Post reply apparatus and post reply program

Abstract

To provide a post reply apparatus that gives an appropriate reply for a store in response to a post submitted to the store.SOLUTION: A post reply apparatus that replies to a posting text submitted to a post target includes: posting text acquisition means for acquiring a posting text; instruction sentence creation means for creating, on the basis of information relating to the post target, an instruction sentence for causing a language model to generate a reply sentence to the posting text; instruction sentence transmission means for transmitting the instruction sentence to the language model; reply sentence reception means for receiving, from the language model, a reply sentence corresponding to the instruction sentence; and reply sentence posting means for posting the reply sentence as a reply to the posting text.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a post response device and a post response program. [Background technology]

[0002] In recent years, systems that allow users (consumers, etc.) to post word-of-mouth and reviews about stores have become widely known on the Web. For example, when choosing a restaurant or purchasing a product, many users refer to word-of-mouth and reviews, and the content of these posts is known to have a significant impact on consumer behavior. In addition, such systems have a reply function, allowing store operators to respond to posted word-of-mouth and reviews.

[0003] As a related technology, for example, Patent Document 1 describes a response output system that utilizes a device or system equipped with a tool using a machine learning model such as a large language model (LLM) such as the well-known Chat GPT, to generate responses to customer reviews using specified language expressions (for example, colloquial language, honorific language, humble language, polite language, and beautifying language). This reduces the time and effort required for store managers to respond to individual reviews, and enables them to efficiently respond to reviews using language expressions specified by the store. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7400181 Summary of the Invention [Problem to be solved by the invention]

[0005] Responding to reviews is an important factor in building a relationship of trust between the store and the user, leading to increased customer satisfaction and a better store image. By responding sincerely and courteously to reviews submitted by the store, the store can demonstrate that it values ​​the user's opinion, and it can help users get to know the store's management and staff and create a sense of familiarity, ultimately increasing users' willingness to visit the store. Therefore, it is desirable for reviews to be original and reflect the store's characteristics and individuality.

[0006] However, in the invention described in Patent Document 1, although the LLM generates a response according to the language expression specified by the store, it only generates a word-of-mouth response in the tone (style) specified by the store.

[0007] The present invention has been proposed in view of the above points, and one object of the present invention is to provide a posting response device that responds to messages posted to a store in a manner that is appropriate for that store. [Means for solving the problem]

[0008] In order to solve the above problems, the post response device of the present invention is a post response device that replies to a post posted to a posting target, and includes: a post message acquisition means for acquiring the post message; an instruction message creation means for creating an instruction message for causing a language model to generate a response message to the post message based on information related to the posting target; an instruction message sending means for sending the instruction message to the language model; a response message receiving means for receiving a response message corresponding to the instruction message from the language model; and a response message posting means for posting the response message to the post message. [Effects of the Invention]

[0009] According to an embodiment of the present invention, it is possible to provide a post reply device that responds to messages posted to a store in a manner suited to the store. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating an example of a network configuration of a posting and replying system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating an example of the hardware configuration of a post reply server according to the present embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of the software configuration of a post reply server according to the present embodiment. [Figure 4] FIG. 2 is a diagram illustrating an example of a data configuration of a store setting DB according to the embodiment. [Figure 5] FIG. 10 is a flowchart illustrating a word-of-mouth response process of the post response server according to the embodiment. [Figure 6] FIG. 10 is a diagram illustrating an example of a review posting screen according to the present embodiment. [Figure 7] FIG. 10 is a diagram illustrating an example of a prompt according to the embodiment. [Figure 8] FIG. 10 is a diagram showing an example of a reply message (answer message) according to the present embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of a review posting screen according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail with reference to the drawings. <System configuration> (Network configuration) Fig. 1 is a diagram showing an example of a network configuration of a post-reply system according to this embodiment. The post-reply system 100 in Fig. 1 includes a post server 10, a post-reply server 20, an LLM (Large Language Models) 30, a store terminal 40, and a poster terminal 50, which are connected via a network 60.

[0012] The posting server 10 is a server device on the Web that allows posters to post word-of-mouth comments and reviews (hereinafter simply referred to as word-of-mouth) about stores. The posted word-of-mouth comments are made publicly available for viewing by store managers, posters, and other general users. The posting server 10 also has a function for responding to each posted word-of-mouth comment, allowing store managers to reply (post) comments to posted word-of-mouth comments.

[0013] The posting response server 20 is a server device that, when a word-of-mouth comment is posted by a poster on the posting server 10, causes the LLM 30 to generate a response (answer) to the word-of-mouth comment and replies (posts) the generated response as a comment on the word-of-mouth comment to the posting server 10. When word-of-mouth comments are posted about their own store, store managers can reliably respond to each word-of-mouth comment posted by the poster while reducing the time and effort required to respond to the comment.

[0014] The LLM (Large-Scale Language Model) 30 is a general-purpose natural language processing model that uses a large amount of text data to pre-train grammar, word meanings, etc. without a teacher. For the general-purpose natural language processing model, for example, a commercial language model (generative AI) based on BERT (Bidirectional Encoder Representations from Transformers), GPT (Generative Pre-trained Transformer)-3.5, GPT-4, etc. can be used.

[0015] The store terminal 40 is, for example, a smartphone, tablet terminal, or PC (Personal Computer), and is a user terminal device on the store side that uses the posting server 10 and the post-reply server 20. The store manager (or store staff) can use the store terminal 40 to access and log in to the posting server 10, register a profile of their store (store name, location, URL, telephone number, store introduction, products, menu, etc.), and view word-of-mouth and reviews about their store posted by posters. The store manager can also use the store terminal 40 to access and log in to the post-reply server 20 and configure various settings to generate a response (answer) to a word-of-mouth comment in the LLM 30.

[0016] The poster terminal 50 is, for example, a smartphone, a tablet terminal, or a PC, and is a terminal device of a poster who uses the posting server 10. The poster can use the poster terminal 50 to access and log in to the posting server 10 and post word-of-mouth comments, reviews, etc. about a store.

[0017] The network 60 is a communication network that may be wired or wireless, and may include, for example, the Internet, a public line network, Wi-Fi (registered trademark), Bluetooth (registered trademark), and the like.

[0018] In order to access the posting server 10 or the posting reply server 20, a predetermined application program (general-purpose web browser, dedicated application, etc.) is installed in advance on the store terminal 40 and the poster terminal 50.

[0019] (Hardware configuration) 2 is a diagram showing an example of the hardware configuration of a post response server according to this embodiment. The post response server 20 includes a CPU (Central Processing Unit) 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, an HDD (Hard Disk Drive) 24, and a communication device 25.

[0020] The CPU 21 executes various programs and performs arithmetic processing. The ROM 22 stores programs required at startup, etc. The RAM 23 is a work area for temporarily storing processing by the CPU 21 and storing data. The HDD 24 stores various data and programs. The communication device 25 communicates with other devices via the network 60.

[0021] (Software configuration) 3 is a diagram showing an example of the software configuration of the post response server according to this embodiment. The post response server 20 has, as its main functional units, a posted message acquisition unit 201, an instruction message creation unit 202, an instruction message sending unit 203, a response message receiving unit 204, a response message posting unit 205, and a storage unit 209.

[0022] The posted message acquisition unit 201 has a function of acquiring word-of-mouth reviews posted to a store from the posting server 10. The instruction message creation unit 202 has a function of creating a prompt for the LLM 30 to generate a response message to a word-of-mouth review based on the store setting information. The instruction message sending unit 203 has a function of sending a prompt to the LLM 30. The response message receiving unit 204 has a function of receiving a response message (answer) in response to the prompt from the LLM 30. The response message posting unit 205 has a function of posting a response message to a word-of-mouth review. The memory unit 209 stores the store setting DB.

[0023] Each functional unit is realized by a program executed on hardware resources such as the CPU, ROM, and RAM of the computer that constitutes the post reply server 20. These functional units may be interpreted as "means," "module," "unit," or "circuit." Each DB of the storage unit 209 may also be placed in an external storage device on the network 60. The computer program and application program may also be stored in a computer-readable storage medium.

[0024] (Database) 4 is a diagram showing an example of the data configuration of a store setting DB according to this embodiment. The store setting DB is a DB in which store setting information is registered in advance. The store setting DB according to this embodiment has data items such as "Store ID," "Store Name," "Location," "URL," "TEL," "Email," "Password," "Store Staff Name," "Store Features," "Keywords," "Response Character Count," and "Automatic Response."

[0025] The "Store ID" is a unique identifier assigned to each store. The "Store Name," "Location," "URL," "TEL," "Email," and "Store Personnel Name" are the store's name, address, web URL, telephone number, email address, and store personnel name (name of the person who responded to the review), respectively. The "Store ID" and "Password" are a common ID and password for users to access and log in to the posting server 10 and the post response server 20. However, if the ID and password for users to access and log in to the posting server 10 are different from the ID and password for users to access and log in to the post response server 20, they should register the respective ID and password. Furthermore, since a store may have multiple URLs simultaneously, such as its official website, information site, social networking site, special site, etc., multiple URLs can be registered for "URL," such as "URL1," "URL2," etc.

[0026] "Store characteristics" is information that indicates the characteristics of a store (such as a store description that indicates the store characteristics). "Store characteristics" can be registered and set by the store operator themselves, or can be edited based on automatically obtained store characteristics. In addition, when a "URL" is registered and set, information indicating store characteristics such as the store's strengths, concept, and atmosphere (casual, luxurious, homey, etc.) can be automatically generated by the LLM30 based on the store website at the URL.

[0027] "Keywords" are designated words that LLM30 must use in its response to reviews. "Keywords" are registered and set by the store operator themselves, and are keywords or key phrases that the store wants to convey to posters who view their responses and other general users.

[0028] The "response character count" is the number of characters in the response text (answer text) that LLM30 generates in response to reviews. It is registered and set by the store operator. The "response character count" may be a guideline indicating the length of the response text (short, normal, long), or it may be the upper limit of the response characters.

[0029] "Automatic Reply" is a setting value that indicates whether or not, when a reply (answer) to a review is generated by the LLM 30, the generated reply is automatically sent to the posting server 10 as a comment on the review. When "Automatic Reply" is OFF, the posting reply server 20 automatically sends the generated reply to the posting server 10 after confirmation by the store manager.

[0030] These data items in the store setting DB can be registered at the time of initial registration in the post response server 20, such as when the store manager registers a user.

[0031] <Review response processing> 5 is a flowchart showing the word-of-mouth response process of the post response server according to this embodiment. The CPU 21 reads and executes a program capable of realizing the flowchart shown in FIG. 5, thereby realizing each step (hereinafter referred to as "S").

[0032] S1: The posting response server 20 determines (detects) whether or not a new word-of-mouth review for the target store has been posted on the posting server 10. FIG. 6 is a diagram showing an example of a review posting screen according to this embodiment. A poster can access and log in to the posting server 10 using a poster terminal 50 to post a review 51 about a store. The posted review 51 is made publicly available for viewing by the store operator, the poster, and other general users. The post-response server 20 can detect when a new review about a target store is posted (published) on the posting server 10.

[0033] S2: If the post response server 20 determines that a new word-of-mouth review has been posted for the store, it acquires the new word-of-mouth information posted from the posting server 10. The acquired new word-of-mouth information is, for example, "Taro, rating 4, the atmosphere is welcoming, so I often go alone. The food is also delicious." (Fig. 5).

[0034] S3: The post response server 20 acquires the setting information corresponding to the store from the store setting DB using the store ID (e.g., ID0001) as a key. The acquired new word-of-mouth information includes, for example, the "store name," "location," "URL," "store staff name," "store features," "keywords," "number of response characters," and "automatic response" (Fig. 4).

[0035] S4: The post reply server 20 creates a prompt (instruction sentence) for causing the LLM 30 to generate a reply sentence (answer sentence) to the new word-of-mouth based on the acquired new word-of-mouth information and setting information. 7 is a diagram showing an example of a prompt according to this embodiment. As shown in FIG. 7, the prompt according to this embodiment is an instruction sentence that causes the LLM 30 to generate a response (answer sentence) to the new review based on the new review information and setting information.

[0036] More specifically, it is as follows. By including the "store name" and "store representative name" in the prompt from the setting information, you can clearly indicate who is responding to a new review in the response (answer) to the new review. By including "store characteristics" in the prompt settings, responses to new reviews can be made more original, reflecting the characteristics and individuality of the store. By including a "URL" in the prompt settings, responses to new reviews can be more comprehensive and original, reflecting the latest store characteristics and individuality. Also, by including multiple "URLs" in the prompt, such as the store's official website, information sites, and social media, responses to new reviews can be more comprehensive and original, reflecting the latest store characteristics and individuality. For example, campaigns announced only on social media can also be reflected in the response. By including the "keyword" setting in the prompt, you can ensure that the keyword is used in the response to a new review. By including the "Number of response characters" setting in the prompt, you can specify the length of the response text (short, normal, long) in the response text (answer text) to new reviews.

[0037] S5: The post reply server 20 sends the created prompt to the LMM 30.

[0038] S6: The post reply server 20 receives the reply message (answer message) from the LMM 30. 8 is a diagram showing an example of a reply message (answer message) according to this embodiment. As shown in FIG. 8, it can be seen that the reply message (answer message) according to this embodiment is generated in accordance with the prompt (instruction message) in FIG.

[0039] S7: The post reply server 20 determines whether the setting information "automatic reply" in the store setting DB is ON.

[0040] S8: If "Automatic Reply" is ON, the posting response server 20 accesses and logs in to the posting server 10 using the "Store ID" and "Password" and posts a response as a store's comment to the new review.

[0041] S9: If "Automatic Reply" is not ON, the posting response server 20 proceeds to S8 after the store manager has performed an operation to allow a reply, and posts a reply as a store's comment to the new review. The posting response server 20 notifies the store manager of the generated reply, and the store manager can confirm or modify the reply and then allow a reply.

[0042] 9 is a diagram showing an example of a review posting screen according to this embodiment. It can be seen that a reply (a response from the store) 52 has been posted (published) as a comment from the store in response to the posted review 51.

[0043] As described above, when a review is posted by a poster on the posting server 10, the posting response server 20 according to this embodiment causes the LLM 30 to generate a response (answer) to the review, and replies (posts) the generated response as a comment on the review to the posting server 10. When reviews are posted about their own store, store managers can reduce the time and effort required to respond to reviews, while also being able to provide original responses to each review posted by the poster that reflect the characteristics and individuality of the store.

[0044] (supplement) Responses to reviews should generally be written using polite language, but by using polite language, tone of voice, or language appropriate to the profile of the store representative (responder), it is possible to create a more original response that better reflects the characteristics and individuality of the store. In this case, in addition to the "store representative" in the store setting DB (Figure 4), additional data items such as the store representative's gender (male, female, or not specified), age (20s, 30s-40s, 50s or older), and personality (friendly, serious, humorous, etc.) can be added as profile information for the store representative. Then, a prompt can be created to have the LLM 30 generate a response (answer) using, for example, the following expressions according to the store representative's profile information (S4).

[0045] If you are male, use masculine responses and concise responses. If you are female, use feminine responses and polite responses. For those in their 20s, use casual language, trendy phrases and emojis, and friendly expressions. For those in their 30s and 40s, use polite language, calm expressions, and honorific language. For those in their 50s and older, use polite and formal language, calm expressions, honorific language, and do not use trendy phrases or emojis. If you are friendly, you will use friendly language, trendy slang, emojis, and familiar expressions. If you are serious, you will use polite and formal language and honorifics. If you are humorous, you will use jokes, trendy slang, emojis, and familiar expressions.

[0046] Furthermore, by creating a response using polite language, tone of voice, or language appropriate to the store's characteristics, it is possible to provide a more original response that reflects the store's characteristics and individuality. Store characteristics can be obtained from the "Store Characteristics" field in the store setting DB (Figure 4) or from the store's website based on the "URL." Alternatively, in addition to the "Store Characteristics," the store setting DB (Figure 4) may also include data items such as "casual," "high-end," and "down-to-earth (homey)" as genre information that indicates one of the store characteristics. Then, depending on these store characteristics, a prompt can be created to have the LLM 30 generate a response (answer) using, for example, the following expressions (S4). Note that the store's price range information may also be used as one of the store characteristics.

[0047] · For casual, use casual language, buzzwords and emojis. For high-end items, use polite and formal language, calm expressions, honorific language, and avoid using trendy phrases and emojis. · For common people, use familiar expressions, dialects, and friendly language.

[0048] Alternatively, in addition to the "store characteristics," the store setting DB (FIG. 4) may also include data items such as "luxury," "ideal for entertaining," "home-like relaxing store," and "health-oriented store" as concept information indicating one of the store characteristics. Then, depending on the store characteristics, a prompt can be created to have the LLM 30 generate a response sentence (answer sentence) using, for example, the following expressions (S4):

[0049] · For luxury settings and business entertainment, use polite and formal language, calm expressions, honorific language, and avoid buzzwords and emojis. · For common people, use familiar expressions, dialects, and friendly language. If your store is a home away from home, use casual language, buzzwords, emojis, and friendly expressions.

[0050] The following embodiments are also mentioned. The post response server 20 can perform language analysis and sentiment analysis when creating a prompt (instruction) in S4 for the new review information acquired in S2. The sentiment classification results are classified, for example, into positive, neutral, and negative, as well as more specifically into praise, complaint, impression, question, and request. The post response server 20 then creates a prompt (instruction) for the LLM 30 to generate a response (answer) according to the sentiment classification result. More specifically, if the sentiment classification result is, for example, positive or praise, the post response server 20 creates an instruction to generate a response expressing gratitude (appreciation) for the review; if the sentiment classification result is negative or complaint, the post response server 20 creates an instruction to generate a response apologizing for and suggesting improvements to the review; if the sentiment classification result is a question, the post response server 20 creates an instruction to generate a response that answers the review's question; and if the sentiment classification result is a request, the post response server 20 creates an instruction to generate a response that suggests improvements to the review.

[0051] The setting information in the store setting database may be at least one of "store characteristics" or "URL." By including at least one of the setting information, "store characteristics" or "URL," in the prompt, the response (answer) to a new review can be an original response that reflects the characteristics and individuality of the store.

[0052] Generally, it is desirable for the response to be in the same language as the new review, but to accommodate foreign languages, you can specify the language in which the response should be generated in the prompt, such as "Please write the response in English." or "Please write the response in the same language as the new review, as well as in English."

[0053] If the poster is a repeat user who has posted multiple times (if there is a history of posts by the same poster in the past), you may instruct the prompt to generate a special response that is different from that of a first-time poster (for example, "Thank you for your reviews" or "Thank you for your valuable opinions").

[0054] <Summary> In the post-reply system 100 according to this embodiment, when a user posts a review on the posting server 10, the post-reply server 20 causes the LLM 30 to generate a response (answer) to the review and replies (posts) the generated response to the review as a comment on the review to the posting server 10. When a review is posted about their store, a store manager can provide an original response to each review posted by the user that reflects the characteristics and individuality of the store, while reducing the time and effort required to respond to the review. That is, this embodiment makes it possible to provide a post-reply device that provides a response tailored to each store in response to posts sent to the store.

[0055] Although the present invention has been described with reference to specific examples according to the preferred embodiments of the present invention, it is apparent that various modifications and changes can be made to these examples without departing from the broad spirit and scope of the present invention as defined in the appended claims. In other words, the details of the examples and the accompanying drawings should not be construed as limiting the present invention.

[0056] This system can be widely applied to systems that allow businesses to post replies to posts from users, such as word-of-mouth about various facilities and services, word-of-mouth about stores on online shopping sites, and reviews of individual products and services. Word-of-mouth and reviews can also be called VOC (Voice of Customer).

[0057] (Appendix 1) A posting response device that responds to a posting (e.g., word-of-mouth, review) posted to a posting target (e.g., a store, product, or service), a posted message acquisition means for acquiring the posted message; an instruction generation means for generating an instruction (e.g., a prompt) for causing a language model (e.g., LLM) to generate a response to the posted message based on information about the posting target (e.g., store setting information); an instruction statement sending means for sending the instruction statement to the language model; a response message receiving means for receiving a response message (a response message generated by the LLM) corresponding to the instruction message from the language model; a reply message posting means for posting the reply message to the posted message; A post reply device comprising: [Explanation of symbols]

[0058] 10 Submission Server 20 Post-Response Server 30 LLM (Large Scale Language Models) 40 Store terminals 50 poster terminal 60 Network 100 Post Reply System 201 Post Acquisition Department 202 Instruction Writing Department 203 Instruction Message Transmission Unit 204 Reply Message Receiving Unit 205 Reply Posting Section 209 Memory section

Claims

1. A post reply device that replies to a message posted to a posting target, a posted message acquisition means for acquiring the posted message; an instruction generation means for generating an instruction to cause a language model to generate a response to the posted message based on information about the posting target; an instruction statement sending means for sending the instruction statement to the language model; a response message receiving means for receiving a response message corresponding to the instruction message from the language model; a reply message posting means for posting the reply message to the posted message; A post reply device comprising:

2. The information about the posting target includes at least one of the name of the posting target or the name of the respondent; 2. The post reply device according to claim 1,

3. The information about the posting subject includes information indicating characteristics of the posting subject; 2. The post reply device according to claim 1,

4. The information about the posting subject includes a URL where information indicating characteristics of the posting subject is posted; 2. The post reply device according to claim 1,

5. the URL includes a plurality of URLs; 5. The post reply device according to claim 4,

6. The information about the posting subject includes keywords to be used in the reply; 2. The post reply device according to claim 1,

7. The information about the posting subject includes an emotion classification result for the posting text; 2. The post reply device according to claim 1,

8. the instruction sentence creation means creates an instruction sentence for generating a special reply sentence different from a reply sentence from a poster who has no history of posting in the past, when the posted message is from a poster who has a history of posting in the past to the posting target; 2. The post reply device according to claim 1,

9. The instruction sentence creation means Creating an instruction sentence for generating a reply sentence in a written form according to the profile information of the respondent; 3. The post reply device according to claim 2, wherein:

10. The instruction sentence creation means Creating an instruction sentence for generating a reply sentence in a written expression according to the characteristics of the posting subject; 5. The post reply device according to claim 3 or 4, wherein:

11. On the computer, a posted message acquisition means for acquiring a message posted to a posting target; an instruction generation means for generating an instruction to cause a language model to generate a response to the posted message based on information about the posting target; an instruction statement sending means for sending the instruction statement to the language model; a response message receiving means for receiving a response message corresponding to the instruction message from the language model; a reply message posting means for posting the reply message to the posted message; Post-reply program to function as.

Images