Conversation support system, conversation support method, and conversation support program

Abstract

The conversation support system comprises at least one processor. The at least one processor acquires target request information indicating a target request for which an answer is to be generated, determines whether all of the one or more items necessary to generate target answer information are included in the target request information, and if it is determined that all of the one or more items are included in the target request information, it inputs the target request information to the generation AI and executes a first generation process to generate target answer information. If it is determined that all of the one or more items are not included in the target request information, it executes a second generation process and outputs the generated target answer information. The second generation process includes the steps of outputting question information indicating a question about the missing items, acquiring additional information indicating the user's answer to the question information, and inputting the target request information and the additional information to the generation AI to generate target answer information.

Description

Technical Field

[0005]

[0001] The present disclosure relates to a conversation support system, a conversation support method, and a conversation support program.

Background Art

[0002] Techniques for supporting conversations with users have been conventionally known. For example, Patent Document 1 describes a computer system that executes a process of receiving a question sentence from a user and outputting an answer sentence to the question sentence via a chatbot. In this computer system, the question sentence received from the user is analyzed by a large language model such as GPT, for example, and the above answer sentence is generated.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the above-described computer system, in order to generate an accurate answer, the question sentence from the user needs to contain sufficient information to generate an answer. That is, when inputting a prompt to a generative AI to generate an answer, the amount of information contained in the prompt affects the accuracy of the generated answer. For example, when a prompt with insufficient information is input to a generative AI, the accuracy of the generated answer will decrease. Therefore, a technique for suppressing a decrease in the accuracy of the generated answer is desired.

Means for Solving the Problems

[0006] A conversation support method relating to one aspect of this disclosure is executed by a conversation support system equipped with at least one processor. This conversation support method includes the steps of: storing in advance in an item database multiple combinations of request information indicating a request from a user and item information including one or more items necessary to generate response information indicating a response to the request; acquiring target request information indicating a target request which is the request to be answered; acquiring target item information which is item information corresponding to the target request information by referring to the item database; determining whether all of the one or more items included in the target item information are included in the target request information; if it is determined that all of the one or more items included in the target item information are included in the target request information, executing a first generation process which inputs the target request information as a first prompt to the generation AI and generates target response information indicating a response to the target request; if it is determined that all of the one or more items included in the target item information are not included in the target request information, executing a second generation process which generates target response information based on the target request information and the missing items in the target request information; and outputting the generated target response information. The second generation process includes the steps of: outputting question information indicating questions about missing items from one or more items; obtaining additional information indicating user responses to the question information; and inputting the target request information and additional information as a second prompt to the generation AI to generate target response information.

[0007] A conversation support program relating to one aspect of this disclosure causes a computer to perform the following steps: store in advance in an item database multiple combinations of request information indicating a request from a user and item information containing one or more items necessary to generate response information indicating a response to the request; obtain target request information indicating a target request which is the request to be answered; obtain target item information which is item information corresponding to the target request information by referring to the item database; determine whether all of the one or more items included in the target item information are included in the target request information; if it is determined that all of the one or more items included in the target item information are included in the target request information, execute a first generation process to input the target request information as a first prompt to the generation AI and generate target response information indicating a response to the target request; if it is determined that all of the one or more items included in the target item information are not included in the target request information, execute a second generation process to generate target response information based on the target request information and the missing items in the target request information; and output the generated target response information. The second generation process includes the steps of: outputting question information indicating questions about missing items from one or more items; obtaining additional information indicating user responses to the question information; and inputting the target request information and additional information as a second prompt to the generation AI to generate target response information.

[0008] In this respect, if the target request information contains all the items necessary to generate the target response information, that target request information is input to the generation AI as the first prompt, and the target response information is generated. Furthermore, if the target request information does not contain all the items necessary to generate the target response information, question information indicating questions about the missing items is output. Then, additional information indicating the answer to that question is obtained, and this additional information, along with the target request information, is input to the generation AI as the second prompt. Therefore, it is possible to prevent prompts with insufficient information from being input to the generation AI. As a result, it is possible to suppress a decrease in the accuracy of the generated response. [Effects of the Invention]

[0009] According to this disclosure, it is possible to suppress a decrease in the accuracy of the generated responses. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 shows an example of the functional configuration of a conversation support system. [Figure 2] Figure 2 shows an example of a computer hardware configuration used in a conversation support system. [Figure 3] Figure 3 shows an example of the data structure of an item database. [Figure 4] Figure 4 shows an example of the data structure of the response database. [Figure 5] Figure 5 shows an example of the data structure of a contact database. [Figure 6] Figure 6 shows an example of the data structure of a cycle database. [Figure 7] Figure 7 is a schematic diagram illustrating an example of a business cycle. [Figure 8] Figure 8 is a flowchart showing an example of the process for outputting target response information. [Figure 9] Figure 9 is a flowchart showing an example of the second generation process. [Figure 10] Figure 10 shows an example of a screen displayed on a user's terminal. [Figure 11] Figure 11 shows an example of a screen displayed on a user's terminal. [Figure 12] Figure 12 is a flowchart showing an example of a process for outputting transition operation information. [Figure 13] Figure 13 shows an example of a screen displayed on a user's terminal. [Modes for carrying out the invention]

[0011] Hereinafter, various examples in the present disclosure will be described in detail while referring to the attached drawings. In the description of the drawings, the same or equivalent elements are denoted by the same reference numerals, and redundant descriptions are omitted.

[0012] [Overview of the System] The conversation support system according to the present disclosure is a computer system that outputs response information indicating a response to a request from a user. In one example, the conversation support system inputs a prompt for generating the above response to a request from the user to a generative artificial intelligence (Generative Artificial Intelligence) and outputs response information. In the present disclosure, generative AI refers to a technology that processes input data to generate output data indicating predictions, responses, etc., and a prompt refers to the input data input to the generative AI.

[0013] The generative AI is configured using, for example, a neural network model such as a deep learning model. Examples of generative AI include large language models (LLMs) that are constructed using a large amount of datasets and deep learning and scale up three elements: the amount of computation, the amount of data, and the number of model parameters. Another example of generative AI is a vision language model or a large-scale vision language model that can integrally process visual data (images) and text data to generate output data.

[0014] In order to generate an accurate response in such generative AI, the prompt input to the generative AI needs to contain sufficient information for generating the response. That is, when a prompt is input to the generative AI to generate a response, the amount of information contained in the input prompt affects the accuracy of the generated response. For example, when a prompt with insufficient information is input to the generative AI, the accuracy of the generated response will decrease.

[0015] In view of this, in the conversation support system according to the present disclosure, when generating a target response indicating a response to a target desire that is a target of response generation, a item database storing a plurality of combinations of desire information and item information is referred to. Here, the desire information is information indicating a desire from a user, and the item information is information including one or more items necessary for generating response information indicating a response to the desire.

[0016] In the conversation support system according to the present disclosure, by referring to the above item database, it is determined whether all the items necessary for generating target response information indicating a target response are included in the target desire information indicating the target desire. And when not all of the necessary items are included in the target desire information, a question regarding the missing items is asked to the user, and a prompt is input to the generation AI in a state where all of the necessary items are available. Thus, in the conversation support system according to the present disclosure, it is possible to prevent a prompt with insufficient information amount from being input to the generation AI. As a result, it is possible to suppress a decrease in the accuracy of the generated response.

[0017] [Configuration of System] Referring to FIG. 1, an example of application of a conversation support system 10 according to an example will be described. FIG. 1 is a diagram showing an example of a functional configuration of the conversation support system 10.

[0018] In one example, the conversation support system 10 comprises an acquisition unit 11, a determination unit 12, a generation unit 13, and an output unit 14 as functional modules. The acquisition unit 11 is a functional module that acquires target request information indicating the target request. The determination unit 12 is a functional module that determines whether all of the one or more items included in the target item information are included in the target request information. The target item information is information that includes one or more items necessary to generate response information indicating the answer to the target request. The generation unit 13 is a functional module that generates target response information using a language model 15. The language model 15 is an example of the generation AI described above, for example, an interactive AI that enables text-based chat with the user. Examples of such interactive AI include ChatGPT (GPT(registered trademark)-3.5, etc.). The output unit 14 is a functional module that outputs the processing result.

[0019] Figure 2 shows an example of the hardware configuration of a computer 100 that constitutes the conversation support system 10. For example, the computer 100 includes a processor 101, a main memory unit 102, an auxiliary memory unit 103, a communication control unit 104, an input device 105, and an output device 106. The processor 101 executes the operating system and application programs. The main memory unit 102 consists of, for example, ROM and RAM. The auxiliary memory unit 103 consists of, for example, a hard disk or flash memory, and generally stores a larger amount of data than the main memory unit 102. The communication control unit 104 consists of, for example, a network card or a wireless communication module. The input device 105 consists of, for example, a keyboard, mouse, touch panel, etc. The output device 106 consists of, for example, a monitor and a speaker.

[0020] Each functional module of the conversation support system 10 is realized by a conversation support program 110 pre-stored in the auxiliary storage unit 103. Each functional module is realized by loading the conversation support program 110 onto the processor 101 or the main memory unit 102 and having the processor 101 execute the conversation support program 110. The processor 101 operates the communication control unit 104, the input device 105, or the output device 106 according to the conversation support program 110, and reads and writes data to the main memory unit 102 or the auxiliary storage unit 103.

[0021] The conversation support program 110 may be provided on a non-temporary recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory. Alternatively, the conversation support program 110 may be provided via a communication network as a data signal superimposed on a carrier wave.

[0022] The conversation support system 10 may consist of one computer 100 or multiple computers 100. When multiple computers 100 are used, these computers 100 are connected via a communication network such as the Internet or an intranet to logically construct a single conversation support system 10. The conversation support system 10 may also be constructed by combining multiple types of computers.

[0023] Return to Figure 1. In one example, the conversation support system 10 connects to the item database 21, the response database 22, the contact database 23, the cycle database 24, and the user terminal 30 via a communication network such as the Internet or an intranet.

[0024] The item database 21 is a database that stores the item data necessary to generate response information. Figure 3 shows an example of the data structure of the item database. As shown in Figure 3, each record of the item data contains request information and item information. In other words, the item database 21 stores multiple combinations of request information and item information.

[0025] In one example, the requests indicated by the request information include requests related to research or development. Examples of such requests include, for example, "support for research and development," "idea generation," and "patent search." As mentioned above, the item information includes one or more items necessary to generate response information that indicates the answer to the corresponding request. In one example, the number of items included in the item information may differ for each request.

[0026] In the example shown in Figure 3, for the request for "support for research and development," the necessary items to generate an answer to that request include "the purpose of the research," "the technical field in which the research and development will be conducted," and "the problem to be solved through the research and development." For the request for "creation of ideas," the necessary items to generate an answer to that request include "the purpose of idea creation," "the technical field in which the idea was created," "the problem to be solved by the idea," and "the cause of the problem to be solved." For the request for "patent search," the necessary items to generate an answer to that request include "the purpose of the patent search," "the technical field to be searched," "the problem that the invention of the patent to be searched aims to solve," and "the organization that is trying to solve the problem."

[0027] In one example, multiple combinations of request information and item information are stored in the item database 21 in advance. In this example, the storage of multiple combinations of request information and item information is set by the user before the acquisition unit 11 acquires the target request information. The multiple combinations set by the user are sent, for example, from the user terminal operated by the user to the conversation support system 10, and the acquisition unit 11 acquires the transmitted multiple combinations. The acquisition unit 11 then stores the acquired multiple combinations of request information and item information in the item database 21.

[0028] The response database 22 is a database that stores expected response data, which indicates anticipated answers to questions asked when not all required items are included in the target request information. Figure 4 shows an example of the data structure of the response database. As shown in Figure 4, each data record of the response data contains question information and expected response information. In other words, the response database 22 stores multiple combinations of question information and expected response information. The question information indicates a question about an item missing from the request information, and the expected response information indicates an anticipated answer to the question indicated by the corresponding question information. In one example, question information is set for all items stored in the item database 21.

[0029] The questions provided by the question information may, for example, be simple affirmative or negative questions, or they may be questions that ask for more specific answers. If the question is simply an affirmative or negative question, the expected answer will be set to something like "yes" or "no." If the question is a more specific question, the expected answer will be set to that specific answer.

[0030] In one example, multiple combinations of question information and expected answer information are stored in the answer database 22 in advance. In this example, before the acquisition unit 11 acquires the target request information, multiple combinations of question information and expected answer information are set by, for example, the user, and the set combinations are stored in the answer database 22. The multiple combinations set by the user are sent, for example, from the user terminal operated by the user to the conversation support system 10, and the acquisition unit 11 acquires the transmitted multiple combinations. The acquisition unit 11 then stores the acquired multiple combinations of question information and expected answer information in the answer database 22.

[0031] The contact database 23 is a database that stores contact data indicating the contact information of a person in charge. Figure 5 shows an example of the data structure of the contact database. Each data record of the contact data includes a person in charge ID and contact information. In other words, the contact database 23 stores multiple combinations of a person in charge identifier, which is an identifier that identifies a person in charge, and contact information indicating the contact information of that person in charge.

[0032] In one example, a person in charge identifier is issued to everyone belonging to a group that uses the conversation support system 10. The group is a collection of multiple people, for example, an organization composed of multiple members. Such an organization may be a legal entity such as a company, or it may be a department within a legal entity. In this example, the person in charge identifier may be the employee number of each person in charge within the organization, or it may be the name of each person in charge.

[0033] In one example, the contact information includes at least one of the associated contact person's phone number and email address. In this example, the contact information may further include information about the associated contact person's attributes, such as their affiliation.

[0034] In one example, as shown in Figure 5, each data record in the contact data further includes a field of expertise. That is, in this example, the contact database 23 stores multiple combinations of contact person identifiers, contact information, and fields of expertise. In this disclosure, a field of expertise refers to a technical area in which the contact person identified by the corresponding contact person identifier has specialized knowledge, or a business area in which the contact person has business responsibilities. Such fields of expertise may be technical categories such as "materials engineering" and "information systems," or business categories such as "marketing" and "intellectual property." Alternatively, a field of expertise may be a specific project name or business theme within the organization using the conversation support system 10. In one example, each contact person identified by the corresponding contact person identifier may be associated with one field of expertise, or with multiple fields of expertise.

[0035] In one example, multiple combinations of a person in charge identifier and contact information are stored in the contact database 23 in advance. In this example, the storage of multiple combinations of person in charge identifiers and contact information is set by the user before the acquisition unit 11 acquires the target request information. The multiple combinations set by the user are sent, for example, from the user terminal operated by the user to the conversation support system 10, and the acquisition unit 11 acquires the transmitted multiple combinations. The acquisition unit 11 then stores the acquired multiple combinations of person in charge identifiers and contact information in the contact database 23.

[0036] The cycle database 24 is a database that stores cycle data related to business cycles. Figure 6 shows an example of the data structure of the cycle database. In this disclosure, a business cycle refers to a series of step-by-step processes from the start to the end of a business. A business cycle can also be described as a cyclical framework in which tasks and decisions at each stage are carried out sequentially to achieve the objective of the business, and which returns to previous stages or proceeds to the next stage as needed.

[0037] As shown in Figure 6, each data record in the cycle data includes a first-stage identifier, a second-stage identifier, and a keyword. That is, the cycle database 24 stores multiple combinations of the first-stage identifier, the second-stage identifier, and the keyword. In this disclosure, the first-stage identifier is an identifier that identifies a stage in the business cycle. The second-stage identifier is an identifier that identifies the next stage after the stage identified by the first-stage identifier in the business cycle. The keyword is a keyword set for the stage identified by the first-stage identifier, and for example, at least one keyword is set for a single first-stage identifier.

[0038] For example, in the cycle database 24, the stages identified by the first-stage identifier and the second-stage identifier correspond to the stages in the innovation management system. Here, with reference to Figure 7, the stages in the innovation management system and the keywords corresponding to each stage will be explained in more detail.

[0039] Figure 7 is a schematic diagram illustrating an example of a business cycle. An innovation management system is an international standard that provides a framework for organizations to create new value and translate that value into actual business results. The innovation management system includes five stages within the framework described above: "Identifying Opportunities (ST01)", "Creating Concepts (ST02)", "Validating Concepts (ST03)", "Developing Solutions (ST04)", and "Implementing Solutions (ST05)".

[0040] "Identifying opportunities" is the stage of identifying opportunities to generate innovation. For example, at this stage, opportunities to create new innovations are identified by analyzing the organization's internal and external environment. Examples of the organization's internal environment include the organization's existing assets, strengths, and challenges, while examples of the organization's external environment include market trends, technological trends, and social issues. For example, keywords set for "identifying opportunities" include "market analysis," "trend analysis," and "customer insights."

[0041] "Concept creation" is the stage in which a concept is created based on a identified opportunity. For example, in this stage, a concept is created that satisfies a identified need by identifying and analyzing the challenges in that need. Keywords set for "concept creation" include "idea generation," "brainstorming," and "design thinking."

[0042] "Concept validation" is the stage where the created concept is verified to determine whether it is actually feasible. For example, at this stage, it is verified whether the created concept is feasible as a business based on the company's resources, etc., and whether it is technically feasible based on technological trends, etc. Keywords set for "concept validation" include "user testing," "proof of concept (PoC)," and "business model validation."

[0043] "Solution development" is the stage in which an actual product or service is developed based on a validated concept. For example, in this stage, the actual product or service is developed by repeatedly producing prototypes that meet the validated concept and evaluating those prototypes. Keywords set for "solution development" include "prototype," "pilot product," and "detailed design."

[0044] "Solution deployment" is the stage where a developed solution is introduced into the actual market. For example, at this stage, the developed solution is introduced into the market through marketing and sales activities to customers. Keywords used for "solution deployment" include "market introduction," "launch," and "rollout."

[0045] When innovation creation proceeds appropriately, the stages of "opportunity identification," "concept creation," "concept validation," "solution development," and "solution implementation" progress in this order. However, in reality, the stages in an innovation management system do not always progress in the order described above. For example, if a concept is deemed unfeasible during "concept validation," the innovation management system will return to "opportunity identification" or "concept creation" rather than proceeding to "solution development." Thus, the transitions between stages in an innovation management system are reversible, not irreversible. Therefore, in the cycle database 24, one or more second-stage identifiers are associated with a given first-stage identifier, and keywords are set for each combination of first-stage identifiers and second-stage identifiers. In the example shown in Figure 6, the second-stage identifiers associated with the first-stage identifier "ST02" are "ST03" and "ST04," and keywords are set for each combination of the first-stage identifier "ST02" and the second-stage identifier "ST03," and for each combination of the first-stage identifier "ST02" and the second-stage identifier "ST04."

[0046] Returning to Figure 1, the user terminal 30 is a computer used by a user of the conversation support system 10. The user terminal 30 can be any type of computer, such as a tablet, smartphone, laptop computer, or personal computer. The user terminal 30 includes a display device such as a display and an input device such as a keyboard. For example, the user terminal 30 may include a touch panel that functions as both a display device and an input device. The user terminal 30 may be provided as a component of the conversation support system 10, or it may be provided outside of the conversation support system 10.

[0047] [System operation] Next, an example of processing by the conversation support system 10 will be explained, along with the estimation method related to this disclosure. In one example, the conversation support system 10 performs processing to output target response information and processing to output transition operation information. Transition operation information is information that indicates the actions necessary to transition the state of the work being performed by the user to the next stage in the business cycle.

[0048] (Output of target response information) The following section describes the process for outputting target response information, referring to Figure 8. Figure 8 is a flowchart showing an example of the process for outputting target response information.

[0049] In step S11, the acquisition unit 11 acquires the target request information. In one example, the acquisition unit 11 acquires the target request information transmitted from the user terminal 30. In this example, the target request information is entered into the user terminal 30 by the user, and the user terminal 30 transmits the entered target request information to the conversation support system 10. In one example, the target request information may include, in addition to the target request for which a response is to be generated, at least one of one or more items necessary to generate a response to that target request. That is, the user may enter, in addition to their own request, items necessary to generate a response to that request into the user terminal 30.

[0050] In step S12, the acquisition unit 11 acquires the target item information. The acquisition unit 11 refers to the item database 21 and acquires the target item information corresponding to the target request information. In one example, the acquisition unit 11 acquires the target item information as follows. First, the acquisition unit 11 calculates the similarity between the target request information and each request information stored in the item database 21, and identifies the request information with the highest similarity to the target request information. Next, the acquisition unit 11 acquires the item information associated with the identified request information as the target item information.

[0051] The similarity between target request information and request information is an index that indicates how similar the requests indicated by the target request information are to the requests indicated by the request information. The more similar the requests indicated by the target request information and the requests indicated by the request information are, the higher the similarity score will be. For example, the target request information and request information indicate corresponding requests using strings, and the acquisition unit 11 calculates the similarity of these strings as the similarity between the target request information and the request information.

[0052] Alternatively, the acquisition unit 11 may identify request information that contains a specific word in the request indicated by the target request information. In this case, the acquisition unit 11 may acquire the item information associated with the request information containing the specific word as the target item information.

[0053] In step S13, the determination unit 12 determines whether the target request information contains all of the one or more items included in the target item information. For example, the determination unit 12 determines that an item is included in the target request information if the target request information contains a word or phrase corresponding to an item included in the target item information. Note that "word or phrase corresponding to an item" does not necessarily mean only a word or phrase that exactly matches the string of the item. For example, a synonym or a phrase that expresses a similar meaning to the string of the item may be considered a word or phrase corresponding to the item. In this example, the determination unit 12 determines whether the target request information contains all of the one or more items included in the target item information by performing the above determination process for each of the one or more items included in the target item information.

[0054] If it is determined that the target request information includes all of the items included in the target item information (YES in step S13), the process proceeds to step S14. In other words, the process proceeds to step S14 if there is sufficient information to generate a response to the target request.

[0055] If it is determined that the target request information does not include all of the items included in the target item information (NO in step S13), the process proceeds to step S15. In other words, if there is insufficient information to generate a response to the target request, the process proceeds to step S15.

[0056] In step S14, the generation unit 13 executes the first generation process. The first generation process is the process of inputting target request information as a first prompt to the generation AI and generating target response information. In one example, the generation unit 13 executes the first generation process as follows: First, the generation unit 13 inputs target request information containing sufficient information to generate a response to the target request to the language model 15 as a first prompt. Then, the language model 15 generates target response information based on the input first prompt, and the generation unit 13 obtains the generated target response information.

[0057] In step S15, the generation unit 13 executes the second generation process. The second generation process generates target response information based on the target request information and any missing items in the target request information. The second generation process will be explained in detail with reference to Figure 9. Figure 9 is a flowchart showing an example of the second generation process.

[0058] In step S1501, the output unit 14 outputs question information. The question information is information that indicates questions about items that are missing from the target request information, among one or more items included in the target item information. In one example, the generation unit 13 generates question information, and the output unit 14 outputs the generated question information to the user terminal 30. In this example, if there are multiple missing items, the generation unit 13 generates multiple question information, each indicating a question about one of the individual missing items. That is, in this case, each generated question information includes a question about one of the multiple missing items. Alternatively, it may generate question information indicating questions about all of the missing items. That is, in this case, the generated question information indicates a question about each of the multiple missing items.

[0059] For example, the question information may include not only a question about the missing item, but also candidate answers to that question. That is, the output unit 14 may output the question about the missing item and candidate answers to that question as question information. By presenting candidate answers to the question information to the user in this way, the user can easily answer the question information. In such a case, for example, the generation unit 13 may input a prompt to the generation AI to generate candidate answers to the question, thereby generating the candidate answers.

[0060] In step S1502, the acquisition unit 11 acquires additional information. The additional information is information indicating the user's response to the question information. In one example, the acquisition unit 11 acquires additional information transmitted from the user terminal 30. In this example, the additional information is entered by the user into the user terminal 30, and the user terminal 30 transmits the entered additional information to the conversation support system 10. If the question information indicates questions about all the missing items, the additional information only needs to include the answer to at least one of the questions indicated by the question information. That is, in this case, the user does not have to answer all the questions indicated by the question information at once, but may answer them sequentially, for example.

[0061] In step S1503, the generation unit 13 acquires the expected answer information. The generation unit 13 refers to the answer database 22 and acquires the expected answer information corresponding to the question information output in the previous step S1501.

[0062] In step S1504, the generation unit 13 calculates the similarity between the additional information and the expected answer information. The similarity between the additional information and the expected answer information is an indicator of how similar the answer shown by the additional information is to the answer shown by the expected answer information. The more similar the answer shown by the additional information is to the answer shown by the expected answer information, the higher the similarity score. For example, the additional information and the expected answer information indicate corresponding answers using strings, and the generation unit 13 calculates the similarity between these strings as the similarity between the additional information and the expected answer information.

[0063] In step S1505, the determination unit 12 determines whether all of the items missing from the target request information among the one or more items included in the target item information have been acquired as additional information. In one example, the determination in step S1505 is performed in the same way as the determination in step S13. That is, the determination unit 12 determines that the missing item is included in the additional information if the additional information contains a word or sentence corresponding to the missing item. The determination unit 12 performs this determination process for each missing item to determine whether all of the missing items are included in the additional information.

[0064] If it is determined that the additional information includes all of the missing items listed above (YES in step S1505), the process proceeds to step S1506. In other words, the process proceeds to step S1506 if there is sufficient information to generate a response to the target request.

[0065] If it is determined that the additional information does not include all of the missing items listed above (resulting in NO in step S1505), the process proceeds to step S1507. In other words, if there is insufficient information to generate a response to the target request, the process proceeds to step S1507.

[0066] In step S1506, the generation unit 13 generates the target response information. The generation unit 13 inputs the target request information and additional information as a second prompt to the generation AI and generates the target response information. In one example, the generation unit 13 generates the target response information as follows: First, the generation unit 13 inputs the target request information and additional information as a second prompt to the language model 15. Then, the language model 15 generates the target response information based on the input second prompt, and the generation unit 13 obtains the generated target response information.

[0067] In step S1507, the determination unit 12 determines whether the number of times the similarity calculated in step S1504 is above the threshold is greater than or equal to a predetermined number of times. If the number of times the similarity calculated in step S1504 is above the threshold is greater than or equal to a predetermined number of times (YES in step S1507), the process proceeds to step S1508. If the number of times the similarity calculated in step S1504 is above the threshold is less than the predetermined number of times (NO in step S1507), the process returns to step S1501.

[0068] In step S1508, the generation unit 13 generates counter-information. Counter-information is information that shows an answer that is the opposite of the additional information. For example, "an answer that is the opposite of the additional information" refers to an answer that takes the opposite stance or a different perspective to the user's answer to the question information. In this example, the counter-information may show an opposing opinion or an alternative option to the user's answer to the question information, or it may show an opinion that negates or complements the user's answer, or it may show an opinion from a new perspective that the user has not considered.

[0069] The generation unit 13 inputs information instructing the generation of opposing answers as a third prompt to the generation AI, and generates opposing information. In one example, the generation unit 13 generates opposing information as follows: First, the generation unit 13 inputs the above information instructing the generation of opposing answers as a third prompt to the language model 15. Then, the language model 15 generates opposing information based on the input third prompt, and the generation unit 13 acquires the generated opposing information.

[0070] In step S1509, the generation unit 13 identifies the target person identifier, which is the person identifier of the person in charge related to the answer indicated by the opposing information. In one example, the generation unit 13 identifies the target person identifier by referring to the contact database 23. In this example, the generation unit 13 calculates the degree of relevance between the answer indicated by the opposing information and each area of ​​responsibility stored in the contact database 23, and identifies the person identifier associated with the area of ​​responsibility that has the highest degree of relevance to the answer indicated by the opposing information as the target person identifier. The degree of relevance between the answer indicated by the opposing information and the area of ​​responsibility is an indicator that shows how related the answer indicated by the opposing information is to the area of ​​responsibility. The more related the answer indicated by the opposing information is to the area of ​​responsibility, the higher the degree of relevance.

[0071] In step S1510, the generation unit 13 obtains target contact information, which is contact information corresponding to the target person identifier. The generation unit 13 obtains target contact information by referring to the contact database 23.

[0072] In step S1511, the output unit 14 outputs the opposing party information and the target contact information. In one example, the output unit 14 transmits the opposing party information and the target contact information to the user terminal 30. The user terminal 30 receives and displays the transmitted opposing party information and target contact information. This allows the user to find out the contact information of the person in charge who is knowledgeable about the answer indicated by the opposing party information.

[0073] After step S1511 is executed, the process returns to step S1501. In this way, the conversation support system 10 repeatedly outputs question information, acquires additional information, acquires expected answer information, and calculates the similarity between the additional information and the expected answer information until all of the items that are missing from the target request information among the one or more items included in the target item information are acquired as additional information. This ensures that sufficient information is gathered to generate an answer to the target request.

[0074] Returning to Figure 8, in step S16, the output unit 14 outputs the target response information. In one example, the output unit 14 sends the target response information to the user terminal 30. The user terminal 30 receives and displays the transmitted target response information. This allows the user to obtain an answer to their request.

[0075] In one example, a user (user terminal 30) who has referenced the target response information generated as described above may send new target request information to the conversation support system 10. In other words, in this example, the exchange of target request information and target response information between the user and the conversation support system 10 may be repeated. In such cases, for example, the target request information and additional information used to generate a certain target response information (first target response information) may be reused when generating another target response information (second target response information) in the next instance. In one example, if the items necessary to generate the second target response information are included in the previously acquired target request information or additional information, the second target response information may be generated without outputting question information, even if the target request information corresponding to the second target response information does not include those items. In this example, for example, the target request information corresponding to the second target response information and the previously acquired target request information or additional information may be input to the generation AI (language model 15) as a second prompt to generate the second target response information.

[0076] Next, the series of processes from steps S11 to S16, when the number of times the similarity between the additional information and the expected answer information exceeds a threshold is less than a predetermined number, will be explained in more detail using Figure 10 as a specific example. Figure 10 is a diagram showing an example of a screen displayed on the user terminal. Specifically, Figure 10 shows how the exchange between the conversation support system 10 and the user (user terminal 30) is displayed on the display device of the user terminal 30.

[0077] In Figure 10, the target request entered by the user is illustrated as target request information RQ1, and the question information from the conversation support system 10 is illustrated as question information QE1 to QE5. In addition, in Figure 10, additional information showing the user's response to question information QE1 to QE5 is illustrated as additional information RP1 to RP5, and the target response generated by the conversation support system 10 is illustrated as target response information AS1.

[0078] In the example shown in Figure 10, the target request indicated by target request information RQ1 is "I want to look up patents," and target request information RQ1 does not contain any items necessary to generate a response to the target request it indicates. When such target request information RQ1 is sent to the conversation support system 10, the acquisition unit 11 calculates the similarity between the string "I want to look up patents" and the strings of each request information stored in the item database 21, and identifies "patent search," which is the request information with the highest similarity.

[0079] As described above, the target request information RQ1 does not contain the necessary items to generate a response (target answer) to the target request it indicates. Therefore, in the example shown in Figure 10, questions regarding these necessary items are sent to the user terminal 30 as question information QE1 to QE5, and additional information RP1 to RP5 corresponding to each of the five question information QE1 to QE5 is sent from the user terminal 30 to the conversation support system 10.

[0080] As shown in Figure 3, for a request for "patent search," the following items are set as necessary to generate an answer to that request: "Purpose of the patent search," "Technical field to be searched," "Problem that the invention of the patent to be searched aims to solve," and "Organization that is trying to solve the problem." Therefore, in the example shown in Figure 10, "Purpose of the patent search," "Technical field to be searched," "Problem that the invention of the patent to be searched aims to solve," and "Organization that is trying to solve the problem" are presented to the user in order as question information QE1 to QE5.

[0081] Specifically, the user is first asked about the "purpose of the patent search" through question information QE1. In the example shown in Figure 10, the question information includes questions about missing items, as well as candidate answers to those questions. More specifically, candidate answers to the "purpose of the patent search" include "confirmation of whether or not it infringes on other companies' patents," "obtaining hints for product development," and "obtaining hints for problem-solving methods."

[0082] Then, the additional information RP1 is sent to the conversation support system 10 as the user's response to the question information QE1. Subsequently, the question information QE2 asks the user about the "technical field to be investigated" and the "problem that the invention of the patent to be investigated aims to solve." In other words, in the example shown in Figure 10, the question information indicates questions about two or more items that are missing from the target request information.

[0083] In the example shown in Figure 10, the additional information RP2, which is the answer to the question information QE2 concerning two items, "the technical field to be investigated" and "the problem that the invention of the patent to be investigated aims to solve," only shows the answer to "the technical field to be investigated." Therefore, the question information QE3 prompts the user to ask again about "the problem that the invention of the patent to be investigated aims to solve." In this way, if the user does not provide an answer to the question indicated by the question information, the conversation support system 10 (output unit 14) may output question information indicating the same question again to the user terminal 30.

[0084] In the exchange of question information and additional information between the user (user terminal 30) and the conversation support system 10, although the additional information is an answer to the question information, its level of detail may not be sufficient for that answer. In the example shown in Figure 10, the question information QE3 asks the user about "the problem that the invention of the patent to be investigated aims to solve," and the user answers the question with problem "B." However, if the level of detail of problem "B" is not sufficient to generate an answer to the request for "patent investigation," the user may be presented with a question requesting a more detailed answer. That is, the conversation support system 10 may determine whether the level of detail of the acquired additional information is above a predetermined threshold, and if the level of detail is below the threshold, it may output question information indicating a question to request a more detailed answer.

[0085] In the example shown in Figure 10, question information QE4 is output as a question requesting a more detailed answer. Then, additional information RP4 is sent to the conversation support system 10 as the user's response to question information QE4. Subsequently, question information QE5 asks the user about the "organization that is trying to solve the problem".

[0086] Then, the additional information RP5 is sent to the conversation support system 10 as the user's response to the question information QE5. With this, all the necessary items for generating a response to the target request indicated by the target request information are available, so the conversation support system 10 (generation unit 13) inputs the target request information RQ1 and the additional information RP1~RP5 as second prompts to the language model 15. The language model 15 then generates the target response information AS1 based on the input second prompts, and the conversation support system 10 displays the target response information AS1 on the display device of the user terminal 30. The user can refer to the target response information AS1 displayed on the display device and obtain a response to their request. In the example shown in Figure 10, in response to the target request information RQ1, "I want to look up patents," the user is presented with "Documents from Company C: Patent NNNN01, Documents from Company D: Patent NNNN02."

[0087] Next, the series of processes from steps S11 to S16, when the number of times the similarity between the additional information and the expected answer information exceeds a threshold is greater than a predetermined number, will be explained in more detail using Figure 11 as another specific example. Figure 11 is a diagram showing an example of a screen displayed on the user terminal. Specifically, Figure 11 shows how the exchange between the conversation support system 10 and the user (user terminal 30) is displayed on the display device of the user terminal 30.

[0088] In Figure 11, the question information from the conversation support system 10 is shown as question information QE5, and additional information from the user in response to question information QE5 is shown as additional information RP5. The following explanation will assume that the interaction between the user terminal 30 and the conversation support system 10 prior to question information QE5 was the same as the example shown in Figure 10.

[0089] Here, as shown in Figure 11, the question information QE5 is a simple yes or no question: "Companies C and D are major companies that have B1 as a problem. Would you like to investigate how these companies are solving the problem?" In response to the question information QE5, which presents such a question, an affirmative answer, "Yes, please," is sent to the conversation support system 10 as additional information RP5. Here, the affirmative answer "Yes" is set as the expected answer information for the question information QE5. In the example shown in Figure 11, the response "Yes, please," provided by the additional information RP5, is shown to have caused the number of times the similarity between the additional information and the expected answer information exceeds a predetermined threshold to exceed a certain number of times.

[0090] As shown in Figure 11, in the example, the number of times the similarity score indicated by the additional information RP5 exceeds the threshold has exceeded a predetermined number. Therefore, following the additional information RP5, the opposite information, which is the opposite of the answer to the additional information RP5, is shown as the opposite information OI1. Here, the additional information RP5 corresponding to the opposite information OI1 is based on the question information QE5, which asks about a task called "B1". As an example, the opposite information OI1 shows a task called "B2", which is different from "B1", as the opposite answer to the additional information RP5. Furthermore, in the example shown in Figure 11, the contact information of the person in charge related to the task "B2" is shown as the target contact information CI1. In this example, the target contact information CI1 shows the contact information of "Mr. E", who is conducting research on the task "B2".

[0091] In the example shown in Figure 11, a user (user terminal 30) seeking further information regarding such opposing information OI1 sends target request information RQ2 to the conversation support system 10 as new target request information, and the conversation support system 10 outputs target response information AS2 to the user terminal 30 as a response to target request information RQ2. In this case, the conversation support system 10 may use any one of the target request information RQ1 and additional information RP1 to RP5, which were acquired before target request information RQ2, to generate target response information AS2.

[0092] (Output of transition operation information) Next, we will explain the process of outputting transition operation information, referring to Figure 12. Figure 12 is a flowchart showing an example of the process of outputting transition operation information.

[0093] In step S21, the output unit 14 outputs business question information. Business question information is information that indicates questions about the status of the work the user is doing. "Questions about the status of work" are questions that ask, for example, what kind of work the user is currently doing. Such questions may be abstract questions such as, "Please tell me the status of your work." Alternatively, the above questions may be specific questions such as, "Please tell me what experiment you are currently conducting." In step S21, the output unit 14 outputs the business question information to the user terminal 30.

[0094] In step S22, the acquisition unit 11 acquires business status information. Business status information is information indicating the user's response to business question information. In one example, the acquisition unit 11 acquires business status information transmitted from the user terminal 30. In this example, the business status information is entered into the user terminal 30 by the user, and the user terminal 30 transmits the entered business status information to the conversation support system 10.

[0095] In step S23, the generation unit 13 identifies the first-stage identifier. In one example, the generation unit 13 refers to the cycle database 24 to identify the first-stage identifier corresponding to the keyword related to the acquired business status information. In this example, the generation unit 13 calculates the degree of relevance between the response indicated by the business status information and each keyword stored in the cycle database 24, and identifies the first-stage identifier associated with the keyword with the highest degree of relevance to the response indicated by the business status information. In this case, the generation unit 13 may calculate the degree of relevance to each keyword stored in the cycle database 24 using only a portion of the response indicated by the business status information, or it may calculate the degree of relevance to each keyword using the entire response indicated by the business status information.

[0096] The correlation between responses and keywords in business status information refers to an indicator that shows how closely the responses and keywords in business status information are related. The higher the correlation between the responses and keywords in business status information, the higher the correlation score.

[0097] Identifying such a first-stage identifier identifies which stage of the business cycle the user's current task belongs to. As mentioned above, the stage identified by the first-stage identifier is, for example, a stage in the innovation management system. Therefore, it can also be said that in step S23, the generation unit 13 identifies which stage of the innovation management system the user's current task belongs to.

[0098] In step S24, the generation unit 13 generates transition operation information. The generation unit 13 inputs business status information and information requesting the necessary actions to transition the stage indicated by the identified first-stage identifier to the next stage indicated by the second-stage identifier corresponding to the first-stage identifier as a fourth prompt to the generation AI, and generates transition operation information indicating the necessary actions. For the sake of explanation, the information requesting the necessary actions to transition the stage indicated by the identified first-stage identifier to the next stage indicated by the second-stage identifier corresponding to the first-stage identifier may be simply referred to as action request information.

[0099] In one example, the generation unit 13 generates transition operation information as follows: First, the generation unit 13 inputs business status information and operation request information as fourth prompts to the language model 15. Then, the language model 15 generates operation transition information based on the input fourth prompts, and the generation unit 13 acquires the generated operation transition information.

[0100] In step S25, the output unit 14 outputs transition operation information. In one example, the output unit 14 transmits the transition operation information to the user terminal 30. The user terminal 30 receives and displays the transmitted transition operation information. In this example, the transition operation information may further indicate the stage indicated by the first stage identifier identified in step S23, and the stage indicated by the second stage identifier associated with the first stage identifier. That is, the output unit 14 may output the stage to which the user's current task belongs, the destination and next stage, and the actions necessary to transition to the next stage as transition operation information. This allows the user to refer to the transition operation information and transition their task to the next stage, thereby accelerating the business cycle.

[0101] Next, the series of processes from step S21 to step S25, which output the transition operation information described above, will be explained in more detail using Figure 13 as a specific example. Figure 13 is a diagram showing an example of a screen displayed on the user terminal. Specifically, Figure 13 shows how the exchange between the conversation support system 10 and the user (user terminal 30) is displayed on the display device of the user terminal 30 before the transition operation information is output.

[0102] In Figure 13, the business question information output from the conversation support system 10 is illustrated as business question information BI1, and the business status information from the user is illustrated as business status information SI1. In the example shown in Figure 13, the user responds to the business question information with the business status information, "Currently, in order to solve problem A, experiment B was conducted, and result C was obtained." In this example, the conversation support system 10 (generation unit 13) uses, for example, the part "In order to solve problem A, experiment B was conducted," to calculate the degree of relevance to each keyword stored in the cycle database 24 and identify the first-stage identifier.

[0103] Once the first-stage identifier is identified, it is determined which stage of the business cycle the user's current task belongs to. Therefore, the conversation support system 10 inputs the acquired business question information BI1 and information requesting the action necessary to transition from the identified stage to the next stage as the fourth prompt to the language model 15. This generates transition action information TI1 indicating the necessary action, and the generated transition action information TI1 is output to the user terminal 30.

[0104] In the example shown in Figure 13, the transition operation information TI1 indicates the operation required to transition from the specified stage to the next stage, the specified current stage, the next stage to which the transition will occur, and the operation required to transition from the current stage to the next stage. In this example, the current stage is indicated as "Stage D," the next stage is indicated as "Stage E," and the required operation is indicated as "Experiment F." In the example shown in Figure 13, a user (user terminal 30) seeking further information regarding such transition operation information TI1 sends target request information RQ3 to the conversation support system 10 as new target request information, and the conversation support system 10 outputs target response information AS3 to the user terminal 30 as a response to target request information RQ3.

[0105] [effect] As described above, the conversation support system relating to one aspect of this disclosure is a conversation support system that outputs response information indicating an answer to a user's request. This conversation support system comprises at least one processor. The at least one processor pre-stores in an item database multiple combinations of request information indicating a user's request and item information containing one or more items necessary to generate response information indicating an answer to the request, retrieves target request information indicating a target request which is the request to be answered, refers to the item database to retrieve target item information which is item information corresponding to the target request information, determines whether all of the one or more items included in the target item information are included in the target request information, and if it is determined that all of the one or more items included in the target item information are included in the target request information, it inputs the target request information as a first prompt to the generation AI and executes a first generation process to generate target response information indicating an answer to the target request, and if it is determined that all of the one or more items included in the target item information are not included in the target request information, it executes a second generation process to generate target response information based on the target request information and the items missing from the target request information, and outputs the generated target response information. The second generation process includes the steps of: outputting question information indicating questions about missing items from one or more items; obtaining additional information indicating user responses to the question information; and inputting the target request information and additional information as a second prompt to the generation AI to generate target response information.

[0106] A conversation support method relating to one aspect of this disclosure is executed by a conversation support system equipped with at least one processor. This conversation support method includes the steps of: storing in advance in an item database multiple combinations of request information indicating a request from a user and item information including one or more items necessary to generate response information indicating a response to the request; acquiring target request information indicating a target request which is the request to be answered; acquiring target item information which is item information corresponding to the target request information by referring to the item database; determining whether all of the one or more items included in the target item information are included in the target request information; if it is determined that all of the one or more items included in the target item information are included in the target request information, executing a first generation process which inputs the target request information as a first prompt to the generation AI and generates target response information indicating a response to the target request; if it is determined that all of the one or more items included in the target item information are not included in the target request information, executing a second generation process which generates target response information based on the target request information and the missing items in the target request information; and outputting the generated target response information. The second generation process includes the steps of: outputting question information indicating questions about missing items from one or more items; obtaining additional information indicating user responses to the question information; and inputting the target request information and additional information as a second prompt to the generation AI to generate target response information.

[0107] A conversation support program relating to one aspect of this disclosure causes a computer to perform the following steps: store in advance in an item database multiple combinations of request information indicating a request from a user and item information containing one or more items necessary to generate response information indicating a response to the request; obtain target request information indicating a target request which is the request to be answered; obtain target item information which is item information corresponding to the target request information by referring to the item database; determine whether all of the one or more items included in the target item information are included in the target request information; if it is determined that all of the one or more items included in the target item information are included in the target request information, execute a first generation process to input the target request information as a first prompt to the generation AI and generate target response information indicating a response to the target request; if it is determined that all of the one or more items included in the target item information are not included in the target request information, execute a second generation process to generate target response information based on the target request information and the missing items in the target request information; and output the generated target response information. The second generation process includes the steps of: outputting question information indicating questions about missing items from one or more items; obtaining additional information indicating user responses to the question information; and inputting the target request information and additional information as a second prompt to the generation AI to generate target response information.

[0108] In this respect, if the target request information contains all the items necessary to generate the target response information, that target request information is input to the generation AI as the first prompt, and the target response information is generated. Furthermore, if the target request information does not contain all the items necessary to generate the target response information, question information indicating questions about the missing items is output. Then, additional information indicating the answer to that question is obtained, and this additional information, along with the target request information, is input to the generation AI as the second prompt. Therefore, it is possible to prevent prompts with insufficient information from being input to the generation AI. As a result, it is possible to suppress a decrease in the accuracy of the generated response.

[0109] In conversation support systems relating to other aspects, at least one processor may pre-store in an answer database multiple combinations of question information and expected answer information indicating expected answers to the questions indicated by the question information. The second generation process may further include the steps of: referring to the answer database to obtain expected answer information corresponding to the question information output immediately before; calculating the similarity between the obtained expected answer information and additional information; repeating the steps of outputting, obtaining additional information, obtaining expected answer information, and calculating until all missing items are obtained as additional information; and, if the number of times the calculated similarity shows a threshold or higher exceeds a predetermined number of times in the repeated steps, inputting information as a third prompt to the generation AI instructing it to generate an answer opposite to the additional information, thereby generating opposite information indicating an opposite answer; and outputting the generated opposite information. If the similarity between the additional information and the expected answer information exceeds a certain threshold a predetermined number of times, it may indicate that the question is appropriate for an item where the question information is insufficient. On the other hand, because the question information is appropriate, it may be difficult for users to come up with new ideas, and users may not be given the opportunity to reconsider their requests from a different perspective. In contrast, this conversation support system generates contrasting information that is the opposite of the additional information when the number of times the similarity score exceeds a predetermined threshold exceeds a predetermined number. By deliberately outputting such contrasting information, the system can encourage users to change their perspective and provide them with an opportunity to reconsider their requests from a different viewpoint.

[0110] In conversation support systems relating to other aspects, at least one processor may pre-store in a contact database multiple combinations of an employee identifier, which is an identifier that identifies an employee, and contact information indicating the contact information of said employee. The second generation process may further include the steps of: identifying a target employee identifier, which is the employee identifier of the employee related to the response indicated by the opposing information; obtaining target contact information, which is the contact information corresponding to the target employee identifier, by referring to the contact database; and outputting the target contact information. As mentioned above, contrasting information can provide users with new ideas, but users may have little understanding of those ideas. If users have little understanding of contrasting information, they may have difficulty finding a relationship between the information and their own needs, and the output of contrasting information may not provide an opportunity for users to reconsider their needs from a different perspective. In contrast, this conversation support system outputs contact information for the relevant personnel related to the answers indicated by the contrasting information, so that users can refer to this contact information and obtain insights about the contrasting information from those personnel. Therefore, it is possible to reliably provide users with an opportunity to reconsider their own needs from a different perspective.

[0111] In conversation support systems relating to other aspects, at least one processor may pre-store in a cycle database a combination of a first-stage identifier that identifies a stage in a business cycle, a second-stage identifier that identifies the next stage in the business cycle, and a set of keywords for the stage identified by the first-stage identifier. The system may output business question information indicating a question about the status of a task performed by the user, obtain business status information indicating the user's response to the business question information, refer to the cycle database to identify a first-stage identifier corresponding to the keyword related to the business status information, input the business status information and information requesting the action necessary to transition the stage indicated by the identified first-stage identifier to the next stage indicated by the second-stage identifier corresponding to the first-stage identifier as a fourth prompt to the generating AI, generate transition action information indicating the necessary action, and output the generated transition action information. To generate new innovations in business operations, it is necessary to increase the number of cycles in the business process. However, it is difficult for users to identify which stage of the business process their current state belongs to, and even more difficult to grasp the actions required to transition to the next stage. In other words, if users are required to identify the stage and grasp the necessary actions themselves, there is a problem in that it is difficult to increase the number of cycles in the business process. In contrast, this conversation support system acquires business state information and, based on that business state information, identifies the first stage identifier, that is, which stage of the business process the user's current state belongs to. Then, the system generates transition action information, which indicates the actions required to transition from that stage to the next stage, and outputs the generated transition action information. As a result, users can easily grasp the actions required to transition to the next stage by referring to the outputted transition action information. Therefore, it is possible to increase the number of cycles in the business process and improve the possibility of generating new innovations.

[0112] [Differentiation] The technology relating to this disclosure has been described in detail above based on various examples. However, this disclosure is not limited to the examples given above. The technology relating to this disclosure can be modified in various ways without departing from its essence.

[0113] In the embodiment described above, the conversation support system 10 includes an item database 21, a response database 22, a contact database 23, and a cycle database 24. However, the conversation support system 10 does not necessarily include the item database 21, the response database 22, the contact database 23, and the cycle database 24. In this case, the item database 21, the response database 22, the contact database 23, and the cycle database 24 may be located in a separate computer system different from the conversation support system 10.

[0114] In the embodiment described above, in step S1508, the generation unit 13 inputs information instructing the generation of a contrasting answer as a third prompt to the generation AI, thereby generating contrasting information. However, the generation of contrasting information in step S1508 is not limited to the example described above.

[0115] In one example, the generation unit 13 may generate opposing information from the perspective of a field different from the field of expertise of the user who entered the target request information. In this example, opposing information may be generated as follows: First, the generation unit 13 identifies the field of expertise of the user who entered the target request information. In one example, in step S11, the user may input their field of expertise along with the target request information into the user terminal 30. Alternatively, in step S1508, the output unit 14 may display an input screen on the user terminal 30 for the user to input their field of expertise, and the acquisition unit 11 may acquire the field entered by the user on the input screen as the user's field of expertise.

[0116] Next, the generation unit 13 identifies a field of expertise different from that of the user who entered the target request information. In this case, the generation unit 13 may input information indicating the user's field of expertise and information instructing the generation AI to identify a field different from that field as prompts to identify the different field. Alternatively, the generation unit 13 may use natural language processing to vectorize the fields of expertise of each user stored in the contact database 23, and identify a field whose vector direction is different from that of the user who entered the target request information as a field different from that field of expertise.

[0117] Next, the generation unit 13 generates the opposing answers from the perspective of the specified field. In one example, the generation unit 13 may input information to the generation AI as a prompt instructing it to generate the opposing answers from the perspective of the specified field, and generate the opposing information. Alternatively, the generation unit 13 may refer to the contact database 23 and send an input form to the person in charge associated with the specified field to input the information necessary to generate the opposing answers. In this example, for example, the input form may be sent to the contact indicated by the contact information associated with that person in charge, or it may be displayed on the user terminal 30 of that person in charge. The generation unit 13 may generate the opposing information based on the information entered in the input form.

[0118] In step S1508, if the opposing information is generated from the perspective of a field different from the field of responsibility of the user who entered the target request information, in step S1509, the person in charge identifier associated with that identified field may be identified as the target person in charge identifier. In other words, in this case, it is not necessary to calculate the degree of relevance between the answer indicated by the opposing information and each field of responsibility stored in the contact database 23.

[0119] The processing steps for a method executed by at least one processor are not limited to the examples above. For example, some of the steps described above may be omitted, or each step may be performed in a different order. Also, any two or more of the steps described above may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be performed in addition to each of the steps described above.

[0120] In comparing the relative magnitudes of two numerical values ​​in this disclosure, either of the two criteria, "greater than or equal to" and "greater than," may be used, or either of the two criteria, "less than or equal to" and "less than," may be used.

[0121] In this disclosure, the expression "at least one processor executes a first process, a second process, ... and the nth process," or a corresponding expression, refers to a concept that includes cases where the entity executing the n processes from the first process to the nth process, i.e., the processor, changes along the way. In other words, this expression refers to a concept that includes both cases where all n processes are executed by the same processor and cases where the processor changes at an arbitrary rate for the n processes. [Explanation of Symbols]

[0122] 10...Conversation support system, 11...Acquisition unit, 12...Determination unit, 13...Generation unit, 14...Output unit, 15...Language model (Generative AI), 21...Item database, 22...Response database, 23...Contact database, 24...Cycle database, 30...User terminal, 110...Conversation support program.

Claims

1. A conversation support system that outputs response information indicating an answer to a user's request, Equipped with at least one processor, The at least one processor, Multiple combinations of request information indicating user requests and item information containing one or more items necessary to generate response information indicating responses to those requests are pre-stored in the item database. Obtain target request information that indicates the target request, which is the request that will be the subject of response generation. By referring to the aforementioned item database, obtain the target item information, which is the item information corresponding to the target request information. Determine whether all of the one or more items included in the target item information are included in the target request information. If it is determined that all of the one or more items included in the target item information are included in the target request information, the target request information is input to the generating AI as a first prompt, and a first generation process is executed to generate target response information indicating a response to the target request. If it is determined that all of the one or more items included in the target item information are not included in the target request information, a second generation process is executed to generate the target response information based on the target request information and the items missing from the target request information. Output the generated target response information, The second generation process is, The steps include outputting question information indicating questions related to the missing items among the one or more items mentioned above, The steps include obtaining additional information indicating the user's response to the aforementioned question information, A conversation support system comprising the step of inputting the aforementioned target request information and the aforementioned additional information as a second prompt to the generating AI to generate the aforementioned target response information.

2. The at least one processor, Multiple combinations of the aforementioned question information and the expected answer information indicating the expected answer to the question indicated by the aforementioned question information are stored in advance in the answer database. The second generation process is as follows: The steps include: referring to the aforementioned answer database to obtain the assumed answer information corresponding to the question information output immediately before; A step of calculating the similarity between the obtained assumed response information and the additional information, The steps of outputting, acquiring additional information, acquiring expected answer information, and calculating are repeated until all of the missing items are acquired as additional information. In the repeated step described above, if the number of times the calculated similarity score exceeds a threshold exceeds a predetermined number of times, information instructing the generation AI to generate an answer opposite to the additional information is input as a third prompt to generate opposite information that represents the opposite answer, The conversation support system according to claim 1, further comprising the step of outputting the generated opposite information.

3. The at least one processor, Multiple combinations of an identifier that identifies the person in charge and contact information indicating the contact details of that person are pre-stored in the contact database. The second generation process is as follows: A step of identifying the target person identifier, which is the person identifier of the person in charge related to the answer indicated by the counter information, The steps include: obtaining target contact information, which is the contact information corresponding to the target person identifier, by referring to the aforementioned contact database; The conversation support system according to claim 2, further comprising the step of outputting the aforementioned target contact information.

4. The at least one processor, Multiple combinations of a first-stage identifier that specifies a stage in the business cycle, a second-stage identifier that specifies the next stage in the business cycle, and keywords set for the stage specified by the first-stage identifier are stored in advance in the cycle database. The system outputs business question information indicating questions about the status of the tasks performed by the user. Obtain business status information that shows the user's response to the business inquiry information. By referring to the cycle database, the first-stage identifier corresponding to the keyword related to the business status information is identified, The business status information and information requesting the necessary actions to transition the stage indicated by the identified first stage identifier to the next stage indicated by the second stage identifier corresponding to the first stage identifier are input to the generating AI as a fourth prompt to generate transition action information indicating the necessary actions. A conversation support system according to any one of claims 1 to 3, which outputs the generated transition operation information.

5. A conversation support method performed by a conversation support system comprising at least one processor, The steps include pre-storing multiple combinations of request information indicating user requests and item information containing one or more items necessary to generate response information indicating responses to those requests in an item database, A step of obtaining target request information that indicates the target request which is the request to be used for response generation, The steps include: obtaining target item information, which is the item information corresponding to the target request information, by referring to the item database; A step of determining whether all of the one or more items included in the target item information are included in the target request information, If it is determined that all of the one or more items included in the target item information are included in the target request information, the first generation process is executed, which involves inputting the target request information as a first prompt into the generating AI and generating target response information indicating a response to the target request. If it is determined that all of the one or more items included in the target item information are not included in the target request information, the second generation process is performed to generate the target response information based on the target request information and the items missing from the target request information. The steps include outputting the generated target response information, Includes, The second generation process is, The steps include outputting question information indicating questions related to the missing items among the one or more items mentioned above, The steps include obtaining additional information indicating the user's response to the aforementioned question information, A conversation support method comprising the step of inputting the aforementioned target request information and the aforementioned additional information as a second prompt to the generating AI to generate the aforementioned target response information.

6. The steps include pre-storing multiple combinations of request information indicating user requests and item information containing one or more items necessary to generate response information indicating responses to those requests in an item database, A step of obtaining target request information that indicates the target request which is the request to be used for response generation, The steps include: obtaining target item information, which is the item information corresponding to the target request information, by referring to the item database; A step of determining whether all of the one or more items included in the target item information are included in the target request information, If it is determined that all of the one or more items included in the target item information are included in the target request information, the first generation process is executed, which involves inputting the target request information as a first prompt into the generating AI and generating target response information indicating a response to the target request. If it is determined that all of the one or more items included in the target item information are not included in the target request information, the second generation process is performed to generate the target response information based on the target request information and the items missing from the target request information. The steps include outputting the generated target response information, Have the computer run it, The second generation process is, The steps include outputting question information indicating questions related to the missing items among the one or more items mentioned above, The steps include obtaining additional information indicating the user's response to the aforementioned question information, A conversation support program comprising the step of inputting the aforementioned target request information and the aforementioned additional information as a second prompt to the generating AI to generate the aforementioned target response information.

Images