JP7813012B2 - Chatbot control device, chatbot control method, and program - Google Patents

Description

The present invention relates to a chatbot control device, a chatbot control method, and a program.

Chatbots, communication robots that can automatically hold conversations with users over chat, have been known for some time. In recent years, efforts have been made to utilize chatbot technology to provide information useful for decision-making regarding the purchase of various products and services, and to carry out the procedures required for purchases through automated conversations.

Japanese Patent Application Laid-Open No. 2022-77779

Prior art employs a method of proposing products and other items suited to each user by analyzing user information (interests, tastes, purchasing objectives, etc.) obtained through conversations with the user over chat. To achieve this type of user-appropriate product and other suggestion, a designer (communication designer) must manually create in advance a scenario with a conditional branching structure (tree-type structure) that defines a series of conversations with the user (see Patent Document 1). Because such scenarios tend to be complex, the workload of preparing the scenario increases. Furthermore, if the scenario manually generated by the designer is inappropriate or does not comprehensively take into account the user's information, there is a risk that products and other items unsuitable for the user will be suggested.

On the other hand, it is also possible to suggest products and other items based on user information by using learning models based on machine learning techniques such as deep learning. However, preparing such learning models requires associating (labeling, annotating) user information with a huge amount of data to be analyzed (product data), which increases the workload. Furthermore, if the association is inappropriate or if the data cannot be associated with a wide variety of user information, there is a risk that products and other items that are not suitable for the user will be suggested.

The present invention was made in consideration of these circumstances, and one of its objectives is to provide a chatbot control device, chatbot control method, and program that can reduce the operational burden of enabling a chatbot to suggest products or services, while improving the accuracy of suggestions.

One aspect of the present invention is a chatbot control device that controls a chatbot that makes product or service suggestions to a user. The chatbot control device includes: a first tag generation unit that generates multiple first tags based on output data obtained by inputting, into a first language model, information indicating the content of the product or service and a first prompt that instructs the generation of multiple first tags associated with the product or service; a second tag generation unit that generates multiple second tags based on output data obtained by inputting, into a second language model, information indicating the conversation between the chatbot and the user and a second prompt that instructs the generation of multiple second tags associated with the user; and a proposal generation unit that generates product or service suggestions tailored to the user based on the generated multiple first tags and multiple second tags.

One aspect of the present invention reduces the operational burden required to enable chatbots to suggest products or services, while improving the accuracy of the suggestions.

FIG. 1 is a diagram illustrating an example of a system configuration for realizing a chatbot service according to an embodiment. FIG. 1 is a functional block diagram illustrating an example of the configuration of a chatbot server 1 according to an embodiment. FIG. 10 is a diagram illustrating input and output data of a first language model (product tag generation) according to the embodiment. FIG. 10 is a diagram illustrating input and output data of a second language model (user tag generation) according to the embodiment. FIG. 10 is a diagram illustrating input and output data of a third language model (individual offer generation) according to the embodiment. FIG. 2 is a functional block diagram showing an example of the configuration of a user terminal device 100 according to the embodiment. FIG. 10 is a diagram showing an example of a conversation screen P1 (initial conversation) displayed on the user terminal device 100 according to the embodiment. FIG. 10 is a diagram showing an example of a conversation screen P1 (individual offer) displayed on the user terminal device 100 according to the embodiment. 10 is a flowchart illustrating an example of a product tag generation process performed by the chatbot server 1 according to the embodiment. FIG. 2 is a diagram showing an example of product data PD according to the embodiment. FIG. 10 is a sequence diagram illustrating an example of a chat control process in the chatbot service according to the embodiment. FIG. 2 is a diagram showing an example of user data UD according to the embodiment. A figure explaining the relationship between the user tag UT and the product tag PT when an individual offer is generated by the chatbot server 1 according to the embodiment.

Embodiments of the chatbot control device, chatbot control method, and program of the present invention will be described below with reference to the drawings. The chatbot generation device of the embodiment uses machine learning techniques such as PromptLearning, RetrievalAugmentedGeneration, and FineTuning using Large Language Models (LLM), thereby reducing the operational burden required to enable a chatbot to suggest products or services and improving the accuracy of the suggestions.

<Chatbot service>
FIG. 1 illustrates an example of a system configuration for implementing a chatbot service according to an embodiment. The chatbot service uses a chatbot to automatically converse with users, such as businesses providing products or services (hereinafter referred to as "products, etc."), to learn about the user's hobbies, preferences, and personal background (also referred to as "insights") and propose information on products, etc., tailored to the user. Examples of businesses include cosmetics retailers, food retailers, hospitals, human resources service providers, financial institutions, insurance companies, real estate agents, and telecommunications companies. The chatbot service is implemented primarily around a chatbot server 1. The chatbot server 1 communicates with, for example, one or more service servers 3, one or more message servers 5, one or more user terminal devices 100, and one or more operation terminal devices 200 via a communication network NW. The communication network NW includes, for example, the Internet, a local area network (LAN), a wireless base station, a provider device, and the like.

<Device configuration>
<Chatbot Server 1>
The chatbot server 1 provides a function for generating a scenario that defines a series of conversations with a user to a designer D who operates the operation terminal device 200. The chatbot server 1 also provides a chatbot service to a user U who operates a user terminal device 100. Figure 2 is a functional block diagram showing an example of the configuration of the chatbot server 1 according to an embodiment. The chatbot server 1 includes, for example, a communication unit 11, a control unit 13, and a storage unit 15. The chatbot server 1 is an example of a "chatbot control device." The communication unit 11 is a communication interface for connecting to a communication network NW. The communication unit 11 is, for example, a network interface card.

The control unit 13 includes, for example, an acquisition unit 131, a conversation generation unit 132, a transmission unit 133, a product tag generation unit 134, a user tag generation unit 135, an individual offer generation unit 136, a learning unit 137, a scenario generation unit 138, and a display control unit 139. The components of the control unit 13 are realized, for example, by a hardware processor such as a CPU executing a program (software). Some or all of these components may be realized by hardware such as an LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or GPU (Graphics Processing Unit), or may be realized by a combination of software and hardware. The program may be stored in advance in a storage device (non-transitory computer-readable storage medium) such as a hard disk drive (HDD) or flash memory, or it may be stored in a removable storage medium (non-transitory computer-readable storage medium) such as a DVD or CD-ROM, and installed in the storage device by inserting the storage medium into a drive device.

The acquisition unit 131 acquires various information (data) from the service server 3, message server 5, user terminal device 100, and operation terminal device 200 via the communication network NW. For example, the acquisition unit 131 acquires conversation information indicating the content of the conversation entered in response to the operation of user U from the user terminal device 100. Also, for example, the acquisition unit 131 acquires instruction information regarding tag generation and scenario generation entered in response to the operation of designer D from the operation terminal device 200.

The conversation generation unit 132 generates conversation text to be sent to the user terminal device 100 (conversation text to be presented to the user) based on conversation information acquired from the user terminal device 100 and scenario data SD for each business operator pre-stored in the storage unit 15. The conversation generation unit 132 generates conversation text including suggestions for products, etc. tailored to the user. The conversation generation unit 132 is an example of a "conversation generation unit."

The transmission unit 133 transmits the conversation text generated by the conversation generation unit 132 to the user terminal device 100. The transmission unit 133 is an example of a "transmission unit."

The product tag generation unit 134 generates multiple product tags based on output data obtained by inputting information indicating the contents of a product, etc., and a product tag generation prompt (command, first prompt) that instructs the generation of multiple product tags (first tags) associated with the product, etc., into the first language model AI1 pre-stored in the storage unit 15. The product tag generation unit 134 plays a role in expanding the product tag PT by generating multiple product tags PT based on information indicating the contents of a certain product, etc. The product tag generation unit 134 is an example of a "first tag generation unit."

Figure 3 is a diagram illustrating input and output data for a first language model AI1 (product tag generation) according to an embodiment. The first language model AI1 is a large-scale language model (LLM) generated by training a large amount of text data, including product data indicating the contents of products, etc., included in catalogs provided by businesses, etc., as training data. The first language model AI1 is a model that predicts and arranges words that are likely to follow a specific word based on the input text data, and outputs text. Examples of first language models AI1 include ChatGPT (registered trademark), LaMDA, Llama, and Claude. As shown in Figure 3, when used for product tag generation, the input data for the first language model AI1 are product data PD and a product tag generation prompt PR1, and the output data is multiple product tags PT.

A product tag PT is a string of characters that represents the characteristics of a product or the like. Product tags PTs are expressed by parts of speech, such as nouns, adjectives, verbs, and adverbs. The product tag generation prompt PR1 includes instructions for generating multiple product tags PTs. The product tag generation prompt PR1 includes, for example, at least one of a specification of the category of the multiple product tags PTs to be generated and a specification of the number of multiple product tags PTs to be generated (e.g., 10 for each category). Product tag PT categories include, for example, features, occasions, user lifestyle, and user interests. For example, if the product is "shoes," product tag PTs in the features category include, for example, retro, leather upper, 1980s, black, and durability. Similarly, for example, if the product is "shoes," product tag PTs in the occasions category include, for example, casual outdoor, everyday use, walking around town, shopping, leisure, and city life.

For example, the product tag generation prompt PR1 includes, in addition to a character string specifying the product data PD, instructions for generating multiple product tags PT, such as "Please identify the product tags to be used to suggest products. Please describe the tags in words. Generate 10 product tags for each category."

Returning to FIG. 2 , the user tag generation unit 135 generates multiple user tags based on output data obtained by inputting information indicating a conversation with user U conducted via the chat function and a user tag generation prompt (command, second prompt) that instructs the generation of multiple user tags (second tags) associated with user U into a second language model AI2 previously stored in the storage unit 15. The user tag generation unit 135 generates multiple user tags UT based on output data obtained by inputting a user tag generation prompt PR2 that includes information indicating an answer from user U to a question included in the conversation with user U (for example, information indicating one answer from user U to one of the multiple questions) into the second language model AI2. The user tag generation unit 135 plays a role in expanding the user tags UT by generating multiple user tags UT based on information indicating the conversation between a certain user. The user tag generation unit 135 is an example of a "second tag generation unit."

Figure 4 is a diagram illustrating input and output data for the second language model AI2 (user tag generation) according to an embodiment. The second language model AI2 is a large-scale language model (LLM) generated by training a large amount of text data, including data representing conversations with users, as training data. The second language model AI2 is a model that predicts and arranges words that are likely to follow specific words based on the input text data. Examples of the second language model AI2 include ChatGPT (registered trademark), LaMDA, Llama, and Claude. As shown in Figure 4, when used for user tag generation, the input data for the second language model AI2 are conversation data TD and a user tag generation prompt PR2, and the output data is a plurality of user tags UT.

A user tag UT is a character string that represents a user's characteristics. A user tag UT is represented by a part of speech, such as a noun, adjective, verb, or adverb. The user tag generation prompt PR2 includes instructions for generating multiple user tags UT associated with the user U. The user tag generation prompt PR2 includes, for example, a specification of the categories of the multiple user tags UT to be generated and a specification of the number of multiple user tags UT to be generated (e.g., 10 for each category). Categories of user tags UT include, for example, features, user interests, user lifestyle, and occasions. If the product is "shoes," user tags UT in the features category include, for example, stylish, comfortable, formal, and durable. Similarly, if the product is "shoes," user tags UT in the user interests category include, for example, fashion, brand, and functionality.

For example, user tag generation prompt PR2 includes a string specifying the content of the user's response contained in the conversation with the user, as well as instructions for generating multiple user tags UT, such as "Based on the user's responses, identify the user tags to be used to suggest products. Describe the tags in words. Generate five user tags for each category. Try to minimize overlap in the tags generated for each option as much as possible."

The individual offer generation unit 136 generates proposals for products or services personalized for the user (hereinafter referred to as "individual offers"). The individual offer generation unit 136 generates proposals for products or services tailored to the user based on the generated multiple product tags PT (first tags) and multiple user tags UT (second tags). For example, the individual offer generation unit 136 generates proposals tailored to user U based on output data obtained by inputting the generated multiple product tags (first tags) and multiple user tags (second tags) and an offer generation prompt (command, third prompt) that instructs the generation of proposals tailored to user U into a third language model AI3 stored in advance in the storage unit 15. The individual offer generation unit 136 is an example of a "proposal generation unit." The individual offer generation unit 136 may generate proposals tailored to user U by comparing the generated multiple product tags PT (first tags) with the multiple user tags UT (second tags).

Figure 5 is a diagram illustrating input and output data for the third language model AI3 (individual offer generation) according to an embodiment. The third language model AI3 is a large-scale language model (LLM) generated by training a large amount of text data, including product tag and user tag data, as training data. The third language model AI3 is a model that predicts and arranges words that are likely to follow a specific word based on the input text data. Examples of the third language model AI3 include ChatGPT (registered trademark) and LaMDA. As shown in Figure 5, when used for individual offer generation, the third language model AI3 receives input data consisting of multiple product tags, multiple user tags, and an offer generation prompt PR3, and outputs an individual offer OF. The offer generation prompt PR3 includes an instruction to generate a proposal tailored to the user U. The offer generation prompt PR3 may include a specification of the number of products or services to be included in the proposal tailored to the user U (e.g., three products). The offer generation prompt PR3 also includes an instruction to generate a proposal tailored to the user based on the degree of association between the multiple product tags PT and the multiple user tags UT.

For example, the offer generation prompt PR3 includes a character string specifying multiple product tags and multiple user tags, as well as the instruction to generate an individual offer: "Based on the user tags and product tags, please identify three products to be used in proposing to the user." Note that instead of the third language model AI3, a machine learning model such as a neural network trained to output an individual offer OF when multiple product tags and multiple user tags are input may be used.

The first language model AI1, second language model AI2, and third language model AI3 may be configured as separate models, or as a model in which their functions are integrated. For example, the first language model AI1 and the second language model AI2 may be integrated into a single model. Furthermore, for example, the first language model AI1, second language model AI2, and third language model AI3 may be integrated into a single model. It has been confirmed that the performance of each LLM model varies depending on the collected data and its storage format, and the optimal model may be selected and implemented from that perspective.

Returning to Figure 2, the learning unit 137 performs learning using a large amount of data as learning data, generates a first language model AI1, a second language model AI2, and a third language model AI3, and stores them in the memory unit 15. The learning unit 137 separates and tokenizes the text data included in the learning data into its smallest units, learns the context and meaning of words from the obtained tokens, and generates the first language model AI1, the second language model AI2, and the third language model AI3.

The scenario generation unit 138 generates a scenario based on instructions from designer D input via the operation terminal device 200, and stores the generated scenario as scenario data SD in the storage unit 15. The scenario includes an initial conversation, such as a greeting message or questions for user U, and offers proposing products or the like based on user information obtained through the conversation with the user.

The display control unit 139 generates information for displaying various operation screens for generating product tags PT and scenarios, transmits this information to the operation terminal device 200, and causes the operation terminal device 200 to display the various operation screens.

The memory unit 15 is a HDD, flash memory, RAM (Random Access Memory), etc. The memory unit 15 may also be a NAS (Network Attached Storage) device that the chatbot server 1 can access via a network. The memory unit 15 stores the first language model AI1, the second language model AI2, the third language model AI3, scenario data SD, product data PD, user data UD, etc. The product data PD is obtained from the service server 3 or other sales management servers via the communication network NW and stored in the memory unit 15.

<Service Server 3>
The service server 3 is managed by a business that provides products and the like. The service server 3 provides the user terminal device 100 with a site such as a homepage where the user can complete the purchase procedure for the product and the like. When the user U performs a preset trigger operation on this site, processing for providing the chatbot service is initiated. Examples of the trigger operation include pressing a button that starts the chatbot service displayed in a pop-up on the site, pressing an advertising banner displayed on the conversation application, or pressing an advertising display placed by a company that uses the conversation application.

<Message Server 5>
The message server 5 provides a message service (chat service) that realizes a conversation function by sending and receiving messages to the user terminal device 100, the chatbot server 1, and the service server 3. The message server 5 provides a conversation application that can be installed in the user terminal device 100 to the user terminal device 100, and the user terminal device 100 sends and receives messages with other user terminal devices 100 and various servers by executing this conversation application. The message server 5 also provides an API that enables use of the message service from an external device (server), and the chatbot server 1 and the service server 3 send and receive messages with the user terminal device 100 by using this API.

<User Terminal Device 100>
The user terminal device 100 is operated by a user U who purchases and uses products and the like provided by a business operator. The user terminal device 100 is, for example, a portable terminal device such as a smartphone or a tablet terminal. FIG. 6 is a functional block diagram showing an example of the configuration of the user terminal device 100 according to the embodiment. The user terminal device 100 includes, for example, a communication unit 101, a display unit 103, an operation unit 105, a control unit 107, and a storage unit 109.

The communication unit 101 is a wireless communication module that performs wireless communication with a wireless base station connected to the communication network NW. The communication unit 101 communicates with the chatbot server 1, service server 3, message server 5, etc. via the wireless base station. The display unit 103 is a display device such as a liquid crystal display device. The operation unit 105 is an input interface that accepts operation instructions from the user U. Note that the display unit 103 and operation unit 105 may be configured as touch panel displays.

The control unit 107 controls the overall operation of the user terminal device 100. The functions of the control unit 107 are realized, for example, by a hardware processor such as a CPU executing a program (software). The control unit 107 launches a conversation application (program) AP stored in the memory unit 109 to realize a conversation function via a network. The conversation application may be an application dedicated to conversation, or it may also have other functions (such as a call function, an image upload function, and a message upload function). For example, the conversation application may be an application on the provider's side that is provided by a business that provides products, etc. and installed on the user terminal device 100 and has a conversation function built in. Alternatively, the conversation application may be realized using a general-purpose application program such as a web browser. For example, the conversation application may be realized using a chat function on the provider's homepage provided by a service server 3 managed by the business.

Figure 7A is a diagram showing an example of a conversation screen P1 (initial conversation) displayed on a user terminal device 100 according to an embodiment. Figure 7A shows a conversation screen P1 of an initial conversation in which a chatbot service provided by business operator A is started and greeting messages and questions are sent to user U according to a scenario.

Figure 7B is a diagram showing an example of a conversation screen P1 (individual offer) displayed on the user terminal device 100 according to the embodiment. As a result of acquiring user U's information (interests, tastes, purchasing purpose, etc.) in the initial conversation shown in Figure 7A, products etc. (in this example, three individual offers OF for product A, product B, and product C) according to user U's information are presented, as shown in Figure 7B.

<Operation Terminal Device 200>
The operation terminal device 200 is operated by a designer D who sets product tags PT in the chatbot service and designs scenarios. The operation terminal device 200 is, for example, a personal computer, a smartphone, a tablet terminal, etc. The operation terminal device 200 has at least a display function, an input function, a storage function, etc.

<Processing flow>
<Product tag generation process>
Next, various processes executed in the chatbot service will be described. First, the process on the operation side, which is performed based on the operation of designer D, will be described. FIG. 8 is a flowchart showing an example of a product tag generation process by the chatbot server 1 according to an embodiment. The process shown in FIG. 8 is started, for example, by a program being automatically triggered in response to an operation by user U in a chat on the user terminal device 100 during the chat control process described below. Alternatively, the process shown in FIG. 8 may be started, for example, when designer D operates the operation terminal device 200 to input an instruction to start product tag generation, and the chatbot server 1 receives this instruction to start.

(Step S101) First, the acquisition unit 131 of the chatbot server 1 acquires product data PD, for example, from the memory unit 15. Note that the acquisition unit 131 may also acquire product data PD from the service server 3, a website such as a homepage provided by the service server 3, other sales management servers, etc., via the communication network NW.

(Step S103) Next, the product tag generation unit 134 generates multiple product tags based on the output data obtained by inputting the product data PD and the product tag generation prompt PR1 into the first language model AI1. Note that designer D can also operate the operation terminal device 200 to change the instructions included in the product tag generation prompt. For example, designer D can specify the categories of the multiple product tags PT to be generated and the number of multiple product tags PT to be generated (e.g., 10 for each category).

(Step S105) Next, the product tag generation unit 134 stores the generated multiple product tags PT in association with the product data PD in the storage unit 15. FIG. 9 is a diagram showing an example of product data PD according to an embodiment. In the example shown in FIG. 9, various information is registered in association with a product ID that identifies the product. The product name, price, classification, category, and feature items included in the product data PD are obtained from the service server 3, websites such as homepages provided by the service server 3, other sales management servers, etc. Meanwhile, the product tags (first tag, second tag, ..., Xth tag) included in the product data PD are generated by the product tag generation unit 134. The first tag, second tag, ..., Xth tag are categorized by the category specified in the product tag generation prompt PR1. For example, product ID "1001" includes tags such as "retro," "leather," "black," "durable," and "low-cut" as first product tags (features). This completes the processing of this flowchart.

<Chat control processing>
Next, the process on the service side where the user U experiences the chatbot service will be described. Fig. 10 is a sequence diagram showing an example of chat control process in the chatbot service according to the embodiment. Here, the process will be described taking as an example a situation where the user U operates the user terminal device 100 and browses a provider's website using a web browser.

(Step S201) First, when a trigger operation is performed on the site in response to an operation by user U, the user terminal device 100 transmits a trigger operation signal indicating that a trigger operation has been performed to the service server 3. An example of a trigger operation is pressing a button that starts a chatbot service displayed in a pop-up on the site.

(Step S203) Next, in response to the trigger operation signal received from the user terminal device 100, the service server 3 sends a conversation start instruction to the chatbot server 1 to start a series of conversations in accordance with a scenario in the chatbot service.

(Step S205) Next, when the conversation generation unit 132 of the chatbot server 1 receives a conversation start instruction from the service server 3, it generates a conversation based on the scenario data SD stored in the memory unit 15. Here, for example, the conversation generation unit 132 generates the first conversation (first conversation) set as the initial conversation with the user in the scenario associated with the business operator.

(Step S207) Next, the sending unit 133 of the chatbot server 1 sends a request to send conversation data indicating the generated conversation to the message server 5. The chatbot server 1 sends the request to send conversation data to the message server 5, for example, by using an API provided by the message server 5. (Step S209) Having received the request to send conversation data from the chatbot server 1, the message server 5 sends this conversation data to the user terminal device 100.

(Step S211) Next, the user terminal device 100 launches the conversation application AP based on the conversation data received from the chatbot server 1 via the message server 5, and displays the conversational text based on the conversational data on the display unit 103. This allows the user U to check the conversational text automatically sent from the chatbot server 1.

(Step S213) Next, if the conversation data received from the chatbot server 1 via the message server 5 is a question requesting an answer (for example, a question requesting a selection from options showing answer candidates), the user terminal device 100 transmits answer data indicating the answer determined in response to the user U's operation of the operation unit 105 to the message server 5. Note that if the conversation data received from the chatbot server 1 via the message server 5 is a simple message or image that does not request an answer, the user terminal device 100 may transmit receipt confirmation data indicating receipt to the message server 5. (Step S215) Having received the answer data from the user terminal device 100, the message server 5 transmits this answer data to the chatbot server 1.

(Step S217) Next, the acquisition unit 131 of the chatbot server 1 stores the response data received from the user terminal device 100 via the message server 5 in the user data UD of the storage unit 15. Figure 11 is a diagram showing an example of user data UD according to an embodiment. The user data UD stores response data acquired from users, linked to a user ID that identifies each user. In the example of Figure 11, based on the response data from the user to the question about intended use (first question), "1 (office casual)" is stored as the response data to the first question (intended use), linked to user ID "0001".

(Step S219) Next, the user tag generation unit 135 of the chatbot server 1 generates multiple user tags based on the output data obtained by inputting the response data to the first question (usage) and the user tag generation prompt PR2 into the second language model AI2, and stores these in the user data UD of the storage unit 15. In the example of Figure 11, user tags UT such as "stylish," "formal," "business," "office," and "casual event" are stored as first user tags (usage) linked to user ID "0001" and linked to the response data "1 (office casual)" to the first question (usage).

The series of processing QA from steps S205 to S219 above is repeatedly executed according to the scenario. That is, the transmission of conversations and the reception of responses from objects set in order in the scenario are repeatedly executed. For example, in step S219, the user tag generation unit 135 of the chatbot server 1 generates multiple user tags (second user tags (selection points)) based on the output data obtained by inputting the response data to the second question (selection points) and the user tag generation prompt PR2 into the second language model AI2, and stores them in the user data UD of the memory unit 15. Finally, when the acquisition unit 131 of the chatbot server 1 receives the response data from user U to the last question, the process proceeds to the next step S221. Furthermore, in parallel with the series of processing QA above, the product tag generation process is executed, and the generated multiple product tags PT are linked to the product data PD in the memory unit 15 and stored.

(Step S221) Next, the individual offer generation unit 136 generates an individual offer. The individual offer generation unit 136 generates an offer tailored to user U based on output data obtained by inputting the generated multiple product tags PT and multiple user tags UT, and an offer generation prompt PR3 that instructs the generation of an offer tailored to user U, into the third language model AI3.

12 is a diagram illustrating the relationship between user tags UT and product tags PT when an individual offer is generated by the chatbot server 1 according to an embodiment. In the example shown in FIG. 12, the user tag UT set for user ID "0001" and the product tag PT set for product ID "1001" share common tags such as "leather," "black," "durable," "office," "business," and "men's." In this way, the individual offer generation unit 136 generates proposals for user U based on the degree of association between the generated multiple product tags PT and multiple user tags UT. For example, the individual offer generation unit 136 generates, as an individual offer, a product that shares a large number of product tags PT with the user tag UT of a certain user U (e.g., a threshold value or more). Note that the associated user tag UT and product tag PT do not necessarily have to be tags with the same wording; tags that are related in terms of the content indicated by the tag wording may be associated. The method of association can be specified in instructions included in the offer generation prompt PR3.

(Step S223) Next, the sending unit 133 of the chatbot server 1 sends a request to send conversation data including the generated individual offer to the message server 5. (Step S225) Having received the request to send conversation data including the individual offer from the chatbot server 1, the message server 5 sends this conversation data to the user terminal device 100.

(Step S227) Next, the user terminal device 100 displays on the display unit 103 the conversational text based on the conversation data including the individual offer received from the chatbot server 1 via the message server 5. This allows the user U to check the conversational text including the individual offer automatically sent from the chatbot server 1.

(Step S229) Next, the user terminal device 100 transmits to the message server 5 a purchase instruction determined from the individual offer in response to the user U's operation of the operation unit 105. (Step S231) The message server 5, which has received the purchase instruction from the user terminal device 100, transmits this purchase instruction to the chatbot server 1. (Step S233) The transmission unit 133 of the chatbot server 1, which has received the purchase instruction from the message server 5, transmits this purchase instruction to the service server 3.

(Step S235) When the service server 3 receives a purchase instruction from the chatbot server 1, it generates a purchase screen for performing payment processing, etc. (Step S237) Next, the service server 3 transmits purchase screen data for displaying the generated purchase screen to the user terminal device 100.

(Step S239) Next, the user terminal device 100 displays the purchase screen on the display unit 103 based on the purchase screen data received from the service server 3. (Step S241) Next, the user terminal device 100 transmits the purchase information entered by the user U in response to operation of the operation unit 105 to the service server 3.

(Step S243) Next, the service server 3 performs purchase processing, including payment processing and shipping processing, based on the purchase information received from the user terminal device 100. This completes the processing in this sequence diagram.

The embodiments described above reduce the operational burden of enabling chatbots to suggest products or services, while also improving the accuracy of the suggestions. In particular, by generating product tags and user tags using a language model, the burden of manual tag generation can be reduced. Furthermore, by generating suggestions for products or services tailored to the user based on the product tags and user tags generated in this way, the accuracy of the suggestions can be improved.

The above describes the form for carrying out the present invention using an embodiment, but the present invention is in no way limited to such an embodiment, and various modifications and substitutions can be made without departing from the spirit of the present invention.

1...Chatbot server, 11...Communication unit, 13...Control unit, 15...Storage unit, 131...Acquisition unit, 132...Conversation generation unit, 133...Transmission unit, 134...Product tag generation unit, 135...User tag generation unit, 136...Individual offer generation unit, 137...Learning unit, 138...Scenario generation unit, 139...Display control unit, 100...User terminal device, 101...Communication unit, 103...Display unit, 105...Operation unit, 107...Control unit, 109...Storage unit, 200...Operation terminal device, 3...Service server, 5...Message server, NW...Communication network

Claims (8)

A chatbot control device that controls a chatbot that makes suggestions about products or services to a user,
a first tag generation unit that generates the plurality of first tags based on output data obtained by inputting information indicating the content of the product or service and a first prompt that instructs generation of the plurality of first tags associated with the product or service into a first language model;
a second tag generation unit that generates the plurality of second tags based on output data obtained by inputting information indicating a conversation between the chatbot and the user and a second prompt that instructs the generation of a plurality of second tags associated with the user into a second language model;
a proposal generation unit that generates a proposal regarding a product or service according to the user based on the generated plurality of first tags and the generated plurality of second tags;
Equipped with
the proposal generation unit generates a proposal regarding a product or service suited to the user based on output data obtained by inputting the generated plurality of first tags and the plurality of second tags and a third prompt that instructs generation of a proposal regarding a product or service suited to the user based on the relevance between the plurality of first tags and the plurality of second tags into a third language model;
the first prompt includes specifying a category of the first plurality of tags to be generated;
the second prompt includes specifying a category of the plurality of second tags to be generated;
Chatbot control device . a conversation generation unit that generates a conversation sentence including a proposal for a product or service according to the generated user;
a transmitting unit that transmits the generated conversation sentence to the user's terminal device;
Further provided with
The chatbot control device according to claim 1 . the first prompt includes specifying the number of the plurality of first tags to be generated ;
The chatbot control device according to claim 1 . the second prompt includes specifying the number of the plurality of second tags to be generated ;
The chatbot control device according to claim 1 . the third prompt includes specifying the number of product or service suggestions to generate according to the user;
The chatbot control device according to claim 1 . the second tag generation unit generates the plurality of second tags based on output data obtained by inputting the second prompt, which includes an answer from the user to a question included in a conversation sentence with the user, into the second language model.
The chatbot control device according to claim 1 . A chatbot control method for controlling a chatbot that makes suggestions about products or services to a user, comprising:
The computer
generating the plurality of first tags based on output data obtained by inputting information indicating the content of the product or service and a first prompt instructing generation of the plurality of first tags associated with the product or service into a first language model;
generating the plurality of second tags based on output data obtained by inputting information indicating a conversation between the chatbot and the user and a second prompt instructing generation of a plurality of second tags associated with the user into a second language model;
generating a product or service proposal tailored to the user based on the generated plurality of first tags and the generated plurality of second tags;
A chatbot control method , comprising:
generating a proposal for a product or service tailored to the user based on output data obtained by inputting the generated plurality of first tags and the plurality of second tags and a third prompt that instructs the generation of a proposal for a product or service tailored to the user based on the relevance between the plurality of first tags and the plurality of second tags into a third language model;
the first prompt includes specifying a category of the first plurality of tags to be generated;
the second prompt includes specifying a category of the plurality of second tags to be generated;
Chatbot control method . A program for controlling a chatbot that makes product or service suggestions to a user,
On the computer,
generating the plurality of first tags based on output data obtained by inputting information indicating the content of the product or service and a first prompt instructing generation of the plurality of first tags associated with the product or service into a first language model;
generating the plurality of second tags based on output data obtained by inputting information indicating a conversation between the chatbot and the user and a second prompt instructing generation of a plurality of second tags associated with the user into a second language model;
generating a proposal for a product or service tailored to the user based on the generated plurality of first tags and the generated plurality of second tags;
A program ,
generating a proposal for a product or service tailored to the user based on output data obtained by inputting the generated plurality of first tags and the plurality of second tags and a third prompt that instructs the generation of a proposal for a product or service tailored to the user based on the relevance between the plurality of first tags and the plurality of second tags into a third language model;
the first prompt includes specifying a category of the first plurality of tags to be generated;
the second prompt includes specifying a category of the plurality of second tags to be generated;
program .