Information processing device, control method for information processing device, and program

Abstract

This invention provides an information processing device, a control method for the information processing device, and a program that can accurately acquire user information in a more natural conversational format by generating questions based on the content of the conversation with the user. [Solution] The management server 1 includes: a reception processing unit 31 as a reception means for receiving target information which is the content of a conversation with the user; a first cooperation processing unit 36 ​​that outputs a prompt including the target information and commands to a trained model 50 including a large-scale language model and acquires the hearing items output by the trained model 50; a second cooperation processing unit 37 that outputs a prompt including the target information, hearing items and commands to the trained model 50 including a large-scale language model and acquires the hearing content output by the trained model 50; and a presentation processing unit 34 as a presentation means for presenting questions regarding unfilled hearing items based on the hearing items and hearing content.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, a control method for the information processing apparatus, and a program.

Background Art

[0002] Conventionally, information processing technologies for obtaining information from users using chatbots or the like are known. For example, Patent Document 1 describes this type of technology.

[0003] Patent Document 1 relates to an information processing method that contributes to simplifying the interaction with a user regarding the transaction record of products or services on the web. The information processing method of Patent Document 1 includes a bot display step of displaying an interactive chatbot on a predetermined page related to the supply of products or services, a predetermined information acquisition step of receiving an input of predetermined information including identification information, which is information for identifying a user, in response to the output of information through the displayed chatbot, a transaction record information acquisition step of acquiring transaction record information, which is information regarding the user's transaction record of products or services, using the received identification information, a dialogue control step of controlling the output of information through the chatbot based on the acquired transaction record information, and a reaction acquisition step of acquiring the user's reaction to the controlled output of information.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By using chatbots, user information can be acquired efficiently. However, the way user information is acquired using chatbots is in the form of filling out a presented questionnaire, which can hardly be called a natural conversation. While conventional technologies such as Patent Document 1 acquire user responses, there was room for improvement in terms of acquiring user information within the flow of a natural conversation with the chatbot.

[0006] This invention has been made in view of the above circumstances, and aims to provide an information processing device, a control method for the information processing device, and a program that can accurately acquire user information in a more natural conversational format by generating questions according to the content of the conversation with the user. [Means for solving the problem]

[0007] To achieve the above objective, one aspect of the present invention relates to an information processing device having: a receiving means for receiving target information which is the content of a conversation with a user, and for which a plurality of items to be obtained from the user are managed; a first cooperation processing unit which outputs a prompt including the target information and commands to a trained model including a large-scale language model and obtains the hearing items output by the trained model; a second cooperation processing unit which outputs a prompt including the target information, the hearing items and commands to the trained model including the large-scale language model and obtains the hearing content output by the trained model; and a presentation means which presents questions regarding the hearing items that have not been obtained among the plurality of items based on the hearing items and the hearing content.

[0008] Furthermore, one aspect of the present invention is a control method for an information processing device, which includes: a reception step of receiving target information that is the content of a conversation with a user, wherein multiple items to be obtained from a user are managed; a first collaboration processing step of outputting a prompt including the target information and commands to a trained model including a large-scale language model and obtaining the hearing items output by the trained model; a second collaboration processing step of outputting a prompt including the target information, the hearing items and commands to the trained model including the large-scale language model and obtaining the hearing content output by the trained model; and a presentation step of presenting questions regarding the hearing items that have not been obtained among the multiple items, based on the hearing items and the hearing content.

[0009] Furthermore, one aspect of the present invention is a program that causes a computer to execute the following steps, which include: a reception step that receives target information which is the content of a conversation with the user, where multiple items to be obtained from the user are managed; a first collaborative processing step that outputs a prompt including the target information and commands to a trained model including a large-scale language model and obtains the hearing items output by the trained model; a second collaborative processing step that outputs a prompt including the target information, the hearing items and commands to the trained model including the large-scale language model and obtains the hearing content output by the trained model; and a presentation step that presents questions regarding the hearing items that have not been obtained among the multiple items, based on the hearing items and the hearing content. [Effects of the Invention]

[0010] According to the present invention, it is possible to provide an information processing device, a control method for the information processing device, and a program that can accurately acquire user information in a more natural conversational format by generating questions according to the content of the conversation with the user. [Brief explanation of the drawing]

[0011] [Figure 1]This figure shows a transaction support system to which an information processing device according to one embodiment of the present invention is applied. [Figure 2] This is a block diagram showing the hardware configuration of the information processing device according to this embodiment. [Figure 3] This is a functional block diagram showing an example of the functional configuration of the information processing device according to this embodiment. [Figure 4] This flowchart shows an example of the overall flow of the process for acquiring customer information using the information processing device according to this embodiment. [Figure 5] This flowchart shows an example of the selection process from evaluation to action when the previous action is "Next". [Figure 6] This flowchart shows an example of the selection process from evaluation to action when the previous action was concrete. [Figure 7] This flowchart shows an example of the selection process from evaluation to action when the previous action was exploration. [Figure 8] This is a sequence diagram showing the process by which interview items and interview content are obtained from conversations with users. [Figure 9] This flowchart shows an example of the process for scoring useful information. [Figure 10] This is the first example showing a typical chat screen displayed on the user's terminal. [Figure 11] This is a second example showing a typical chat screen displayed on the user's terminal. [Figure 12] This is a third example showing a chat screen displayed on the user's terminal. [Modes for carrying out the invention]

[0012] One embodiment of the present invention will be described below with reference to the drawings.

[0013] <System Configuration> First, the overall system configuration will be described. FIG. 1 is a diagram showing a transaction support system 100 to which a management server 1 according to an embodiment of the present invention is applied. The transaction support system 100 provides a service (hereinafter referred to as "this service") that supports the transactions of providers who provide predetermined goods or services. The users of the transaction support system 100 are not particularly limited, but in the following description, the transaction support system 100 will be described as being used by a service provider S that provides this service, a real estate business operator B that sells or intermediates real estate as a predetermined good or service, and a user U who is a customer for the real estate business operator B.

[0014] The transaction support system 100 is realized by a management server 1 that transmits and receives various information to and from a user terminal 2, a business operator terminal 3, and an AI processing server 4 via a communication network such as the Internet. The management server 1 is a server that performs various processes for supporting transactions with respect to the user terminal 2, the business operator terminal 3, and the AI processing server 4.

[0015] The user terminal 2 is an information processing device used by, for example, a user U who is a customer considering purchasing or renting real estate. The user terminal 2 is composed of a smartphone, a tablet, a personal computer, etc. The user terminal 2 may exchange various information with the management server 1 by a pre-installed program, or may exchange various information through a web browser.

[0016] The business operator terminal 3 is an information processing device used by, for example, a transaction staff member of a real estate business operator who sells or intermediates real estate. The business operator terminal 3 is composed of a tablet, a personal computer, a smartphone, etc. The business operator terminal 3 may exchange various information with the management server 1 by a pre-installed program, or may exchange various information through a web browser.

[0017] The AI ​​processing server 4 is an information processing device that performs information processing using AI (Artificial Intelligence) technology. For example, the AI ​​processing server 4 has a pre-trained model 50 that has undergone training processing, and when a prompt including natural language is input, it has the function of generating and outputting output corresponding to the input prompt based on that prompt. Specifically, for example, the pre-trained model 50 is a large-scale language model capable of generating output as a so-called AI chatbot. The AI ​​processing server 4 is managed by an AI provider (not shown). However, it is not limited to this, and the AI ​​processing server 4 may also be managed by a service provider S. Furthermore, the information processing using AI technology may be performed by the management server 1.

[0018] Specifically, the management server 1 works in conjunction with a pre-trained model 50, which includes a large-scale language model, and engages in conversational interactions with user U using user terminal 2 to acquire customer information that can be used in transactions. The customer information acquired by the management server 1 is shared with transaction staff who handle customer service at real estate businesses that sell or broker real estate.

[0019] The management server 1 may be built on a physical computer, or it may be implemented using a cloud computer built on the internet, event-driven programming, a containerized application, or a combination thereof. In either case, the management server 1 consists of one or more computers.

[0020] <Hardware Configuration> Next, an example of the hardware comprising the management server 1 will be described. Figure 2 is a block diagram showing the hardware configuration of the management server 1 according to this embodiment. The management server 1 includes a CPU (Central Processing Unit) 11 as a processor, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input / output interface 15, an output unit 16, an input unit 17, a storage unit 18, a communication unit 19, and a drive 20.

[0021] The CPU 11 executes various processes according to the program stored in the ROM 12 or the program loaded from the storage unit 18 into the RAM 13. The RAM 13 also stores data necessary for the CPU 11 to execute various processes. The CPU 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output interface 15 is also connected to this bus 14.

[0022] The input / output interface 15 is connected to an output unit 16, an input unit 17, a storage unit 18, a communication unit 19, and a drive 20. The output unit 16 consists of a display, speakers, etc., and outputs various information as images and sounds. The input unit 17 consists of a keyboard, mouse, etc., and inputs various information. The storage unit 18 consists of a hard disk, DRAM (Dynamic Random Access Memory), etc., and stores various data. The communication unit 19 communicates with other devices via a network, including the Internet.

[0023] The drive 20 is appropriately equipped with removable media 21, which may consist of a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory. Programs read from the removable media 21 by the drive 20 are installed in the storage unit 18 as needed. The removable media 21 can also store various types of data stored in the storage unit 18, just like the storage unit 18.

[0024] The hardware configuration described here is merely an example. The computer described in this embodiment, including the management server 1, may have the same configuration as that in Figure 2, or it may have a different configuration. Furthermore, the computer may consist of two or more computers. The user terminal 2 and the business operator terminal 3 in Figure 1 are, for example, smartphones, tablets, or personal computers having a configuration similar to the hardware configuration shown in Figure 2.

[0025] <Functional configuration> Next, the functional configuration of the management server 1 will be described. Figure 3 is a functional block diagram showing an example of the functional configuration of the management server 1 according to this embodiment. In the CPU 11 of the management server 1, the chat execution unit 30, the reception processing unit 31, the intent determination unit 32, the conversation evaluation unit 33, the presentation processing unit 34, the question evaluation unit 35, the first cooperation processing unit 36, the second cooperation processing unit 37, and the useful information registration unit 38 function as functional units. In one area of ​​the storage unit 18 of the management server 1, the hearing item DB 60, the useful information DB 70, and the user information DB 80 are provided.

[0026] The configuration of each functional part of this embodiment will be described below.

[0027] The Hearing Items DB60 stores and manages information on multiple items that are set based on the information that the user wants to be interviewed about and that the user wants to obtain. Hearing items are information that can be used, for example, to propose a specific product or service that is provided by a business that requests the hearing items. In the example of real estate business B, the specific product or service is a real estate transaction. In this case, the hearing items would be annual income, desired area, family structure, floor plan, desired rent, etc. Based on the operation of the real estate business terminal 3 by real estate business B, the hearing items managed in the Hearing Items DB60 are entered or modified.

[0028] The Useful Information DB70 stores and manages information that is useful to the user but is not provided by the business requesting the interview items (in this case, real estate business B), associating it with tags. This information may also include information on administrative services not provided as part of the business, or explanatory information on regional characteristics, etc. Details of the information stored in the Useful Information DB70 will be described later. Note that the information on specified products or services stored in the Useful Information DB70 may also include information on specified products or services provided by the business requesting the interview items. Specifically, for example, information on how users should consider properties when proposing or contracting for properties provided by real estate business B can be considered information on specified products or services provided by real estate business B, and is an example of information that is useful to the user. Thus, information on specified products or services provided by the business requesting the interview items (in this case, real estate business B) may be stored and managed in the Useful Information DB70 as a candidate for useful information, as it is associated with tags.

[0029] The User Information DB80 stores and manages information associated with each user covered by this service. This user information includes, for example, chat logs, memo information, customer information, and useful information tags. Details of the information stored in the User Information DB80 will be described later.

[0030] The chat execution unit 30 is a chat execution means that performs processing to provide conversational responses and questions to information input by the user, in cooperation with the intent determination unit 32 to the useful information registration unit 38, etc., and the trained model 50, which will be described later. Specifically, for example, the chat execution unit 30 performs processing such as asking and answering questions to the user and receiving responses and questions from the user. Here, the chat execution unit 30 is constructed by coordinating with the trained model 50 using, for example, an API (Application Programming Interface). The chat execution unit can also provide the user with conversation content generated using the results of the intent determination unit 32 to the useful information registration unit 38, which will be described later. The chat execution unit 30 sets a chat endpoint corresponding to the user.

[0031] The reception processing unit 31 is a reception means that receives target information, which is the content of a conversation with the user. For example, the reception processing unit 31 receives user messages from the user terminal 2 via the communication network through processing by the chat execution unit 30. That is, the reception processing unit 31 receives raw data (unprocessed information) of the conversation with the user as a chat log. The chat log is stored and managed in the user information DB 80 as part of the user information. For example, the reception processing unit 31 accepts notes related to the chat log. Specifically, the reception processing unit 31 outputs a prompt to the trained model 50 that includes the chat log and a command that includes an instruction to make notes from the chat log (for example, an instruction to summarize including the main points), and accepts the notes output by the trained model 50 as note information. The note information is stored and managed in the user information DB 80 as part of the user information. Chat logs and memo information are examples of relevant information, both representing the content of conversations with users.

[0032] The intent determination unit 32 is an intent determination means that determines the intent of a user's conversation. For example, the intent determination unit 32 outputs a prompt to the trained model 50 that includes target information indicating the content of the user's conversation and a command to determine the intent of the conversation, and obtains the intent identified by the trained model 50. In this embodiment, the intent determination unit 32 determines whether or not the user's message contains a question as the intent.

[0033] The conversation evaluation unit 33 is a conversation evaluation means for evaluating the flow of a conversation. For example, the conversation evaluation unit 33 outputs a prompt to the trained model 50 that includes a command to evaluate the user's message and the flow of the conversation, and the trained model 50 obtains an evaluation of the flow of the conversation. The evaluation may be in a Yes / No format based on predetermined evaluation axes, or it may be a score. In this embodiment, the evaluation of the flow of the conversation is performed from different perspectives depending on the situation of the previous conversation. Details of this evaluation of the flow of the conversation will be described later.

[0034] The presentation processing unit 34 is a presentation means that presents questions regarding unfilled hearing items based on pre-set hearing items and hearing content.

[0035] The interview content will consist of specific details corresponding to the interview item. For example, if the interview item is annual income, the interview content will be specific figures or ranges such as 5 million yen or 10 million yen. Similarly, if the interview item is desired area, the interview content will be specific areas such as Chiyoda Ward. If the interview item is family structure, the interview content will be information that identifies the family structure, such as single person or family of three. If the interview item is floor plan, the interview content will be information that identifies the property details, such as 1LDK. If the interview item is desired rent, the interview content will be specific rent ranges or figures such as within 100,000 yen.

[0036] The method by which the presentation processing unit 34 sets questions for the user is not particularly limited. For example, the presentation processing unit 34 may generate questions by combining pre-set phrases in a rule-based manner to correspond to the unfilled hearing items, or it may select pre-set questions based on a rule-based method. Alternatively, the presentation processing unit 34 may output a prompt to the trained model 50 instructing it to generate questions for the unfilled hearing items, and then obtain questions from the trained model 50 to fill in the unfilled hearing items.

[0037] Furthermore, the presentation processing unit 34 can also stop asking questions to the user if predetermined conditions are met. For example, the predetermined conditions are that the percentage of hearing items for which hearing content has been obtained is above a predetermined level (for example, 80%). In this case, even if not all of the set hearing items have been filled in, if the predetermined conditions are met, no new questions will be presented to the user to fill in the hearing items.

[0038] The predetermined conditions may include factors other than the proportion of interview items. For example, the trained model 50 may be used to determine whether further additional information can be obtained from the customer based on the acquired interview items and interview content, and the predetermined condition may be to stop asking additional questions if it is determined that no new information can be obtained. Alternatively, the trained model 50 may be used to determine the likelihood of a contract based on the conversation content and the acquired interview items and interview content, and questions may be asked if the determination indicates a low likelihood of a contract. Conversely, if the likelihood of a contract is considerably high, the system may stop asking questions and contact the transaction representative. Furthermore, the system may stop asking questions and contact the transaction representative if the user appears bored based on the conversation content.

[0039] The question evaluation unit 35 is a question evaluation means that evaluates the questions presented by the presentation processing unit 34 based on pre-set evaluation axes. For example, the question evaluation unit 35 outputs a prompt to the trained model 50 that includes the questions generated by the presentation processing unit 34 and a command to evaluate the questions, and obtains an evaluation of the questions by analyzing the trained model 50.

[0040] The evaluation axis consists of one or more indicators, for example, an indicator showing whether new information can be obtained from the user, an indicator showing whether the user's current situation and needs will become clearer, and an indicator showing whether it is related to the user's previous conversation. The new information referred to here is the content of the interview corresponding to the unfilled interview items and information related to the interview items. The evaluation axis may also be a combination of the indicator showing whether new information can be obtained from the user, the indicator showing whether the user's current situation and needs will become clearer, and the indicator showing whether it is related to the user's previous conversation. The evaluation may be in a Yes / No format based on the evaluation axis, or it may be a score. The evaluation axis, including these indicators, is included in the prompt output to the trained model 50.

[0041] The first integration processing unit 36 ​​outputs a prompt to the trained model 50 containing commands for extracting target information and interview items, and obtains interview items from the analysis results of the trained model 50. The target information is, for example, text information based on user conversations. The interview items obtained by the first integration processing unit 36 ​​are those that have a high probability of yielding interview content corresponding to the interview items from the target information.

[0042] The commands included in the prompt output by the first collaborative processing unit 36 ​​include, for example, text that indicates pre-set interview items for the trained model 50 and requests the extraction of interview items that can be filled in from the target information. The text is entered as a sentence such as, "What interview items can be filled in?"

[0043] The second linkage processing unit 37 outputs a prompt containing target information, interview items, and commands to the trained model 50, and obtains the interview content from the analysis results of the trained model 50. The interview items included in the prompt output by the second linkage processing unit 37 are the interview items obtained by the processing of the first linkage processing unit 36.

[0044] The commands included in the prompt output by the second coordinating processing unit 37 are commands for extracting interview content corresponding to the interview items. For example, the command contains text that specifies the interview items to the trained model 50 and requests that it extract the interview content corresponding to the interview items from the target information.

[0045] The Useful Information Registration Unit 38 is a storage means that stores candidate information that is useful to the user but is not related to a specified product or service, associating it with tags. Useful information for the user includes, for example, articles on web pages and documents such as PDFs (Portable Document Format). In this embodiment, the useful information for the user is not direct information such as identifying the price of a real estate property, but rather indirect information such as public facilities and private facilities in each area. The useful information and its tags registered in the Useful Information Registration Unit 38 are used by the Presentation Processing Unit 34 to match with tags extracted from the target information. The useful information may also be stored as an overview of the useful information and a method of accessing the useful information (for example, the URL of the web page).

[0046] This section explains how to set tags for useful information. First, morphological analysis is performed on the text indicating useful information, and then a TF (Term Frequency)-IDF (Inverse Document Frequency) score is calculated. Next, an embedding process is performed to convert each noun for which a TF-IDF score has been calculated into numerical vector data, and vectors are generated. The vector obtained by taking a weighted average of the TF-IDF scores for each of the calculated noun vectors becomes the tag that is registered in the useful information registration unit 38 in association with the useful information. Tags for matching from target information are also obtained in the same way. Information that identifies the useful information matched for a user is stored and managed in the user information DB 80 as a useful information tag for that user.

[0047] <Overall processing flow> Next, we will refer to Figure 4 and explain the overall processing flow. Figure 4 is a flowchart showing an example of the overall flow of the process by which the management server 1 acquires customer information according to this embodiment. The flowchart in Figure 4 starts when the user initiates a chat on the user terminal 2.

[0048] In step S11, the reception processing unit 31 starts processing the target information (conversation content) entered by the user on the chat screen displayed on the user terminal 2.

[0049] In step S12, the first linked processing unit 36, which works in conjunction with the trained model 50, obtains pre-configured interview items from the target information.

[0050] In step S13, the second linkage processing unit 37, which works in conjunction with the trained model 50, retrieves interview content corresponding to the interview items acquired by the first linkage processing unit 36 ​​from the target information.

[0051] In step S14, the presentation processing unit 34 determines whether the predetermined conditions for stopping questioning have been met. If the predetermined conditions for stopping questioning have not been met, the process proceeds to step S15 (step S14; No). If the predetermined conditions for stopping questioning have been met, the process by the presentation processing unit 34 to ask questions to the user is completed (step S14; Yes).

[0052] In step S15, the presentation processing unit 34 outputs questions to the user terminal 2 to fill in any unfilled hearing items from the set hearing items. The presentation processing unit 34, together with the intent determination unit 32, conversation evaluation unit 33, question evaluation unit 35, and other functional units, generates questions in cooperation with the trained model 50 and conducts a chat-style conversation. After the processing in step S15, the process returns to step S11.

[0053] <Chat-style conversation flow> Next, we will explain the basic operation of the trained model 50 in a chat-style conversation. In this embodiment, the basic operation of the trained model 50 is "evaluation," "selection of action," "execution," and "user response."

[0054] The evaluation process involves "evaluating the user's intent" and "evaluating the flow of the conversation." The action selection process involves choosing an action from four available options. The execution process involves carrying out the action selected during the action selection phase. User response is the process of receiving responses and questions from the user after the execution, and after this process, evaluation is performed again. In other words, evaluation, action selection, execution, and user response are repeated.

[0055] The actions within the basic operation process are classified into four categories: "Next," "Concrete," "Exploration," and "Answer." Next is the action of asking questions about the next question set in the topic. Concrete is the action of asking questions to make the answer to the question more concrete. Exploration is the action of asking another question by developing the question from the previous conversation. Answer is the action of answering the user's question. Answer does not consume a turn and can be interrupted before the three actions of Next, Concrete, and Exploration.

[0056] As described above, the conversation evaluation unit 33 evaluates the conversation from different perspectives depending on the status indicating the situation of the previous conversation. The statuses referred to here are "Next," "Concrete," and "Exploration." Next, referring to Figures 5 to 7, the process of selecting an action from the evaluation of the conversation flow in each case where the previous status is "Next," "Concrete," and "Exploration" will be explained. Note that the communication process between the management server 1 and the trained model 50 will be omitted in the following explanation.

[0057] Figure 5 is a flowchart illustrating an example of the selection process from evaluation to action when the previous action is "Next".

[0058] In step S101, the reception processing unit 31 acquires the user's reply as target information. In step S102, the intent determination unit 32 performs the first stage of evaluation, which is the evaluation of the user's intent. In this embodiment, the evaluation of the user's intent is whether or not the user asked a question. If the intent determination in step S102 is evaluated as indicating that the user has asked a question, the process proceeds to step S103. In step S103, the presentation processing unit 34 performs the action of answering the user's question, which is called Answer.

[0059] If, in the determination of intent in step S102, it is determined that there are no questions from the user, the process proceeds to step S104. In step S104, the conversation evaluation unit 33 performs the second stage of evaluation, which is the evaluation of the flow of the conversation, by evaluating the answers.

[0060] In the evaluation of responses in step S104, the user's responses are classified into "Answered," "Ambiguous," "Refused to Answer," and "Different Topic." "Answered" is selected when the response to the interview item is specific and the interview item can be completed. It is also selected when the user's current situation is unrelated to the interview item and no further questioning is needed. To avoid a situation where the conversation never ends, "Answered" is also selected when a decision cannot be made. "Ambiguous" is selected when the response to the interview item is ambiguous and further questioning is needed. "Refused to Answer" is selected when the user refuses to answer the interview item. "Different Topic" is selected when the situation does not fall into any of the above categories: Answered, Ambiguous, or Refused to Answer.

[0061] If the user's conversation content is determined to have been answered in the response evaluation in step S104, the process proceeds to step S105. In step S105, the presentation processing unit 34 creates a candidate question. Subsequently, in step S106, the question evaluation unit 35 evaluates whether the question is a pass or fail based on pre-set evaluation criteria. If the question is a pass as a result of the evaluation in step S106, the process proceeds to step S107. In step S107, the presentation processing unit 34 executes "Exploration," which develops the previous conversation question into another question, and the status becomes "Exploration." This results in another question being asked that develops from the previous conversation question.

[0062] If the user's conversation is deemed ambiguous during the response evaluation in step S104, the process proceeds to step S108. In step S108, the presentation processing unit 34 executes the Concrete process, which is used to concretize the response, and the status becomes Concrete. This triggers a question to further clarify the response.

[0063] If the user refuses to answer or the topic is changed during the response evaluation in step S104, the process proceeds to step S109. Similarly, if the question is evaluated as unsuccessful in step S106, the process also proceeds to step S109. In step S109, the Next command is executed to ask questions about the next hearing item, and the status becomes Next. This initiates questioning about a different hearing item than the one for which questions were previously asked.

[0064] Figure 6 is a flowchart illustrating an example of the selection process from evaluation to action when the previous action was "Concrete".

[0065] In step S111, the reception processing unit 31 acquires the user's reply as target information. In step S112, the intent determination unit 32 performs the first stage of evaluation, which is the evaluation of the user's intent. In this embodiment, the evaluation of the user's intent is whether or not the user asked a question. If the intent determination in step S112 is evaluated as indicating that the user has asked a question, the process proceeds to step S113. In step S113, the presentation processing unit 34 performs the action of answering the user's question, which is called Answer.

[0066] If, in the determination of intent in step S112, it is evaluated that there are no user questions, the process proceeds to step S114. In step S114, the conversation evaluation unit 33 performs an evaluation of concretization as the second stage of evaluation, which is an evaluation of the flow of conversation. The conversation evaluation unit 33 evaluates whether the hearing items can be filled in as a result of attempting to concretize the answer given in the previous step. For example, the conversation evaluation unit 33 outputs a prompt that includes a command to evaluate whether the hearing items can be filled in with Yes or No to the trained model 50. In addition, in order to avoid a situation where the conversation does not end, the command may also include an instruction to evaluate with Yes if the judgment is difficult.

[0067] If the specificity evaluation in step S114 determines that the user's conversation is specific, the process proceeds to step S115. In step S115, the question evaluation unit 35 performs an evaluation of the potential for development as an evaluation of the flow of the conversation.

[0068] In evaluating the potential for development, first, the presentation processing unit 34 creates advanced questions. Next, the question evaluation unit 35 performs processing to determine whether the advanced questions satisfy pre-set conditions. For example, the question evaluation unit 35 sets commands to output evaluation indicators based on pre-set evaluation axes for the trained model 50.

[0069] If it is determined that there is potential for development in the evaluation of potential for development in step S115, the process proceeds to step S116. In step S116, the presentation processing unit 34 performs an "Exploration" action, which is asking the user the question that was evaluated as having potential for development, and the status becomes "Exploration".

[0070] If it is determined that there is no potential for development in the evaluation of potential for development in step S115, the process proceeds to step S117. In step S117, the presentation processing unit 34 performs the Next action, which involves asking questions about the next hearing item, and the status becomes Next.

[0071] If the specificity evaluation in step S114 determines that the user's response lacks specificity, the process proceeds to step S118. In step S118, the presentation processing unit 34 executes the Concrete process, which is used to make the response more specific, and the status becomes Concrete.

[0072] Figure 7 is a flowchart illustrating an example of the selection process from evaluation to action when the previous action was exploration.

[0073] In step S121, the reception processing unit 31 acquires the user's reply as target information. In step S122, the intent determination unit 32 performs the first stage of evaluation, which is the evaluation of the user's intent. In this embodiment, the evaluation of the user's intent is whether or not the user asked a question. If the intent determination in step S122 is evaluated as indicating that the user has asked a question, the process proceeds to step S123. In step S123, the presentation processing unit 34 performs the action of answering the user's question, which is called Answer.

[0074] In step S122, if the intent determination is evaluated as having no user questions, the process proceeds to step S124. In step S124, the presentation processing unit 34 creates candidate questions. Subsequently, in step S125, the question evaluation unit 35 evaluates whether the question is a pass or fail based on pre-set evaluation axes. If the evaluation in step S125 results in the question being a pass, the process proceeds to step S126. In step S126, the presentation processing unit 34 executes "Exploration," which develops the question from the previous conversation into another question, and the status becomes "Exploration." This results in another question being asked that develops from the question from the previous conversation.

[0075] If the question is evaluated as unsuccessful in step S125, the process proceeds to step S127. In step S127, the presentation processing unit 34 executes Next, which involves asking a question about the next hearing item, and the status becomes Next. This results in a question being asked about a different hearing item than the one for which the previous question was asked.

[0076] <Obtaining interview items and interview content> Next, referring to Figure 8, the details of how the first and second collaboration processing units 36 and 37 obtain interview items and interview content will be explained. Figure 8 is a sequence diagram showing the flow of how interview items and interview content are obtained from conversations with the user. The sequence diagram shown in Figure 8 is started, for example, when the reception processing unit 31 receives a request from the chat execution unit 30 to create user information and select useful information tags.

[0077] In step A1, the reception processing unit 31 outputs a prompt to the trained model 50 requesting the creation of a memo, which will be the target information for analysis based on the most recent conversation with the user for which user information is to be created. In step A2, the trained model 50 analyzes the most recent conversation based on the prompt output by the reception processing unit 31, generates a memo, and outputs the memo to the first linkage processing unit 36.

[0078] In step A3, the first collaborative processing unit 36 ​​outputs a prompt to the trained model 50 requesting it to select interview items from the memo. In step A4, the trained model 50 analyzes the memo based on the prompt received from the first collaborative processing unit 36 ​​and outputs a list of interview items to be updated to the first collaborative processing unit 36. As a result, the first collaborative processing unit 36 ​​obtains the interview items extracted from the target information.

[0079] In step A5, the second linkage processing unit 37 outputs a prompt requesting the extraction of interview content corresponding to the interview items, based on the interview items to be updated and the memo created in this step. In step A6, the trained model 50 analyzes the memo based on the prompt received from the second linkage processing unit 37 and outputs data to be entered as interview content corresponding to the interview items to the second linkage processing unit 37. As a result, the second linkage processing unit 37 obtains the interview content corresponding to the interview items extracted from the target information.

[0080] In this embodiment, the series of processes in steps A5 and A6 are performed for each hearing item to be updated. The series of processes in steps A5 and A6 may be performed sequentially for each hearing item, or they may be performed in parallel for each hearing item.

[0081] In step A7, the presentation processing unit 34 outputs a prompt requesting the trained model 50 to score the useful information. In step A10, the trained model 50 scores the useful information based on the prompt received from the presentation processing unit 34 and outputs the score to the presentation processing unit 34.

[0082] The series of processes in steps A1 to A6 and the series of processes in steps A7 to A8 are performed in parallel. The processes in steps A1 to A8 score the user's information and useful information.

[0083] <Scoring of Useful Information> Next, referring to Figure 9, the scoring process of useful information by the presentation processing unit 34, corresponding to the processes in steps A7 and A8 of Figure 8, will be described. Figure 9 is a flowchart showing an example of the useful information scoring process. The flowchart in Figure 9 is initiated, for example, by a request for a useful information tag.

[0084] In step S131, the presentation processing unit 34 determines the comment type of the target information. The comment type determination determines whether it is text or a link. In the comment type determination in step S131, if the target information is a link, the presentation processing unit 34 proceeds to step S132, and if the target information is a link, the processing proceeds to step S136.

[0085] In step S132, the presentation processing unit 34 analyzes the content of the linked page. In the following step S133, the presentation processing unit 34 determines whether the linked page content is a PDF or something else. If it is determined to be a PDF in step S133, the presentation processing unit 34 extracts text from the PDF in step S134 and then proceeds to step S136. If it is determined to be something other than a PDF in step S133, the presentation processing unit 34 extracts text from the linked web page in step S135 and then proceeds to step S136.

[0086] In step S136, morphological analysis is performed on the text, and nouns are extracted from the target information. For morphological analysis, a morphological analysis library such as MeCab or a morphological dictionary such as IPADic may be used.

[0087] In step S137, the presentation processing unit 34 calculates a TF-IDF score for each extracted noun (word). TF represents the frequency with which the word appears in the target document. IDF represents the rarity of documents containing the word among all documents. In TF-IDF, the score is calculated based on the two indicators, TF and IDF, so that words that appear frequently in the target document and are infrequently found in other documents receive higher scores.

[0088] In step S138, the presentation processing unit 34 performs an embedding process to convert each noun (word) for which a TF-IDF score has been calculated into numerical vector data, thereby generating a vector.

[0089] In step S139, the presentation processing unit 34 calculates a vector by taking a weighted average of the TF-IDF scores for each of the calculated noun vectors.

[0090] In step S140, the presentation processing unit 34 outputs the tag with the highest similarity from among the vectorized tag candidates that have been pre-registered in the useful information registration unit 38.

[0091] In step S141, the presentation processing unit 34 presents useful information to the user based on the tags and scores of the useful information. For example, the presentation processing unit 34 extracts useful information such as articles about properties corresponding to tags with a score above a certain level and high similarity from the useful information registration unit 38 and presents it to the user terminal 2.

[0092] <Example of chat screen> Next, we will describe an example of how the chat screen displayed on user terminal 2 can be shown. The chat screen may be displayed by launching an application running on user terminal 2, or it may be displayed through a web browser.

[0093] Figure 10 is a first display example showing an example of a chat screen displayed on user terminal 2. Figure 10 shows a topic selection unit 101 that allows the user to select a topic, a question display 102 that confirms the content of the hearing regarding rent, and a hearing content selection unit 103 that allows the user to select the content of the hearing. The topic selection unit 101 is a selection means that allows the user to select a topic, and in this example, the user has selected "marriage" as the topic. The question display 102 displays a question to the user asking about rent as the specific hearing content corresponding to the hearing item, rent. The hearing content selection unit 103 is a selection means that allows the user to select a range of rent in order to obtain the user's answer corresponding to the question display 102. The hearing content selection unit 103 shows numerical ranges such as less than 30,000 yen, 30,000 yen or more but less than 50,000 yen, 50,000 yen or more but less than 70,000 yen, 70,000 yen or more but less than 100,000 yen, and 100,000 yen or more but less than 150,000 yen.

[0094] Figure 11 is a second display example showing an example of the chat screen displayed on user terminal 2. Figure 11 shows the user's rent answer display 111 in response to a question, and a floor plan question display 112 to identify the content of the next hearing item based on the content of the user's rent answer display 111. In this example, the rent answer display 111 shows that the user's answer is that the rent is 70,000 yen or more and less than 100,000 yen, based on the selection operation of the hearing content selection unit 103. The floor plan question display 112 displays a question to confirm the floor plan as a question to fill in the next hearing content. The floor plan question display 112 shows specific floor plans that are to be filled in if the hearing item is floor plan, such as 1R-1K, 2DK-2LDK, 3DK-3LDK, 4DK-4LDK, and 5DK or more.

[0095] Figure 12 is a third display example showing an example of a chat screen displayed on user terminal 2. Figure 12 shows the user's floor plan answer display 121 corresponding to the floor plan question display 112, a detailed question display 122 that displays a question based on the floor plan answer display 121, a detailed question answer display 123 that answers the detailed question display 122, and a specific question display 124 based on the detailed question answer display 123. In this example, the floor plan answer display 121 shows that it is 5DK or larger as the answer to the floor plan question display 112. The detailed question display 122 shows a question asking the user how they use the rooms. The detailed question answer display 123 shows that the user wants a room for work as their answer to the detailed question display 122. The specific question display 124 shows a more advanced question. This is because the user's response (the answer to detailed question 123) is specific enough to fill in the interview items (e.g., floor plan), and therefore, in relation to the aforementioned "Exploration," it is preferable for the generating AI to ask further follow-up questions regarding the interview items (e.g., floor plan).

[0096] The first and second linkage processing units 36 and 37 acquire the user's customer information from conversations with the user as shown in Figures 10 to 12, and the presentation processing unit 34 asks questions to the user.

[0097] As described above, the management server 1 of this embodiment includes: a reception processing unit 31 as a reception means for receiving target information which is the content of a conversation with a user; a first cooperation processing unit 36 ​​that outputs a prompt including the target information and commands to a trained model 50 including a large-scale language model and acquires the hearing items output by the trained model 50; a second cooperation processing unit 37 that outputs a prompt including the target information, hearing items and commands to the trained model 50 including a large-scale language model and acquires the hearing content output by the trained model 50; and a presentation processing unit 34 as a presentation means for presenting questions regarding unfilled hearing items based on the hearing items and hearing content.

[0098] Furthermore, the control method for the management server 1 in this embodiment includes: a reception step of receiving target information which is the content of a conversation with the user; a first cooperation processing step of outputting a prompt including the target information and commands to a trained model 50 including a large-scale language model and acquiring the hearing items output by the trained model 50; a second cooperation processing step of outputting a prompt including the target information, hearing items and commands to the trained model 50 including a large-scale language model and acquiring the hearing content output by the trained model 50; and a presentation step of presenting questions regarding unfilled hearing items based on the hearing items and hearing content. Furthermore, the program of this embodiment causes a computer to execute the steps included in the control method for the management server 1.

[0099] In this way, with the management server 1, its control method, and program configured, a question-based generation AI, rather than a typical answer-based generation AI, can fill in the interview content through natural conversation, avoiding a mere interrogation-like Q&A. Furthermore, by performing the extraction of interview items from the target information (which is the content of the conversation) and the extraction of interview content corresponding to those items from the target information in two stages, it is possible to extract interview content that is likely to be filled in accurately. In addition, while customers may hesitate to provide personal information to their transaction representatives (e.g., sales representatives), the question-based generation AI allows for the acquisition of the user's customer information (interview items and interview content) without resistance. Moreover, transaction representatives may want to avoid handling customers that are unlikely to result in a sale from an efficiency standpoint. For example, in real estate sales transactions, the closing rate is very low (e.g., 95%), so there is a desire to know the likelihood of a sale in advance. In this regard, the configuration of this embodiment makes it possible to streamline the process of obtaining customer information in advance and to take appropriate action according to the likelihood of a sale. In this way, it is possible to streamline customer service that bridges the gap between the transaction representative and the user, leading to successful deals.

[0100] Furthermore, in this embodiment, the presentation processing unit 34 does not present any further questions if the interview items and interview content meet predetermined conditions. This allows for a smooth connection of users who have completed sufficient interviews to the transaction representative. For example, it avoids situations where unnecessary questions are presented to the user when there are no more questions to fill in or when sufficient customer information has been gathered for customer service.

[0101] Furthermore, in this embodiment, the presentation processing unit 34 determines whether the question candidates generated based on a pre-set evaluation axis are acceptable or unacceptable, and if a question candidate is acceptable (if the predetermined conditions are met), it asks the question to the user. This reduces the likelihood of the user giving unintended answers and prolonging the conversation due to questions that do not match the evaluation axis, and speeds up and streamlines the process of obtaining necessary customer information from the user. In addition, as in this embodiment, by making the evaluation axis an index consisting of one or more of the following: an index indicating whether new information is likely to be obtained, an index indicating whether the current situation and needs will be clarified, and an index indicating whether it is related to the content of the previous conversation, it is possible to efficiently conduct interviews by limiting them to questions that broaden the content. Furthermore, by including in the evaluation axis an index that eliminates similar questions, such as an index indicating whether it is related to the content of the previous conversation, it is possible to avoid asking the user the same questions and achieve efficient interviews.

[0102] Furthermore, the management server 1 of this embodiment further includes a conversation evaluation unit 33 as a conversation evaluation means for evaluating the content of the user's conversation, and the presentation processing unit 34 presents questions that develop the conversation content based on the evaluation of the conversation evaluation unit 33. This allows for developing the conversation according to the content of the conversation and deepening the customer's information. For example, if the hearing item is the number of people who plan to live there, and the hearing content is the number of people, the answer obtained may be 3 people. In this case, by developing the questions regarding the number of people who plan to live there, such as whether it is a nuclear family or not, whether it is a two-family house or not, or whether it is a shared house or not, more detailed customer information can be obtained. Alternatively, by asking developmental questions to confirm the user's preferences and hobbies, it becomes possible to propose transactions that match the user's preferences and hobbies. For example, in a real estate transaction, if the user's hobby is golf, properties near car sharing services or highway interchanges can be proposed, or if the user's hobby is weight training, properties near gyms can be proposed.

[0103] Furthermore, the management server 1 of this embodiment further includes a conversation evaluation unit 33 as a conversation evaluation means for evaluating the user's answers, and the presentation processing unit 34 presents a new question when the conversation evaluation unit 33 evaluates the user's answer as ambiguous. This prevents situations where inappropriate questions are asked while the customer's answer is still ambiguous. For example, if the second linkage processing unit 37 cannot appropriately extract the hearing content, it can ask a question to prompt the user to provide a more specific answer. If the answer regarding the area where the user plans to live is along the Chuo Line, it can ask a question to confirm the nearest station. Similarly, if the first linkage processing unit 36 ​​cannot appropriately extract the hearing items, the presentation processing unit 34 can ask a question to prompt the user to provide a more specific answer.

[0104] Furthermore, in this embodiment, if the target information includes a question from the user, the presentation processing unit 34 answers that question and then presents a new question. This enables a more natural conversation with the user than simply filling out a survey form with a chatbot.

[0105] Furthermore, the management server 1 of this embodiment has a useful information registration unit 38 as a storage means that stores candidate information that is useful to the user but is not related to the specified product or service, associating it with tags. The presentation processing unit 34 matches the target information with tags and presents the user useful information associated with the matched tags. As a result, appropriate tags are selected from the conversation with the user, and useful information for the user, such as parks and daycare centers near properties in real estate, can be provided. In addition, since information that is not directly related to the specified product or service is presented, it is possible to avoid situations in which information that may affect future transactions is unintentionally provided to the user.

[0106] Although one embodiment of the present invention has been described above, the present invention is not limited to the embodiments described above, and any modifications, improvements, etc. that can achieve the objectives of the present invention are included in the present invention.

[0107] For example, the prompt output by the first linkage processing unit 36 ​​and the prompt output by the second linkage processing unit 37 may be combined into the same prompt and output to the trained model 50. In this case, the trained model 50 will extract interview items from the target information, and then extract interview content based on the extracted interview items and the source target information. In this way, the first linkage processing unit 36 ​​and the second linkage processing unit 37 can also be configured as a single linkage processing unit. That is, the following technical ideas can be grasped from the description of this embodiment. A means of receiving target information, which is the content of conversations with the user, A collaborative processing unit that outputs a prompt to a trained model including a large-scale language model, which includes a command to extract the target information and interview items and to extract the interview content corresponding to the extracted interview items, and acquires the interview items and interview content output by the trained model, A presentation means for presenting questions regarding the unfilled hearing items based on the aforementioned hearing items and the content of the hearing, An information processing device having

[0108] Furthermore, the series of processes described above can be executed by hardware or by software. In other words, the functional configuration described above is merely illustrative and not particularly limiting. That is, it is sufficient that the management server 1 is equipped with the functionality to execute the series of processes described above as a whole, and the type of functional block used to realize this functionality is not particularly limited to the example above. Also, the location of the functional block is not particularly limited and can be arbitrary. For example, the functional block of the management server 1 may be transferred to another device, etc. Conversely, the functional block of another device may be transferred to a server, etc. Also, a single functional block may be composed of hardware alone, software alone, or a combination of both.

[0109] When a series of processes are executed by software, the programs that make up that software are installed on a computer or other device from a network or storage medium. The computer may be a computer built into dedicated hardware. Alternatively, the computer may be a computer capable of performing various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.

[0110] Such recording media containing programs may consist not only of removable media (not shown) distributed separately from the main device to provide the programs, but also of recording media provided pre-installed in the main device. Since programs can be distributed via a network, the recording media may be installed on or accessible from a computer connected to or capable of connecting to a network.

[0111] In this specification, the step of describing a program to be recorded on a recording medium includes not only processes that are performed chronologically in that order, but also processes that are not necessarily performed chronologically, but are executed in parallel or individually. Furthermore, in this specification, the term "system" refers to an overall system composed of multiple devices, means, etc. [Explanation of Symbols]

[0112] 1. Information Processing Device 2 User terminals 3. Business Operator Terminal 30 Chat Execution Unit 31 Reception Processing Section 32 Intention determination section 33 Conversation Evaluation Department 34 Presentation Processing Unit 35 Question and Evaluation Department 36. First Interoperability Processing Unit 37 Second Interoperability Processing Unit 38 Useful Information Registration Section 100 Transaction Support Systems

Claims

1. Multiple items that we want to retrieve from the user are managed, A receiving means for receiving target information including the content of the conversation with the user, A first collaborative processing unit outputs a prompt containing the aforementioned target information and commands to a trained model including a large-scale language model, and acquires the hearing items output by the trained model. A second collaborative processing unit outputs the target information, the hearing items, and a prompt including commands to the trained model including the large-scale language model, and acquires the hearing content output by the trained model. A presentation means that presents questions regarding the hearing items that have not been obtained among the multiple items, based on the hearing items and the content of the hearing, An information processing device having

2. The aforementioned presentation means will not present any further questions if the aforementioned hearing items and the aforementioned hearing content meet predetermined conditions. The information processing apparatus according to claim 1.

3. The presentation means determines whether a candidate question generated based on a pre-set evaluation axis is acceptable or unacceptable, and if the candidate question is deemed acceptable, it executes the question to the user as a predetermined condition. The information processing apparatus according to claim 2.

4. The aforementioned evaluation axis includes, as an indicator, whether the content of the candidate questions can further fill in the information gathered during the interview. The information processing apparatus according to claim 3.

5. The system further includes a conversation evaluation means for evaluating the content of the user's conversation, The presentation means presents questions that develop the conversation content based on the evaluation of the conversation evaluation means. The information processing apparatus according to claim 1.

6. The system further includes a conversation evaluation means for evaluating the user's response, The presentation means presents a new question when the conversation evaluation means evaluates the user's response as ambiguous. The information processing apparatus according to claim 1.

7. The aforementioned presentation means, if the target information includes a question from the user, answers that question and then presents a new question. The information processing apparatus according to claim 1.

8. The aforementioned hearing is for the purpose of allowing businesses to propose specific products or services. The system has a storage means for storing, in association with tags, candidate information related to the specified goods or services that the aforementioned business operator does not provide, which would be useful to the aforementioned user. The presentation means matches the target information with the tags and presents information useful to the user associated with the matched tags. The information processing apparatus according to claim 1.

9. A reception step that receives the target information, which is the content of the conversation with the user, A first collaborative processing step involves outputting a prompt containing the aforementioned target information and commands to a trained model including a large-scale language model, and obtaining the hearing items output by the trained model. A second collaborative processing step involves outputting the target information, the hearing items, and a prompt including commands to the trained model including the large-scale language model, and acquiring the hearing content output by the trained model. A presentation step in which questions regarding the unfilled hearing items are presented based on the aforementioned hearing items and the content of the hearing, A control method for an information processing device, including the device itself.

10. A reception step that receives the target information, which is the content of the conversation with the user, A first collaborative processing step involves outputting a prompt containing the aforementioned target information and commands to a trained model including a large-scale language model, and obtaining the hearing items output by the trained model. A second collaborative processing step involves outputting the target information, the hearing items, and a prompt including commands to the trained model including the large-scale language model, and acquiring the hearing content output by the trained model. A presentation step in which questions regarding the unfilled hearing items are presented based on the aforementioned hearing items and the content of the hearing, A program that causes a computer to execute something.

Images