding to the reply generation operation unit and receives instructions for generating a reply based on the reply policy indicated by the selected option; a means for acquiring the reply policy selected from the options; a means for inputting the reviews and instructions to a large language model to respond to the reviews in accordance with the acquired reply policy, and generating a reply to the reviews based on the output of the large language model; and a means for outputting the generated reply. The reply generation instruction screen may be configured to present a first option indicating a reply policy that automatically generates the reply content and a second option indicating a reply policy that accepts instructions for the reply content, and if the second option is selected, an input field for the instructions content may be further output. [Effects of the Invention]

[0008] According to the present invention, it is possible to support the process of responding to reviews posted by consumers. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows an example of a review response system according to the embodiment. [Figure 2] This figure shows an example of a product purchase screen according to an embodiment. [Figure 3] This figure shows an example of a review check screen according to the embodiment. [Figure 4] This figure shows an example of a review management table according to the embodiment. [Figure 5] This figure shows an example of a purchase history table according to the embodiment. [Figure 6] This figure shows an example of a product information table according to the embodiment. [Figure 7] Figure 1 shows an example of a reply text generation instruction screen according to the embodiment. [Figure 8] Figure 2 shows an example of a reply text generation instruction screen according to the embodiment. [Figure 9]It is FIG. 3 showing an example of a reply text generation instruction screen according to an embodiment. [Figure 10] It is a diagram showing an example of an edit screen of a reply text according to an embodiment. [Figure 11] It is a flowchart showing an example of a generation process of a reply text according to an embodiment. [Figure 12] It is a diagram showing an example of a hardware configuration of a review reply system according to an embodiment. [[ID=1's]]

Embodiments for Carrying Out the Invention

[0010] <Embodiment> Hereinafter, the review reply system 1 according to the present disclosure will be described with reference to the drawings. The review reply system 1 enables a reply to a review posted by a consumer on an EC site or the like with a simple operation.

[0011] (System Configuration) The review reply system 1 includes a user terminal 10, a store terminal 20, and a server terminal 30. The user terminal 10, the store terminal 20, and the server terminal 30 are communicably connected via a network NW. In FIG. 1, one user terminal 10, one store terminal 20, and one server terminal 30 are illustrated, but the review reply system 1 may include a plurality of user terminals 10, a plurality of store terminals 20, and a plurality of server terminals 30.

[0012] The user terminal 10 is, for example, a mobile terminal such as a smartphone used by a user, a tablet terminal, a PC (personal computer), or the like. A user is a consumer who purchases a transaction target such as a product or a service on an EC site. Products include general items such as clothing, food, and home appliances, as well as digital content, event tickets, and the like. Services include reservation services for accommodation facilities and golf courses. The user posts a review of a product or the like using the user terminal 10.

[0013] The store terminal 20 is a mobile device such as a smartphone, tablet, or PC used by the staff of a store that operates an e-commerce site and sells products. The store staff uses the store terminal 20 to post replies to reviews.

[0014] The server terminal 30 is a PC or a server terminal. The server terminal 30 comprises an acquisition unit 31, a control unit 32, a reply text generation unit 33, a large-scale language model 34, a storage unit 35, and a communication unit 36.

[0015] The acquisition unit 31 acquires data transmitted from the user terminal 10 and the store terminal 20. For example, the acquisition unit 31 acquires transaction items such as products and reviews of stores transmitted from the user terminal 10 and records the acquired reviews in the storage unit 35. The acquisition unit 31 also acquires reply messages transmitted from the store terminal 20 and records the acquired reply messages in the storage unit 35.

[0016] The control unit 32 generates replies to reviews posted by users, controls the posting process, and generates web pages and screens to be displayed on the user terminal 10 and the store terminal 20. For example, the control unit 32 generates a product purchase screen 100 (Figure 2) and displays the reviews posted by the user. For example, the control unit 32 generates a review check screen 200 (Figure 3) where a store employee can check the reviews posted by users, and outputs the selected reviews on the review check screen to the reply generation unit 33, which will be described next. The control unit 32 also displays the replies generated by the reply generation unit 33 in association with the reviews on the review check screen 200 and the product purchase screen 100. The control unit 32 may also have the functionality of an e-commerce server. For example, the control unit 32 may perform the purchase process for products purchased on the product purchase screen 100 in response to a request from the user terminal 10.

[0017] The reply generation unit 33 generates a reply to a user's review in accordance with the instructions of the store staff. The reply generation unit 33 has a Large Language Model (LLM) 34. The Large Language Model 34 is a natural language processing model that has been trained on a vast amount of text data. When a sentence called a prompt is input to the LLM, the LLM analyzes the input sentence and responds to the prompt. The reply generation unit 33 generates an instruction sentence that instructs the generation of a reply to the review, and inputs the generated instruction sentence and a prompt containing the text data of the posted review to the Large Language Model 34. The Large Language Model 34 analyzes the input prompt and generates a reply to the input review. The reply generation unit 33 retrieves the reply generated by the Large Language Model 34 and records it in the storage unit 35.

[0018] The storage unit 35 stores various types of data. For example, the storage unit 35 stores a review management table 301 (Figure 4) which registers user reviews and replies to reviews, a purchase history table 302 (Figure 5) which registers users' purchase history, and a product information table 303 (Figure 6) which registers product names, prices, etc. The communication unit 36 ​​communicates data with the user terminal 10 and the store terminal 20. The acquisition unit 31 acquires data from the user terminal 10 and the store terminal 20 through the communication unit 36. The control unit 32 outputs web pages, screens (images), etc., to the user terminal 10 and the store terminal 20 through the communication unit 36.

[0019] In the configuration example in Figure 1, the review management table 301, purchase history table 302, and product information table 303 are stored in the storage unit 35 of the server terminal 30. However, these databases may be stored in an external storage device or stored and managed in another system (e.g., an e-commerce server). Furthermore, these tables 301 to 303 may be distributed across separate servers or storage devices. When tables 301 to 303 are located in an external storage device or on another server, data is registered to and read from tables 301 to 303 based on instructions from the server terminal 30. In addition, in the configuration example in Figure 1, the server terminal 30 is assumed to have a large-scale language model 34. However, the large-scale language model 34 may be provided by an external server (one or more servers). If the external server has a large-scale language model 34, the reply generation unit 33 sends a prompt to the large-scale language model 34 on the external server and retrieves the reply generated by the large-scale language model 34 from the external server.

[0020] Figure 2 shows an example of a product purchase screen 100. The product purchase screen 100 includes a product introduction area 101 that displays the product image, product name, description of the product contents and specifications, and price, a purchase procedure area 102 used by the user when purchasing a product, and a review section 103. The review section 103 includes review display areas 103a and 103b that display reviews posted by users, and a reply display area 103c that displays replies from store staff to reviews. In the example in Figure 2, no replies have been posted to the reviews posted in review display area 103a, while replies have been posted in reply display area 103c to the reviews in review display area 103b. In the example in Figure 2, two reviews are displayed in review area 103, but the number of reviews displayed may be three or more. In this example, reviews are displayed on the product purchase screen 100, but reviews may be configured to be displayed on other screens. For example, a screen that displays only reviews may be provided for each product.

[0021] The control unit 32 reads product images, names, etc., from the product information table 303 (Figure 6) and displays them in the product introduction area 101. It also reads reviews and replies from the review management table 301 (Figure 4) and displays them in the review section 103. The control unit 32 transmits the generated product purchase screen 100 to the user terminal 10 via the communication unit 36. The user can purchase products by referring to the reviews displayed on the product purchase screen 100. The user can also post reviews for purchased products or stores from a review posting screen (not shown). Posted reviews are registered in the review management table 301 by the control unit 32 and displayed in the review section 103 by the control unit 32.

[0022] Figure 3 shows an example of the review check screen 200. The control unit 32 reads user reviews registered in the review management table 301 (Figure 4), generates the review check screen 200 shown in Figure 3, and transmits it to the store terminal 20 via the communication unit 36. The store staff can refer to the review check screen 200 to check what reviews have been posted, whether replies have been made to the reviews, what replies have been posted, etc.

[0023] The review check screen 200 includes areas 201, 202, and 203 (203A, 203B). Area 201 includes three tabs 201a to 201c: Show All, Flags Only, and Unaddressed. When the person in charge selects the "Show All" tab 201a, the control unit 32 displays all reviews. When the person in charge selects the "Flags Only" tab 201b, it displays only the reviews for which the person in charge has set a flag in the flag setting field 203d (described later). When the person in charge selects the "Unaddressed" tab 201c, it displays only the reviews for which no reply has been registered.

[0024] Area 202 is an area for entering search criteria to narrow down the reviews displayed in Area 203. Area 202 includes an input field 202a for the posting date and time of the reviews to be displayed, an input field 202b for setting the rating of the reviews to be displayed (all, 1 to 5 stars), an input field 202c for setting the review category (product review, store review, or both), an input field 202d for search keywords, a search button 202e to instruct the search, and a page specification field 202f. Input field 202b allows you to specify any of the ratings from 1 to 5 stars, or all of them. Input field 202c allows you to specify the type of review to display, such as a product review, a store review, or both. When the person in charge enters search criteria in input fields 202a to 202d and presses the search button 202e, the control unit 32 reads the reviews that match the search criteria from the review management table 301 and displays them in Area 203. Furthermore, when the person in charge specifies a page in the page selection field 202f, the control unit 32 displays the review shown on the specified page in area 203.

[0025] Area 203 is the area that displays user reviews and replies from the staff. Each review is displayed in a separate area, such as Area 203A and Area 203B. Area 203A includes a name display field 203a that displays the name of the transaction item purchased by the user, a display field 203b that displays the date and time the review was posted, a display field 203c that displays the order number, a flag setting field 203d, a display field 203e that indicates whether the review is for a product or a store, a display field 203f that displays the user's rating, a review display field 203g that displays the content of the review posted by the user, a reply display field 203h that displays the reply from the store staff, a call button 203i that calls up a dialog screen (Figures 7-9) that instructs the generation of a reply using AI, and a reply button 203j that instructs the posting of the generated reply. The same applies to Area 203B, although the reference numerals in the diagram are omitted. The control unit 32 reads the content to be displayed in display fields 203a to 203h from the review management table 301 and displays it in display fields 203a to 203h.

[0026] Figure 4 shows an example of the review management table 301. As shown in the figure, the review management table 301 has items such as a review ID, which is the identification information of the review; a user ID, which is the identification information of the user who posted the review; a store ID, which is the identification information of the store where the user purchased the product related to the review; a product ID, which is the evaluation by the user; an order number related to the purchase of the product related to the review; a flag; a review category, which indicates whether the review is for a product or a store; the content of the review and the date and time of posting; and the content of the reply sent by the store staff and the date and time of reply. The flag item registers whether or not it was checked in the flag setting field 203d of the review check screen 200. When a user posts a review using the user terminal 10, the control unit 32 acquires the review etc. sent from the user terminal 10 through the acquisition unit 31, assigns a review ID using a predetermined logic, and registers the acquired information in the review ID, user ID, store ID, product ID, evaluation, order number, review category, review content, and posting date and time items of the review management table 301. Furthermore, when a store employee uses the store terminal 20 to set a flag or post a reply from the review check screen 200, the control unit 32 retrieves the review ID and reply sent from the store terminal 20 via the retrieval unit 31 and registers the retrieved information in the flag, reply content, and reply date and time fields of the corresponding review ID record in the review management table 301.

[0027] Figure 5 shows an example of a purchase history table 302. As shown in the figure, the purchase history table 302 has items such as the order number, the product ID of the purchased product, the store ID of the store where the product was purchased, the quantity of the purchased product, the date and time of purchase, and the user ID of the user who purchased the product. When a user purchases a product from the product purchase screen 100, the control unit 32 assigns an order number using a predetermined logic and registers the information in the purchase history table 302.

[0028] Figure 6 shows an example of a product information table 303. As shown in the figure, the product information table 303 has items such as product ID, product name, product description, product price (unit price), product image, group ID, and store ID. The group ID registers the identification information of the group to which the product belongs. For example, similar products or sister products are registered with the same group ID. For example, products are associated with genres, and products belonging to the same genre may be assigned the same group ID and treated as similar products. In the case of beverages, each product may be associated with one of the genres such as "soup," "water / carbonated water," "vegetable / fruit beverage," or "tea / black tea." The store ID registers the store ID of the store that handles the product. The store staff uses the store terminal 20 to input this product information (product ID, product name, product description, product price, product image, etc.) for the products sold in their store. The acquisition unit 31 acquires the entered product information. The control unit 32 associates the product information acquired by the acquisition unit 31 with the store ID of the sending store and registers it in the product information table 303.

[0029] (Generating a reply) Next, we will explain the process of generating a reply message by referring to Figures 7 to 10. Figures 7 and 8 show an example of the reply message generation instruction screen 400. When a store employee selects the call button 203i on the review check screen 200 in Figure 3, the control unit 32 generates the reply message generation instruction screen 400 as illustrated in Figures 7 to 9 and outputs the reply message generation instruction screen 400 to the store terminal 20 via the communication unit 36.

[0030] The reply generation instruction screen 400 shown in Figure 7 is a screen that instructs the generation of a reply to a review related to the call button 203i (Figure 3) selected by the person in charge. For example, if the person in charge selects the call button 203i in area 203A, a reply to the review displayed in the review display field 203g in area 203A will be generated. If the person in charge selects the call button 203i in area 203B, a reply to the review displayed in the review display field 203g in area 203B will be generated. Hereafter, the person in charge may describe the review corresponding to the selected call button 203i as the selected review.

[0031] The reply text generation instruction screen 400 includes a selection display area 401, a generation instruction button 402, a reply text display field 403, and a reflect instruction button 404. The selection display area 401 displays two options: "Leave it to AI" and "Instruct AI". As illustrated in Figure 7, when a store employee selects "Leave it to AI" and presses the generation instruction button 402, the reply text generation unit 33 uses the large-scale language model 34 to generate a reply text for the selected review. The control unit 32 displays the generated reply text in the reply text display field 403 in Figure 7. When the store employee presses the reflect instruction button 404, the control unit 32 displays the generated reply text in the reply text display field 203h of the review check screen 200. On the review check screen 200, the reply text displayed in the reply text display field 203h can be modified. Furthermore, when the employee presses the reply button 203j, the control unit 32 reflects the reply text on the product purchase screen 100. The reflected reply will be displayed in the reply display area 103c in Figure 2.

[0032] Furthermore, when a store employee selects "Give instructions to the AI," the control unit 32 displays an instruction input field 405 in addition to the areas shown in Figure 7, as illustrated in Figure 8. The instruction input field 405 is where information indicating the sentences to be included in the reply is entered. When the employee enters the instructions in the instruction input field 405 and presses the generate instruction button 402, the reply generation unit 33 uses the large-scale language model 34 to generate a reply based on the instructions in the instruction input field 405. The control unit 32 displays the generated reply in the reply display field 403. When the store employee presses the reflect instruction button 404, the control unit 32 reflects the reply on the review check screen 200.

[0033] (Prompt for large-scale language models) Here, we will explain the process when the generation instruction button 402 is pressed. The reply text generation unit 33 has a prompt template to be input to the large-scale language model 34. The template contains background information (settings) and various instructions, and the reply text generation unit 33 generates a prompt by passing input data, etc., to this template as needed.

[0034] The template background includes instructions such as responding appropriately to user reviews as a store representative.

[0035] The template instructions include details such as the character limit for the reply, the avoidance of prohibited words, the avoidance of consecutive sentences containing apologies, the inclusion of a sentence expressing gratitude for the detailed review for reviews exceeding the specified character limit, the inclusion of a sentence based on the instructions entered in the instruction input field 405, and the generation of a reply for the text data of the submitted review.

[0036] If "Leave it to AI" is selected, the reply generation unit 33 inputs the selected review text data and a template as prompts into the large-scale language model 34.

[0037] If "Instruct AI" is selected, the reply generation unit 33 generates an instruction by providing the instruction content from the instruction input field 405 to a template, and inputs the instruction and the selected review text data as prompts to the large-scale language model 34.

[0038] The large-scale language model 34 generates and outputs a response to the review that follows the instructions, taking into account the context included in the prompt.

[0039] For example, suppose the selected review is something like, "It's wider than I expected, and the smell is bothersome." If "Leave it to AI" is selected on the reply generation instruction screen 400, the large-scale language model 34 will generate a reply such as, "Thank you for your review of our product. Your feedback on the size of the product was very helpful. We also appreciate your valuable feedback regarding the smell. Please let us know if the smell does not improve after using it for a while. We will continue to strive to support your comfortable use of the product. Thank you for your continued support."

[0040] Furthermore, suppose that for the same review, "Instruct AI" is selected on the reply generation instruction screen 400, and the instruction input field 405 reads, "If you are concerned about the smell of the product, it may be alleviated by airing it out in a well-ventilated place for a day." In this case, the large-scale language model 34 generates a reply such as, "Thank you for your review of our product. Your feedback on the size of the product was very helpful. Regarding the smell, it may be alleviated by airing it out in a well-ventilated place for a day, so please try that. We will continue to strive to provide products that you will be satisfied with, so we look forward to your continued patronage." The reply contains text that conforms to the instructions specified in the instruction input field 405.

[0041] Another example is when the selected review says, "It's great that the carefully selected, popular products are constantly being replaced and curated." If "Leave it to AI" is selected on the reply generation instruction screen 400, the large-scale language model 34 will generate a reply such as, "Thank you for your valuable feedback. We are very pleased to hear that you are enjoying our carefully selected products. We will continue to carefully select and offer popular products to ensure your satisfaction. We look forward to your continued patronage."

[0042] Furthermore, suppose that for the same review, "Instruct AI" is selected, and the instruction input field 405 reads, "Thank you for your purchase. We always strive to provide carefully selected products, so we appreciate your continued patronage." In this case, the large-scale language model 34 generates a reply such as, "Thank you for your purchase. We are very pleased that you liked our carefully selected popular product. We always strive to provide carefully selected products, so we appreciate your continued patronage. We will continue to work hard to deliver high-quality products to satisfy our customers. We appreciate your continued support." The reply contains text that conforms to the instructions specified in the instruction input field 405.

[0043] The template includes instructions to avoid consecutive sentences containing apologies, so the reply generated by the large-scale language model 34 will not include sentences like "~I'm sorry. ~I'm sorry." Also, the template includes instructions to include a sentence expressing gratitude for the detailed review for reviews exceeding a specified character count, so if the number of characters in the review given as a prompt exceeds the specified character count, the reply generated by the large-scale language model 34 will include a sentence like "Thank you for your detailed review."

[0044] Figure 9 shows another example of a reply generation instruction screen, a reply generation instruction screen 400'. The reply generation instruction screen 400' includes a choice display area 401', a generation instruction button 402, a reply display field 403, and an apply instruction button 404. The choice display area 401' provides three choices: "Say thank you," "Apologize," and "Instruct the AI." The generation instruction button 402, the reply display field 403, and the apply instruction button 404 are as explained in Figure 7.

[0045] When a store employee selects "Send a thank you" and presses the generation instruction button 402, the reply generation unit 33 generates an instruction by providing text data instructing the system to include a thank you in the reply (for example, "Please include a thank you in your reply") to the template mentioned above, and inputs this instruction, along with the text data of the target review, as a prompt to the large-scale language model 34. The large-scale language model 34 generates a reply that includes a thank you for the selected review.

[0046] Furthermore, when a store employee selects "Apologize" and presses the generation instruction button 402, the reply generation unit 33 provides the template with text data instructing it to include an apology in the reply (for example, "Please include an apology in your reply"), generates an instruction, and inputs it as a prompt to the large-scale language model 34 along with the text data of the target review. The large-scale language model 34 then generates a reply that includes an apology for the selected review.

[0047] The control unit 32 displays the generated reply text in the reply text display field 403. When the store staff member presses the reflect instruction button 404, the control unit 32 reflects the reply text on the review check screen 200.

[0048] For example, suppose the selected review is "It's been over a week since I ordered, but it still hasn't been shipped. This is the first time I've encountered such a store." If "Send a thank you" is selected on the reply generation instruction screen 400', the large-scale language model 34 will generate a reply such as, "Thank you very much for your order. We apologize that we were unable to meet your expectations and that shipping has been delayed. We strive to respond promptly, and we are very disappointed that this has happened. We are currently checking the shipping status and will address the issue as soon as possible, so please wait a little longer. We appreciate you sharing your valuable feedback. We will do our best to ensure your satisfaction in the future, and we look forward to your continued support." Because "Send a thank you" was selected, a sentence such as "We appreciate you sharing your valuable feedback" has been generated.

[0049] For example, if the selected review reads, "I bought the M size for everyday use and liked the look and size, so this time I bought the L size for home and travel. The color is as cute as I expected, and it certainly holds a lot of cosmetics, but unfortunately, it won't close if I put brushes in the pockets on both sides. Also, the brushes are just barely tall enough for the pouch, so even if it closes, the quality of the brush bristles might get damaged. I wish it was just a little bit bigger. I think the product itself is very good," then a reply message like, "Thank you very much for purchasing our product. We are very pleased that you are satisfied with the size and design. However, we sincerely apologize for the inconvenience caused by the insufficient size of the pockets..." is generated. Since "Apologize" was selected, the message "We sincerely apologize for the inconvenience caused by the insufficient size of the pockets" is generated.

[0050] In addition to the above, you may also include instructions in the template to include the purchase history of the user who wrote the review and text relevant to the product being reviewed.

[0051] (Another example of a template 1) The template may include instructions to include a thank-you message based on the purchase history of the user who posted the selected review. For example, the reply generation unit 33 obtains the user ID of the user who posted the selected review, the product ID of the product related to the review, and the store ID of the store where the product was purchased. The reply generation unit 33 searches the purchase history table 302 using the obtained user ID, product ID, and store ID as keys, obtains the number of data entries that have a user ID, product ID, and store ID that match the user ID, product ID, and store ID specified by the keys, and provides that number to the template. The template may include instructions to include a message that, if the number of data entries is multiple or exceeds a predetermined number, expresses gratitude for the multiple or predetermined number of purchases. In response to this instruction, the large-scale language model 34 generates a reply message that includes, for example, a message such as, "Thank you for liking this product." The template may also include instructions to express greater gratitude or joy as the number of purchases increases.

[0052] (Another example of a template 2) The template may include instructions to include a thank-you message based on the user's store usage history after posting the selected review. For example, the reply generation unit 33 obtains the user ID of the user who posted the selected review and the store ID of the store where the product related to the review was purchased. The reply generation unit 33 searches the purchase history table 302 using the obtained user ID and store ID as keys, obtains the number of data entries that match the user ID and store ID specified by the keys, and provides that number to the template. The template may include instructions to include a message that, if the number of data entries is multiple or exceeds a predetermined number, expresses gratitude for the user having used the store multiple times or more than a predetermined number of times. The large-scale language model 34 generates a reply message that includes, for example, a message such as, "Thank you for always using our store." The template may also include instructions to express greater gratitude or joy as the number of uses increases.

[0053] (Another example of the template 3) The template may include instructions to include a thank-you message based on the number of reviews submitted by the user who posted the selected review. For example, the reply generation unit 33 obtains the user ID of the user who posted the selected review and the store ID of the store where the product related to the review was purchased. The reply generation unit 33 searches the review management table 301 using the obtained user ID and store ID as keys, obtains the number of data entries that match the user ID and store ID specified by the keys, and provides that number to the template. The template may include instructions to include a message that, if the number of data entries is multiple or exceeds a predetermined number, expresses gratitude for the multiple or predetermined number of reviews submitted. The large-scale language model 34 generates a reply message that includes, for example, a message such as, "Thank you for your feedback. We are taking it into consideration." The template may also include instructions to express greater gratitude or joy as the number of reviews increases. Additionally, the search function in the review management table 301 may be modified to include a review category (product review or store review). If the number of product reviews exceeds a specified number, the system should be instructed to include a message of gratitude for the numerous product reviews, and if the number of store reviews exceeds a specified number, the system should be instructed to include a message of gratitude for the numerous store reviews.

[0054] (Another template example 4) The template may include instructions to include text introducing other products similar to the product being reviewed. For example, the reply generation unit 33 obtains the product ID of the product related to the selected review and the store ID of the store where the product was purchased. The reply generation unit 33 searches the product information table 303 using the obtained product ID and store ID as keys, extracts data that has a product ID and store ID that match the product ID and store ID specified by the keys, and obtains the group ID to which the key product ID belongs from the extracted data. The reply generation unit 33 searches the product information table 303 using the store ID and the obtained group ID as keys, extracts data that has a store ID and group ID that match the store ID and group ID specified by the keys, and obtains the names of other similar products handled by the same store from the extracted data. The reply generation unit 33 provides the obtained product names to the template. The template includes instructions to include text recommending the given product names. Products are associated with genres, and products belonging to the same genre are considered similar products. For example, a store in the "Soup" genre might carry various soups such as pumpkin soup and potato soup. In this case, when a user posts a review for potato soup, the large-scale language model 34 generates a reply that includes a sentence such as, "Please also try the pumpkin soup."

[0055] (Other template examples 5) The template may include instructions to include ratings from other users regarding the product being reviewed. For example, the reply generation unit 33 obtains the product ID of the product related to the selected review and the store ID of the store where the product was purchased. The reply generation unit 33 searches the review management table 301 using the obtained product ID and store ID as keys, extracts data with product IDs and store IDs that match the product ID and store ID specified by the keys, and obtains the content of the reviews from the extracted data. The reply generation unit 33 inputs the content of the reviews obtained into the large-scale language model 34 and instructs it to analyze, for example, the most frequent rating, the most frequent positive rating, the most frequent negative rating, usage, and user experience. The large-scale language model 34 extracts and outputs the most frequent ratings, etc. The reply generation unit 33 provides the outputted most frequent ratings, etc. as ratings from other users to the template. The template includes instructions to include text introducing the given ratings. For example, if there are many reviews for product A stating that "product A is unbreakable," the large-scale language model 34 will generate a reply that includes a sentence such as, "Other users have also rated product A as highly durable."

[0056] As described above, the reply generated by the reply generation unit 33 is reflected on the product purchase screen 100 and becomes viewable by the user when the reply button 203j is pressed. The reply can be edited and modified even after it has been reflected on the product purchase screen 100. Figure 10 shows the main parts of the review check screen 200. Figure 10(a) shows an example of the area 203 where the reply is reflected after the person in charge presses the reply button 203j. As shown in the figure, below the reply display field 203h, the edit button 203k is displayed instead of the reply button 203j. When the person in charge presses the edit button 203k, they can edit the reply in the reply display field 203h. When the person in charge presses the edit button 203k, the control unit 32 changes to the display mode shown in Figure 10(b). When the person in charge edits the reply in the reply display field 203h and presses the update button 203m, the reply is changed to the content edited by the person in charge. The changes will be reflected on the product purchase screen 100 and the review check screen 200. If the person in charge edits the reply text in the reply text display field 203h and presses the cancel button 203l, the changes made by the person in charge will be canceled.

[0057] (operation) Next, we will explain the process of generating a reply to the review on the server terminal 30. Figure 11 is a flowchart showing an example of the reply text generation process according to the embodiment. The store staff member operates the store terminal 20 to call up the review check screen 200. The control unit 32 searches the review management table 301 using the store ID transmitted from the store terminal 20 as the key, reads the content of the review posted for that store ID, generates the review check screen 200 (step S1), and outputs it to the store terminal 20. The staff member selects the review to reply to and presses the call button 203i corresponding to the selected review.

[0058] When the call button 203i is pressed, the control unit 32 receives an instruction to generate a reply (step S2). For example, the call button 203i is associated with the review ID of the selected review, and the control unit 32 obtains the review ID of the selected review when the call button 203i is pressed. Based on the obtained review ID, the control unit 32 refers to the review management table 301 to obtain the user ID, product ID, store ID, post content, etc. related to the selected review. The control unit 32 outputs this obtained information to the reply generation unit 33.

[0059] When the control unit 32 receives an instruction to generate a reply, it generates a reply generation instruction screen 400 (step S3) and outputs it to the store terminal 20. The store staff member selects either "Leave it to the AI" or "Instruct the AI" from the options as the reply policy for the review. Alternatively, in the case of the reply generation instruction screen 400' in Figure 9, the staff member selects either "Say thank you," "Apologize," or "Instruct the AI." The control unit 32 accepts this selection of reply policy (step S4). If "Instruct the AI" is accepted, the control unit 32 displays an instruction input field 405 on the reply generation instruction screen 400 (Figure 8). The staff member enters the instruction content in the instruction input field 405.

[0060] When the person in charge presses the generation instruction button 402, the control unit 32 receives the instruction to generate a reply message (step S5), outputs the user ID, store ID, product ID, and review content related to the selected review to the reply message generation unit 33, and instructs it to generate a reply message.

[0061] The reply generation unit 33 generates a prompt in accordance with the reply policy (step S6). For example, if "Leave it to AI" is selected as the reply policy, the reply generation unit 33 uses a template to generate an instruction message instructing the store to generate an appropriate reply to the review, subject to constraints such as not including consecutive sentences containing apologies and including a detailed sentence expressing gratitude for the review for reviews exceeding a predetermined number of characters. The generated instruction message and the text data of the selected review are then used as prompts.

[0062] Furthermore, the instructions may include one or more of the following (1) to (5): (1) A thank-you message based on the purchase history of the user who posted the review. (2) A thank-you message based on the number of times the user has used the store. (3) A thank-you message based on the number of reviews the user has written. (4) A message introducing other products similar to the product being reviewed. (5) A message including ratings and other comments from other users regarding the product being reviewed.

[0063] For example, if "Instruct AI" is selected as the reply policy, the reply text generation unit 33 adds to the instruction that it will generate a reply text that includes a sentence based on the instruction content in the instruction input field 405, and then generates a prompt.

[0064] For example, if "Send a thank you" is selected as the reply policy, the reply text generation unit 33 adds an instruction to generate a reply text that includes a thank you message, and then generates a prompt.

[0065] For example, if "apologize" is selected as the reply policy, the reply generation unit 33 adds an instruction to generate a reply that includes an apology, and then generates a prompt.

[0066] The reply generation unit 33 inputs the generated prompt to the large-scale language model 34 (step S7). The large-scale language model 34 receives the prompt, generates a reply, and outputs it (step S8). The reply generation unit 33 retrieves the outputted reply and outputs it to the control unit 32. The control unit 32 displays the reply in the reply display field 403. When the person in charge presses the reflect instruction button 404, the control unit 32 closes the reply generation instruction screen 400 and displays the reply in the reply display field 203h of the review for which reply generation was instructed in step S2 on the review check screen 200 (step S9). The person in charge at the store checks the reply displayed in the reply display field 203h and makes corrections as necessary. When a store employee presses the submit command button 404, the control unit 32 displays the reply text in the reply text display field 103c of the review for which reply text generation was instructed in step S2 on the product purchase screen 100 (step S10).

[0067] (effect) As described above, according to this embodiment, a draft reply can be easily generated by selecting a review to reply to and instructing the system to generate a reply to the selected review. This reduces the effort required by the person in charge to create the reply and shortens the time required for reply work. Furthermore, by utilizing the large-scale language model 34, high-quality replies can be generated quickly. Moreover, instead of simply leaving the generation of the reply to the large-scale language model 34, it is possible to instruct the system on what content should be included in the reply, making it easy to create replies that are appropriate to the situation and the review.

[0068] Figure 12 shows an example of the hardware configuration of a review reply system according to an embodiment. The computer 900 includes a CPU 901, main memory 902, auxiliary memory 903, input / output interface 904, and communication interface 905. The user terminal 10, store terminal 20, and server terminal 30 described above are implemented in the computer 900. The processes described above are stored in the auxiliary memory 903 in the form of a program. The CPU 901 reads the program from the auxiliary memory 903, expands it into the main memory 902, and executes the above processes according to the program. The CPU 901 also allocates a memory area in the main memory 902 according to the program. The CPU 901 also allocates a memory area in the auxiliary memory 903 to store the data being processed according to the program.

[0069] In at least one embodiment, the auxiliary storage device 903 is an example of a non-temporary tangible medium. Other examples of non-temporary tangible media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memory, etc., connected via the input / output interface 904. Furthermore, if this program is distributed to the computer 900 via a communication line, the computer 900 that receives the program may expand it into the main memory 902 and execute the above processing. The program may also be for the purpose of realizing some of the functions described above. Moreover, the program may be a so-called differential file (differential program) that realizes the above-mentioned functions in combination with other programs already stored in the auxiliary storage device 903.

[0070] Furthermore, it is possible to replace the components in the above-described embodiments with well-known components as appropriate, without departing from the spirit of the present invention. Also, the technical scope of this invention is not limited to the above-described embodiments, and various modifications can be made without departing from the spirit of the present invention. The call button 203i is an example of a reply message generation operation unit. Goods and services are examples of items subject to transaction.

[0071] Some or all of the above embodiments may be described as follows, but are not limited to the following:

[0072] (Note 1) The server device includes means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews; means for presenting options for a reply policy to the review corresponding to the reply generation operation unit when the reply generation operation unit receives the instructions; means for acquiring the reply policy selected from the options; means for inputting the reviews and an instruction to a large-scale language model that instructs it to respond to the reviews in accordance with the acquired reply policy, and means for generating a reply to the reviews based on the output of the large-scale language model; and means for outputting the generated reply to the reviews.

[0073] (Note 2) The server device as described in Appendix 1, wherein the options include a reply policy for automatically generating reply content, and when the acquisition means acquires a reply policy for automatically generating reply content, the generation means inputs the review and an instruction sentence instructing the large-scale language model to respond to the review, and generates a response sentence to the review based on the output of the large-scale language model.

[0074] (Note 3) The server device as described in Appendix 1 or Appendix 2, wherein the options include a reply policy that accepts instructions for the reply content, and when the acquisition means acquires a reply policy that accepts instructions for the reply content, the acquisition means further acquires the instructions, and the generation means inputs the review and an instruction sentence that instructs the large-scale language model to make a reply to the review based on the instructions, and generates a reply sentence to the review based on the output of the large-scale language model.

[0075] (Note 4) The server device according to any one of Appendix 1 to Appendix 3, wherein the aforementioned options include an option that instructs to include a thank-you message, and when the acquisition means acquires the option that instructs to include a thank-you message, the generation means inputs the review and an instruction sentence that instructs the large-scale language model to send a reply to the review including a thank-you message, and generates a reply sentence to the review based on the output of the large-scale language model.

[0076] (Note 5) The server device according to any one of Appendix 1 to Appendix 4, wherein the options include an option that instructs to include an apology, and when the acquisition means acquires an option that instructs to include an apology, the generation means inputs the review and an instruction that instructs the large-scale language model to send a reply to the review that includes an apology, and generates a reply to the review based on the output of the large-scale language model.

[0077] (Note 6) The server device according to any one of Appendix 1 to Appendix 5, further comprising means for obtaining the purchase history of the transaction item by the end user who posted the review, wherein if the transaction item related to the review is included multiple times in the purchase history, the generation means inputs an instruction to the large language model, which instructs the model to send a reply including the review and an expression of gratitude for the multiple purchases of the transaction item related to the review, and generates a reply to the review based on the output of the large language model.

[0078] (Note 7) The server device according to any one of Appendix 1 to Appendix 6, further comprising means for obtaining the usage history of stores used by the end user who posted the review when purchasing the transaction item, wherein if the usage history includes multiple stores where the transaction item related to the review was purchased, the generation means inputs an instruction to the large-scale language model instructing it to send a reply including the review and an expression of gratitude for using the store related to the review multiple times, and generates a reply to the review based on the output of the large-scale language model.

[0079] (Note 8) The server device according to any one of Appendix 1 to Appendix 7, further comprising means for obtaining the review history of the end user who posted the review, wherein if the review history includes multiple reviews by the end user for the store or transaction subject related to the review, the generation means inputs the review and an instruction to the large language model instructing it to send a reply including an thanks for the end user's multiple reviews, and generates a reply to the review based on the output of the large language model.

[0080] (Note 9) A server device according to any one of Appendix 1 to Appendix 8, further comprising means for acquiring information on transaction objects similar to the transaction object that was the subject of the review, wherein the generation means inputs an instruction sentence instructing the large-scale language model to provide a reply including the review and content introducing the similar transaction objects, and generates a reply sentence to the review based on the output of the large-scale language model.

[0081] (Note 10) A server device according to any one of Appendix 1 to Appendix 9, further comprising means for obtaining past reviews of the transaction subject subject to the review, wherein the generation means inputs an instruction sentence instructing a large-scale language model to provide a reply including the review and content introducing the content of the past review, and generates a reply sentence to the review based on the output of the large-scale language model.

[0082] (Note 11) The generation means is a server device according to any one of the appendices 1 to 10, which inputs the review and an instruction to a large-scale language model to respond to the review without consecutive apology statements, and generates a response to the review based on the output of the large-scale language model.

[0083] (Note 12) If the review exceeds a predetermined number of characters, the server device according to any one of the appendices 1 to 11 inputs the review and an instruction to a large language model that instructs the model to respond to the review, including a sentence expressing gratitude for the detailed review, and generates a response to the review based on the output of the large language model.

[0084] (Note 13) A method for generating replies to reviews, comprising: a computer outputting multiple reviews for a transaction target, outputting a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, presenting a selection of reply policies for the corresponding review when the reply generation operation unit receives the instructions, obtaining the selected reply policy from the selection, inputting the review and an instruction to a large-scale language model to provide a reply to the review in accordance with the obtained reply policy, generating a reply to the review based on the output of the large-scale language model, and outputting the generated reply.

[0085] (Note 14) A program for causing a computer to function as: a means for outputting a reply generation operation unit that outputs multiple reviews for a transaction target and receives instructions for generating a reply corresponding to each of the reviews; a means for presenting a selection of reply policies for the review corresponding to the reply generation operation unit when the reply generation operation unit receives the instructions; a means for obtaining the reply policy selected from the selection; a means for inputting the reviews and instructions to a large-scale language model to make a reply to the reviews in accordance with the obtained reply policy, and for generating a reply to the reviews based on the output of the large-scale language model; and a means for outputting the generated reply. [Explanation of symbols]

[0086] 1. Review Reply System 10. User terminal 20...Store terminals 30...Server terminals 31...Acquisition part 32.. Control Unit 33... Reply text generation section 34. Large-scale language models 35...Storage section 36. Communications Department 900... Computer 901···CPU 902...Main memory 903...Auxiliary storage device 904... Input / Output Interface 905...Communication Interface

Claims

1. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, A means of obtaining the purchase history of the transaction item by the end user who posted the aforementioned review, Equipped with, The generation means, if the transaction item related to the review is included multiple times in the purchase history, inputs the review and an instruction to the large language model that instructs it to send a reply including a thank you for purchasing the transaction item related to the review multiple times, and generates a reply to the review based on the output of the large language model. Server device.

2. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, A means of obtaining the usage history of the store used by the end user who posted the aforementioned review when purchasing the item in the transaction, Equipped with, The generation means, if the store where the transaction item related to the review was purchased is included multiple times in the usage history, inputs the review and an instruction to the large language model instructing it to send a reply including an expression of gratitude for using the store related to the review multiple times, and generates a reply to the review based on the output of the large language model. Server device.

3. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, A means of obtaining the review history of the end user who posted the aforementioned review, Equipped with, The generation means, when the review history includes multiple reviews by the end user for the store or transaction subject related to the review, inputs the review and an instruction to the large language model instructing it to send a reply including a thank you for the end user's multiple reviews, and generates a reply to the review based on the output of the large language model. Server device.

4. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, A means for obtaining information on transaction targets similar to the transaction target that was the subject of the aforementioned review, Equipped with, The generation means inputs an instruction to the large-scale language model, which instructs the model to respond with the review and content introducing similar trading partners, and generates a response to the review based on the output of the large-scale language model. Server device.

5. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, A means of obtaining past reviews for the transaction subject to the aforementioned review, Equipped with, The generation means inputs the review and an instruction sentence that instructs the large-scale language model to send a reply that includes a description of the content of past reviews, and generates a reply sentence to the review based on the output of the large-scale language model. Server device.

6. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply generation operation unit, A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, Equipped with, If the aforementioned review exceeds the specified number of characters, The generation means inputs the review and an instruction to the large-scale language model, which instructs the model to respond to the review, including a sentence expressing gratitude for the detailed review, and generates a response to the review based on the output of the large-scale language model. Server device.

7. A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply text generation operation unit receives the instruction, the display means presents options for the reply policy to the review corresponding to the reply text generation operation unit, and outputs a reply text generation instruction screen that accepts an instruction to generate a reply text based on the reply policy indicated by the selected option. A means for obtaining the aforementioned reply policy selected from the aforementioned options, A generation means that inputs the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generates a response sentence to the review based on the output of the large-scale language model, A means for outputting a reply to the generated review, Equipped with, The aforementioned reply generation instruction screen presents a first option indicating a reply policy that automatically generates the reply content, and a second option indicating a reply policy that accepts instructions regarding the reply content. If the second option is selected, the system is configured to further output an input field for the instructions. Server device.

8. The aforementioned options include a reply policy that automatically generates the reply content. If the acquisition means acquires a reply policy that automatically generates the reply content, The generation means inputs the review and an instruction sentence instructing the large-scale language model to respond to the review, and generates a response sentence to the review based on the output of the large-scale language model. The server device according to claim 1 or claim 2.

9. The aforementioned options include a response policy that accepts instructions regarding the response. If the acquisition means acquires a reply policy that accepts instructions for the reply content, The acquisition means further acquires the instruction content, The generation means inputs the review and an instruction sentence instructing the large-scale language model to respond to the review based on the instructions, and generates a response sentence to the review based on the output of the large-scale language model. The server device according to claim 1 or claim 2.

10. The aforementioned options include an option that instructs you to include a thank-you message. If the acquisition means acquires an option that instructs to include the thank-you message, The generation means inputs the review and an instruction to the large-scale language model, which instructs the model to send a reply to the review including a thank-you message, and generates a reply to the review based on the output of the large-scale language model. The server device according to claim 1 or claim 2.

11. The aforementioned options include an option that instructs the inclusion of an apology statement. If the acquisition means acquires an option that instructs to include the apology statement, The generation means inputs the review and an instruction to the large-scale language model, which instructs the model to respond to the review, including an apology, and generates a response to the review based on the output of the large-scale language model. The server device according to claim 1 or claim 2.

12. The generation means inputs the review and an instruction to the large-scale language model to respond to the review without consecutive apology statements, and generates a response to the review based on the output of the large-scale language model. The server device according to claim 1 or claim 2.

13. Computers The system outputs multiple reviews for the transaction target, and also outputs a reply generation operation unit that receives instructions to generate a reply corresponding to each of the reviews. When the reply generation operation unit receives the instruction, it presents options for the reply policy to the review corresponding to the reply generation operation unit, The response policy selected from the above options is obtained, The large-scale language model is input with the review and an instruction that tells it to respond to the review in accordance with the acquired response policy, and a response to the review is generated based on the output of the large-scale language model. Output the generated reply text, When generating a reply to the aforementioned review, the purchase history of the transaction item by the end user who posted the review is obtained, and if the transaction item related to the review is included multiple times in the purchase history, the large-scale language model is input with the review and an instruction to send a reply that includes an expression of gratitude for the multiple purchases of the transaction item related to the review, and the reply to the review is generated based on the output of the large-scale language model. How to generate a response to a review.

14. Computers The system outputs multiple reviews for the transaction target, and also outputs a reply generation operation unit that receives instructions to generate a reply corresponding to each of the reviews. When the reply generation operation unit receives the instruction, it presents options for the reply policy to the review corresponding to the reply generation operation unit, and outputs a reply generation instruction screen that accepts an instruction to generate a reply based on the reply policy indicated by the selected option. The response policy selected from the above options is obtained, The large-scale language model is input with the review and an instruction that tells it to respond to the review in accordance with the acquired response policy, and a response to the review is generated based on the output of the large-scale language model. Output the generated reply text, The aforementioned reply generation instruction screen presents a first option indicating a reply policy that automatically generates the reply content, and a second option indicating a reply policy that accepts instructions regarding the reply content. If the second option is selected, the system is configured to further output an input field for the instructions. How to generate a response to a review.

15. Computers, A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply generation operation unit receives the instruction, means for presenting options for the reply policy to the review corresponding to the reply generation operation unit, Means for obtaining the aforementioned reply policy selected from the aforementioned options, A means for inputting the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generating a response sentence to the review based on the output of the large-scale language model, means for outputting the generated reply text, A means of obtaining the purchase history of the transaction item by the end user who posted the aforementioned review, To make it function as, The generating means, if the transaction subject to the review is included multiple times in the purchase history, inputs the review and an instruction to the large language model that instructs it to send a reply including a thank you for purchasing the transaction subject to the review multiple times, and generates a reply to the review based on the output of the large language model. program.

16. Computers, A means for outputting multiple reviews for a transaction target and a reply generation operation unit that receives instructions for generating a reply corresponding to each of the reviews, When the reply text generation operation unit receives the instruction, means for outputting a reply text generation instruction screen that presents options for reply policies to the review corresponding to the reply text generation operation unit and accepts an instruction to generate a reply text based on the reply policy indicated by the selected option. Means for obtaining the aforementioned reply policy selected from the aforementioned options, A means for inputting the review and an instruction sentence instructing the large-scale language model to respond to the review in accordance with the acquired response policy, and generating a response sentence to the review based on the output of the large-scale language model, means for outputting the generated reply text, To make it function as, The aforementioned reply generation instruction screen presents a first option indicating a reply policy that automatically generates the reply content, and a second option indicating a reply policy that accepts instructions regarding the reply content. If the second option is selected, the system is configured to further output an input field for the instructions. program.

Images