Information processing apparatus, answer generation method, and answer generation program

Abstract

To enhance accuracy of an answer to a question.SOLUTION: An information processing apparatus includes: a query generation section for using a generation model generated by being trained by machine learning so as to be able to generate a search query in accordance with an input question, to generate a search query for retrieving information related to a target question; and an answer generation section for using information detected by search with use of the search query to generate an answer to the target question. This information processing apparatus makes it easy to generate an answer that is optimized for a specific application.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device, an answer generation method, and an answer generation program. [Background technology]

[0002] There is known a technology for automatically generating answers to questions using a large-scale language model or the like. For example, Patent Document 1 below describes a technology for generating a prompt for inputting the question into a large-scale language model by adding reference information to an input question. By adding reference information to the question, it is possible to have the large-scale language model generate an answer that takes the reference information into consideration. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7325152 Summary of the Invention [Problem to be solved by the invention]

[0004] In the sentence generation method described in Patent Document 1, in order to obtain the above-mentioned reference information, a feature vector of the sentence is calculated from the input question sentence. Then, from among multiple sentences recorded in a sentence database, a sentence associated with a feature vector similar to the calculated feature vector is detected, and the detected sentence is used as reference information.

[0005] The above-mentioned sentence generation method has room for improvement in that the accuracy of the generated answer depends on the input question. For example, in the sentence generation method described in Patent Document 1, if the input question clearly and succinctly indicates what the asker wants to ask, an appropriate feature vector is calculated and appropriate reference information is acquired. On the other hand, if the input question is unclear, insufficiently worded, or redundant, the calculated feature vector is likely to not reflect the asker's intention. In this case, reference information unrelated to the asker's intention is acquired, which is likely to reduce the accuracy of the answer.

[0006] The present disclosure has been made in view of the above-mentioned problems, and an exemplary purpose thereof is to provide a technology that enables improving the accuracy of answers to questions. [Means for solving the problem]

[0007] An information processing device according to an exemplary aspect of the present disclosure includes a query generation means that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation means that generates an answer to the target question using information detected by a search using the search query.

[0008] An answer generation method according to an exemplary aspect of the present disclosure includes a query generation step in which at least one processor generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation step in which an answer to the target question is generated using information detected by a search using the search query.

[0009] An answer generation program according to an exemplary aspect of the present disclosure causes a computer to function as a query generation means that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation means that generates an answer to the target question using information detected by a search using the search query. [Effects of the Invention]

[0010] According to an exemplary aspect of the present disclosure, an exemplary effect is achieved in that it is possible to improve the accuracy of answers to questions. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 2] 1 is a flow diagram showing the flow of a response generation method according to the present disclosure. [Figure 3] 1 is a diagram illustrating a configuration of a response system according to the present disclosure. [Figure 4] 4 is a block diagram showing a configuration of an information processing device included in the response system shown in FIG. 3. FIG. [Figure 5] FIG. 10 is a diagram illustrating an example of a template used to generate a prompt. [Figure 6] 5 is a flowchart showing an example of processing executed by the information processing device shown in FIG. 4. [Figure 7] FIG. 1 is a block diagram illustrating a configuration of a computer that functions as an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technologies (part or all of the products or methods) employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technologies employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, embodiments that do not exhibit the effects mentioned in the exemplary embodiments shown below may also be included in the scope of the present invention.

[0013] First Exemplary Embodiment A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is a basic form for each of the exemplary embodiments described below. Note that the scope of application of each technique employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technique employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technique shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0014] (Configuration of information processing device 1) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in Fig. 1, the information processing device 1 includes a query generation unit 101 and an answer generation unit 102.

[0015] The query generation unit 101 generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question.

[0016] The generative model is a search query corresponding to an input question, and is a trained model generated by machine learning so as to be able to generate a search query for searching for information related to the question. This generative model may be a general-purpose language model (a model generated by machine learning sequences of words, sentences, etc., which are components of natural language). Alternatively, this generative model may be a model specialized for generating search queries (a model generated by machine learning correspondences between questions and corresponding search queries). Alternatively, this generative model may be a model generated by fine-tuning a general-purpose language model using training data that associates questions with corresponding search queries.

[0017] The answer generation unit 102 generates an answer to the target question using information detected by a search using the search query generated by the query generation unit 101. Note that "a search using the search query generated by the query generation unit 101" includes not only a search performed using the search query as is, but also a search performed using information generated using the search query (for example, a feature vector representing the features of the search query). The search may be performed by the information processing device 1, or may be performed by another device. In the latter case, the query generation unit 101 transmits the generated search query to the other device to perform the search, and the answer generation unit 102 acquires the search results from the other device and generates an answer.

[0018] Furthermore, the method by which the answer generation unit 102 generates an answer is not particularly limited. For example, the answer generation unit 102 may input information detected by a search using the search query generated by the query generation unit 101 and the target question into a language model generated by machine learning of natural language, to generate an answer. Furthermore, if the generation model used to generate the search query is a language model, the answer generation unit 102 may cause the generation model to generate an answer. Furthermore, for example, multiple templates according to the content of the target question may be prepared in advance. In this case, the answer generation unit 102 can generate an answer by inputting information detected by the search into a template according to the content of the target question.

[0019] As described above, the information processing device 1 according to this exemplary embodiment is configured to include a query generation unit 101 that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation unit 102 that generates an answer to the target question using information detected by a search using the search query generated by the query generation unit 101.

[0020] According to the above configuration, rather than using information obtained by directly using the target question in a search, an answer is generated using information detected by a search using a search query generated by a generative model. As a result, even if the target question is unclear, poorly worded, or redundant, for example, it is possible to generate an answer that is in line with the asker's intention by detecting information that is in line with the asker's intention using a search query that is in line with the asker's intention. Therefore, the information processing device 1 has the effect of being able to improve the accuracy of answers to questions.

[0021] Furthermore, the information processing device 1 can easily generate answers optimized for a specific application. When the information processing device 1 is made to generate answers optimized for a specific application, a database in which information corresponding to that application is recorded can be searched. For example, by searching a database in which a product or service user manual is recorded, an answer in accordance with the user manual can be generated. Also, for example, a database in which information on how to deal with injuries or illnesses can be searched. This makes it possible to generate an accurate answer to a question about how to deal with injuries or illnesses. In this way, the information processing device 1 can also be used for healthcare applications.

[0022] (Answer generation program) The functions of the information processing device 1 described above can also be realized by a program. The answer generation program according to this exemplary embodiment causes a computer to function as: a query generation unit that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation unit that generates an answer to the target question using information detected by a search using the search query generated by the query generation unit. This answer generation program makes it possible to improve the accuracy of answers to questions.

[0023] (Answer generation method flow) The flow of the answer generation method according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the answer generation method. Note that the execution entity of each step in this answer generation method may be a processor provided in the information processing device 1, or a processor provided in another device, or the execution entities of each step may be processors provided in different devices.

[0024] In S1 (query generation process), a search query for searching for information related to the target question for which an answer is to be generated is generated using a generative model generated by machine learning so that a search query corresponding to the input question can be generated.

[0025] In S2 (answer generation process), at least one processor generates an answer to the target question using information found by a search using the search query generated in S1.

[0026] As described above, the answer generation method according to this exemplary embodiment includes a query generation process in which at least one processor generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation process in which an answer to the target question is generated using information detected by a search using the search query generated in the query generation process. Therefore, the answer generation method according to this embodiment makes it possible to improve the accuracy of answers to questions.

[0027] Second Exemplary Embodiment A second exemplary embodiment, which is one example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technology shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs.

[0028] (Configuration of response system 100A) The configuration of a response system 100A according to this exemplary embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing the configuration of the response system 100A. The response system 100A is a system having a function of automatically responding to questions received by telephone from its users. As shown in the figure, the response system 100A includes an information processing device 1A, a generation model 2A, a vector generation model 3A, a DB (Data Base) 4A, and a speech recognition device 5A.

[0029] The information processing device 1A is a device with a function for generating an answer to a question. As will be described in detail below, the information processing device 1A uses a generative model 2A to generate a search query for searching for information related to a target question for which an answer is to be generated. The information processing device 1A then searches DB4A using the generated search query and generates an answer to the target question using data detected by the search.

[0030] Like the generative model described in the exemplary embodiment 1, the generative model 2A is a search query corresponding to an input question, and is a trained model generated by machine learning so as to be able to generate a search query for searching for information related to the question. In this exemplary embodiment, an example will be described in which the generative model 2A is a general-purpose language model, that is, a model generated by machine learning of sequences of words, sentences, and the like, which are components of natural language. The generative model 2A may be stored in the information processing device 1A or in another device. In the latter case, the information processing device 1A instructs a device storing the generative model 2A to generate a search query and acquires the generated search query from the device. The same applies to the vector generation model 3A; it may be stored in the information processing device 1A or in another device.

[0031] The vector generation model 3A is a trained model generated by machine learning so that it can generate a feature vector representing the features of input data. The vector generation model 3A is used to generate the feature vector of the above-mentioned search query. Therefore, when a search query written in natural language is used, a model generated by machine learning so that it can generate a feature vector representing the features of the input natural language sentence is used as the vector generation model 3A. The feature vector of the search query generated by the vector generation model 3A is used for the search.

[0032] DB4A is a database that is searched using the above-mentioned search query. As mentioned above, the feature vector of the search query is used for the search, so each piece of information recorded in DB4A is associated with its feature vector in advance.

[0033] As described in the first exemplary embodiment, the content of the generated answer depends on the type of information stored in the database to be searched. As shown in the example of FIG. 3, when the information processing device 1A is to generate an answer to a question about a product handled by the company, DB4A storing information about the product handled by the company can be used. In this case, for example, a document such as a product manual containing the above information may be divided into multiple chunks, and a feature vector may be associated with each chunk. This allows a chunk (a part of the document) associated with a feature vector similar to the feature vector of the search query to be detected as information related to the input question. Note that the method for dividing a document into chunks is arbitrary. For example, the target document may be mechanically divided into chunks every predetermined number of characters, or may be divided into chunks by units such as chapters, sections, paragraphs, or sentences contained in the target document. DB4A may also store a collection of questions and answers about the target product. In this case, each question and answer may be associated with its feature vector.

[0034] The voice recognition device 5A is a device that converts voice data into text data. In the response system 100A, the information processing device 1A acquires voice data, which is a digitalized version of a voice uttered by a user, and transmits the voice data to the voice recognition device 5A. The voice recognition device 5A then converts the received voice data into text data and returns it to the information processing device 1A. Note that the information processing device 1A may be provided with a voice recognition function, in which case the voice recognition device 5A is omitted.

[0035] In the example of Fig. 3, the response system 100A extends the functionality of an existing automated voice response system installed at Company A. In this way, the response system 100A can be easily incorporated into the existing automated voice response system. Furthermore, by incorporating the response system 100A into the existing automated voice response system, it becomes possible to achieve highly accurate automated responses.

[0036] 3 shows an example in which a user U of the response system 100A calls company A to inquire about the price of a new product. In this example, the voice data of the question by user U (specifically, an inquiry about the price of the new product) is transmitted to the information processing device 1A via company A's telephone exchange.

[0037] The information processing device 1A first converts this voice data into text data using a voice recognition device 5A. Next, the information processing device 1A generates a search query for searching for information related to the user U's question using a generation model 2A. Next, the information processing device 1A causes a vector generation model 3A to generate a feature vector of the generated search query. Next, the information processing device 1A searches DB4A using the generated feature vector to detect information related to the user U's question.

[0038] Then, the information processing device 1A generates an answer to the question of the user U using the information detected by the search of the DB4A. Specifically, the information processing device 1A inputs the question of the user U and the information obtained by the search into the generative model 2A to generate an answer. The generated answer is then converted into voice by an automated voice response system and output to the user U via a telephone exchange. Note that the conversion into voice may also be performed by the information processing device 1A.

[0039] In the example of FIG. 3, the answer "The price of product XXX is YYYY" is output to user U. In this way, response system 100A can present a product price that is in line with user U's intention, even when user U asks a vague question that does not specify the product name, such as "What is the price of the new product?". Furthermore, response system 100A can present an answer that is in line with user U's intention, even when user U's question is unclear or long-winded (for example, a question such as "I'm thinking of buying that new thing that came out recently, um, and I was wondering about the price"). In this way, response system 100A can present an answer that is in line with user U's intention, even when user U's question is unclear, unclear, or long-winded.

[0040] Furthermore, although details will be described later, the response system 100A can also switch between responses by the information processing device 1A and responses by the operator Op. This makes it possible to provide an appropriate response even in situations where it is difficult for the information processing device 1A to respond.

[0041] (Configuration of information processing device 1A) The configuration of an information processing device 1A according to this exemplary embodiment will be described with reference to FIG. 4. FIG. 4 is a block diagram showing the configuration of the information processing device 1A. As shown in FIG. 4, the information processing device 1A includes a control unit 10A that controls the various units of the information processing device 1A and a storage unit 11A that stores various data used by the information processing device 1A. The information processing device 1A also includes a communication unit 12A that enables the information processing device 1A to communicate with other devices, an input unit 13A that accepts input to the information processing device 1A, and an output unit 14A that enables the information processing device 1A to output data. The control unit 10A includes a query generation unit 101A, an answer generation unit 102A, a reception unit 103A, a question generation unit 104A, a search unit 105A, a presentation unit 106A, and a switching unit 107A. The switching unit 107A will be described later in the section "Switching to an Operator's Response."

[0042] Similar to the query generation unit 101 in exemplary embodiment 1, the query generation unit 101A generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model 2A generated by machine learning so as to be able to generate a search query according to an input question. More specifically, the query generation unit 101A generates a prompt that instructs the generation of a search query using the target question, and generates the search query by inputting the generated prompt to the generative model 2A. Details of the prompt generated by the query generation unit 101A will be described later with reference to FIG. 5.

[0043] Similar to the answer generation unit 102 in the first exemplary embodiment, the answer generation unit 102A generates an answer to the target question using information detected by a search using the search query generated by the query generation unit 101A.

[0044] As described above, in this exemplary embodiment, a language model generated by machine learning of natural language is used as the generation model 2A. In this case, the answer generation unit 102A may generate an answer to the target question using the generation model 2A. This provides the effect of being able to generate both a search query and an answer using a single generation model 2A, in addition to the effect provided by the information processing device 1.

[0045] To generate an answer, the answer generation unit 102A first generates a prompt that instructs generation of an answer to the target question, using the target question and information detected by a search using the search query generated by the query generation unit 101A. Then, the answer generation unit 102A generates an answer by inputting the generated prompt into the generative model 2A. Details of the prompt generated by the answer generation unit 102A will be described later with reference to FIG. 5.

[0046] The receiving unit 103A receives input of various information. For example, the receiving unit 103A receives input of a question from a questioner. Furthermore, for example, the receiving unit 103A receives feedback from the questioner in response to an answer generated by the answer generation unit 102A or an alternative question, which will be described later. The method of feedback is not particularly limited. For example, the receiving unit 103A may receive feedback by voice, text input, or feedback by a predetermined operation input.

[0047] The question generation unit 104A paraphrases the question entered by the questioner to generate an alternative question. The alternative question generated by the question generation unit 104A is presented to the questioner by the presentation unit 106A. The reception unit 103A also receives feedback from the questioner regarding the presented alternative question. If the feedback regarding the alternative question is positive, the answer generation unit 102A generates an answer using information detected by a search using a search query for searching for information related to the alternative question. An alternative question obtained by paraphrasing the question entered by the questioner may more accurately reflect the questioner's intention than the original question. Therefore, in addition to the effects achieved by the information processing device 1, the information processing device 1A can further improve the accuracy of the generated answer. Note that the search query for searching for information related to the alternative question may be generated by the query generation unit 101A. For example, the query generation unit 101A may input the alternative question to the generation model 2A to generate a search query for searching for information related to the alternative question. Furthermore, for example, the query generation unit 101A may input information used to generate an alternative question (for example, a summary, which will be described later, or the question that is the basis for the alternative question) into the generation model 2A, and generate a search query for searching for information related to the alternative question.

[0048] Here, paraphrasing means rephrasing. Although the paraphrasing method is not particularly limited, it is preferable to apply a method that generates an alternative question that reflects the questioner's intention as accurately as possible. For example, the question generation unit 104A may generate a summary of the question entered by the questioner and generate an alternative question using the generated summary. The generation of the summary may be performed by the generative model 2A or another language model. Furthermore, the question generation unit 104A may generate an alternative question using, in addition to the generated summary, search results obtained by searching DB4A for the summary.

[0049] Furthermore, for example, the question generation unit 104A may generate an alternative question by generating a prompt that includes the question input by the questioner and instructs the generation of a question that is rephrased in an easier-to-understand manner, and inputting the prompt into a language model such as the generation model 2A. Furthermore, the question generation unit 104A may include, in the prompt, relevant information related to the questioner or the person to whom the answer to the question is to be presented (e.g., age, gender, occupation, place of residence, place of origin, language used, knowledge level, previously input questions, answers generated for the question, etc.). This makes it possible to generate more accurate alternative questions.

[0050] The search unit 105A detects information related to the target question by searching using the search query generated by the query generation unit 101A. More specifically, the search unit 105A inputs the search query generated by the query generation unit 101A into the vector generation model 3A to generate a feature vector for the search query. The search unit 105A then searches the DB4A using the generated feature vector. As described above, each piece of information recorded in the DB4A is pre-associated with its feature vector. Therefore, the search unit 105A can detect information associated with a feature vector similar to the generated feature vector as information related to the target question. Note that the similarity between feature vectors can be calculated using, for example, a well-known technique. For example, the search unit 105A may calculate the cosine similarity between the feature vectors. Alternatively, the search unit 105A may detect all information associated with a feature vector whose similarity to the generated feature vector is equal to or greater than a threshold, or may detect each piece of information associated with a predetermined number of feature vectors with the highest similarity to the generated feature vector.

[0051] The presentation unit 106A presents various types of information. For example, the presentation unit 106A presents the answer generated by the answer generation unit 102A to the person to whom the answer is to be presented. As described above, the presentation unit 106A also presents an alternative question to the questioner. In this exemplary embodiment, the questioner and the person to whom the answer is to be presented are the same (e.g., user U in the example of FIG. 3 ), and the presentation format is audio output. Therefore, the presentation unit 106A presents the answer to the questioner in the form of audio output. However, the questioner and the person to whom the answer is to be presented may be different, and information may be presented in other formats. For example, assume that the questioner and the person to whom the answer is to be presented each possess a terminal device equipped with an information display function. In this case, the presentation unit 106A may present information to either or both of the questioner and the person to whom the answer is to be presented by displaying at least one of text and images on the terminal device.

[0052] As described above, information processing device 1A includes query generation unit 101A that generates a search query for searching for information related to a target question for which an answer is to be generated, using generative model 2A generated by machine learning so as to be able to generate a search query according to an input question, and answer generation unit 102A that generates an answer to the target question using information detected by a search using the search query generated by query generation unit 101A. Therefore, information processing device 1A, like information processing device 1, has the effect of being able to improve the accuracy of answers to questions.

[0053] Furthermore, information processing device 1A includes reception unit 103A that receives input of a question, and presentation unit 106A that presents the answer generated by answer generation unit 102A to a person to whom the answer is to be presented, and query generation unit 101A generates a search query using the question received by reception unit 103A as a target question. Therefore, in addition to the effects of information processing device 1, information processing device 1A has the effect of being able to realize processes from receiving input of a question to presenting an answer by a single information processing device 1A.

[0054] (About generating prompts) Fig. 5 is a diagram showing an example of a template used to generate a prompt. Template t1 shown in Fig. 5 is a template used to generate a prompt for generating a search query.

[0055] Template t1 includes a sentence that instructs the generation of a search query: "Please create a search query to search target documents based on the customer's inquiry." Note that the "target document" in this sentence refers to a document stored in DB4A. Template t1 also includes an input field, "Inquiry: {question}," that corresponds to the "customer's inquiry" in this sentence. By inputting a target question into this {question} portion, query generation unit 101A can generate a prompt that instructs the generation of a search query based on the target question.

[0056] Template t1 also includes the sentence "However, when creating a search query, please refer to the example below," which instructs users to refer to the examples described in template t1 when generating a search query. Template t1 also includes two examples, Example 1 and Example 2. Each of these examples corresponds to an inquiry statement (i.e., a question statement) and its corresponding search query. Including specific examples of input and output in multiple prompts and generating output by referring to these specific examples is a technique called few-shot learning. In this technique, including only one example in a prompt is called one-shot learning. While it is not essential to apply few-shot learning or one-shot learning, these methods are effective for generating search queries that match the user's intentions.

[0057] 5 is a template used to generate a prompt for generating an answer to the target question or an alternative question. Template t2 includes a sentence that instructs the generation of an answer: "Please answer the customer's inquiry based on the search results from the target document." Template t2 also includes an input field "Inquiry: {question}" corresponding to the "customer's inquiry" in this sentence, and an input field "Search results: {search results}" corresponding to the "search results." The answer generation unit 102A can generate a prompt that instructs the generation of an answer to the target question or an alternative question by inputting the target question or an alternative question in the {question} part and the search results by the search unit 105A in the {search results} part.

[0058] Furthermore, template t2 includes the sentence, "You are the person in charge of product XXX." This indicates the constraint that an answer should be generated as the person in charge of product XXX. In this way, the answer generation unit 102A may use a prompt that includes constraints, conditions, etc. when generating an answer. This allows an answer that satisfies the constraints and conditions to be generated. Note that the same applies to prompts that generate search queries, in that they may include a deal or conditions. Furthermore, few-shot learning, one-shot learning, etc. may also be applied to prompts that generate answers.

[0059] (Regarding the use of related information) The optimal answer may vary depending on the content of the target question itself, the person asking the target question, and the person to whom the answer to the target question is to be provided. For example, if the person to whom the answer is to be provided has specialized knowledge, it may be preferable to provide a detailed answer using technical terms. On the other hand, if the person to whom the answer is to be provided does not have specialized knowledge, it may be preferable to provide a simple answer without using technical terms.

[0060] Therefore, the query generation unit 101A may input related information related to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented to the target question into the generation model 2A, and generate a search query according to the related information. This provides the effect of further improving the accuracy of the generated answers in addition to the effect provided by the information processing device 1.

[0061] Examples of related information related to the target question include the language used and expressive features (dialect, endings, wording, speaking habits, etc.). Such related information can be obtained by analyzing the target question using a language model such as generative model 2A. Information obtained by analyzing speech data (for example, voice pitch, tone, tempo, etc.) can also be used as related information.

[0062] Furthermore, examples of related information related to the asker of the target question include the asker's age, gender, occupation, place of residence, place of origin, language used, knowledge level, previously entered questions, and answers generated for the questions. Information similar to these can also be used as related information related to a person to whom an answer to the target question will be presented. Such information may be entered by the asker when asking the question, may be registered in advance in the response system 100A, or may be generated by analyzing the behavior of the asker or the person to whom the answer will be presented.

[0063] Alternatively, multiple databases may be prepared for searching. In this case, each database stores data corresponding to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented. The answer generation unit 102A then generates an answer using information detected by searching a database corresponding to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented, among the multiple databases. This configuration also has the effect of further improving the accuracy of the generated answer.

[0064] For example, a first database may store document chunks generated from a product manual that describes a product in detail using technical terms for questioners or presentation targets with specialized knowledge, while a second database may store document chunks generated from a product manual that describes a product in simple language.

[0065] In this case, the search unit 105A may determine whether to search the first database or the second database depending on at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented. For example, the search unit 105A may search the first database if the target question contains technical terms, and may search the second database if the target question does not contain technical terms. Furthermore, for example, the search unit 105A may search the first database if the asker of the question or the person to whom the answer is to be presented has a high level of knowledge about the product, and may search the second database if the level is low.

[0066] Furthermore, a plurality of generative models 2A for generating search queries may be prepared according to at least one of the target question, the asker of the target question, and the recipient of the answer to the target question. In this case, the query generation unit 101A generates a search query using one of the plurality of generative models 2A that corresponds to at least one of the target question, the asker of the target question, and the recipient of the answer to the target question. This configuration also has the effect of further improving the accuracy of the generated answers.

[0067] For example, a first generative model for generating a search query from a target question in standard Japanese and a second generative model for generating a search query from a target question in a specific dialect may be prepared in advance. The second generative model can be generated by machine learning using training data in which a question in a dialect is associated with a corresponding search query. In this case, the query generation unit 101A may determine whether the target question is written in standard Japanese or in a specific dialect. If the query generation unit 101A determines that the target question is written in standard Japanese, it may generate a search query using the second generative model. If the query generation unit 101A determines that the target question is written in a specific dialect, it may generate a search query using the second generative model.

[0068] (Process flow: Accepting questions ~ Presenting answers) The flow of processing executed by the information processing device 1A will be described with reference to Fig. 6. Fig. 6 is a flow diagram showing an example of processing executed by the information processing device 1A. The flow diagram of Fig. 6 includes each process of the answer generation method according to this exemplary embodiment. In the following, an example will be described in which a question is received by telephone from user U shown in Fig. 3.

[0069] In S11, the reception unit 103A receives input of a question from the questioner, user U. Voice data of the question uttered by user U is transmitted to information processing device 1A via a telephone exchange, and reception unit 103A acquires this voice data, transmits the acquired voice data to voice recognition device 5A to convert it into text data, and acquires the question converted into text data.

[0070] It is also assumed that the user U may call to consult with an operator Op. For this reason, when a call is received from the user U, an automated voice response system may be configured to ask the user U whether he / she wishes to be assisted by an operator or not. If the user U inputs a response indicating that he / she does not want to be assisted by an operator, the user's question may be transferred to the information processing device 1A.

[0071] Furthermore, before the user U inputs a question, the content of the question may be narrowed down. For example, an automated voice response system may output a voice prompting the user to select from multiple options what the question is about, and the user may select an option by operating a push button on a telephone. Also, for example, the number of the store to which the inquiry is directed may be input by operating a push button on a telephone. This makes it possible to narrow down the questions the user U wants to ask, further improving the accuracy of the answer.

[0072] It is also conceivable that the user U may make an utterance other than a question (for example, a greeting). For this reason, the answer generation unit 102A may determine whether the input of the user U is a question, and if it is determined that it is a question, the processing from S12 onwards may be performed. In this case, if the answer generation unit 102A determines that it is not a question, it may input the text of the user's utterance content into the generation model 2A to generate an answer (for example, a greeting if it is a response to a greeting), and instruct the presentation unit 106A to present the answer to the user U.

[0073] 6 is a process for generating an alternative question. When it is not necessary to generate an alternative question, after S11, query generation unit 101A generates a search query using the question accepted in S11 as the target question (query generation process). Next, search unit 105A generates a feature vector by inputting the search query into vector generation model 3A, searches DB4A using the generated feature vector, and obtains search results. Then, answer generation unit 102A generates an answer to the target question using information detected by the search (answer generation process).

[0074] In S12, the question generation unit 104A generates a summary of the question received as input in S11. For example, the question generation unit 104A may generate a prompt that includes the question received as input in S11 and instructs the generation of a summary of the question, and input the generated prompt into a language model such as the generation model 2A to generate the summary.

[0075] In S13 (query generation process), the query generation unit 101A uses the generative model 2A to generate a search query for searching for information related to the summary generated in S12. In this way, by generating a search query that searches for information related to the summary rather than the input question itself, it is also possible to find information related to the input question. The prompt used in S13 may be the same as that used when generating a search query for searching for information related to the target question. For example, the query generation unit 101A may generate a search query using a prompt generated by inputting the summary generated in S12 into the {question} part of the template t1 shown in FIG. 5.

[0076] In S14, the search unit 105A generates a feature vector by inputting the search query generated in S13 into the vector generation model 3A. Next, in S15, the search unit 105A searches DB4A using the feature vector generated in S14 and obtains search results. More specifically, the search unit 105A obtains, as search results, data stored in DB4A that is associated with a feature vector that has a high similarity to the feature vector used in the search.

[0077] In S16, the question generation unit 104A generates an alternative question using the summary obtained in S12. The question generation unit 104A may also use the search results obtained in S15 in addition to the summary. For example, the question generation unit 104A may generate a prompt that includes the summary obtained in S12 and the search results obtained in S15 and instructs the user to generate a question based on the contents of these, and input the generated prompt into a language model such as the generation model 2A to generate an alternative question.

[0078] In S17, the presenting unit 106A presents the alternative question generated in S16 to the user U. For example, the presenting unit 106A may generate text by inputting an alternative question into the {question} part of a template such as "Your question is {question}, right?", and then use an automated voice response system to convert the generated text into voice and present it to the user U.

[0079] In S18, the reception unit 103A determines whether or not there was positive feedback (FB) for the alternative question presented in S17. If the determination in S18 is YES, the process proceeds to S19, and if the determination in S18 is NO, the process returns to S11. In S11, to which the transition is made from S18, the reception unit 103A receives input of a new question, for example, by having the automated voice response system output a voice message prompting the user to re-enter the question (for example, a message such as "Excuse me, could you please ask your question again").

[0080] In S18, the feedback may be input by voice, or by operating the push buttons on a telephone, etc. In the former case, the receiving unit 103A may convert the input voice into text using the voice recognition device 5A, and input the text into a language model such as the generative model 2A, and determine whether the content is positive or negative.

[0081] In S19, the search unit 105A searches DB4A using the feature vector generated in S14 and obtains the search results. In S19, similar to S15, the search unit 105A obtains, as the search results, data stored in DB4A that is associated with a feature vector that has a high similarity to the feature vector used in the search.

[0082] In S20 (answer generation process), the answer generation unit 102A generates an answer to the target question using information detected by a search using the search query generated in S13, i.e., the search results obtained in S19. More specifically, the answer generation unit 102A generates a prompt that instructs the generation of an answer to the target question using the search result information of S19 and the target question, and generates an answer by inputting the generated prompt into the generative model 2A. For example, the answer generation unit 102A may generate an answer using a prompt generated by inputting the target question in the {question} part of the template t2 shown in FIG. 5 and the search results in the {search results} part.

[0083] In S21, the presentation unit 106A presents the answer generated in S20 to the user U. More specifically, the presentation unit 106A converts the answer generated in S20 into voice using an automated voice response system and presents the voice to the user U. This ends the processing in FIG. 6.

[0084] After S21, the reception unit 103A may receive feedback on the presented answer, for example, by having the automated voice response system output a voice prompting the user U to input whether or not he or she is satisfied with the presented answer. The reception unit 103A may also receive an additional question, for example, by having the automated voice response system output a voice prompting the user U to input any additional questions. If an additional question is received, the process returns to S12.

[0085] If the follow-up question is related to the previous question, it is preferable to generate an answer that takes into account the previous question and its answer, i.e., the response history, when inputting the answer to the follow-up question. This makes it possible to generate a highly accurate answer that takes into account the history of the dialogue. On the other hand, if the follow-up question is not related to the previous question, it is preferable to generate an answer without taking into account the response history regarding the previous question.

[0086] The service of presenting an answer by the information processing device 1A may be provided only to predetermined targets (for example, subscribers of the service or customers of the subscribers). In this case, user authentication may be performed before the process of S11 is performed, and the processes from S11 onward may be performed on the condition that the user authentication is successful. The method of user authentication is not particularly limited, but for example, the reception unit 103A may determine whether a predetermined passphrase has been entered, and perform the process of S11 if it is determined that the passphrase has been entered. The passphrase can be entered by voice, which is preferable in that authentication can be performed smoothly without requiring character input or the like.

[0087] (Switching to operator assistance) The switching unit 107A switches the response to the inquirer to the response by an operator. More specifically, when a predetermined switching condition is satisfied, the switching unit 107A performs the above-mentioned switching by controlling the telephone exchange device shown in Fig. 3 to put the inquirer and the operator Op into a communication state.

[0088] The switching condition may be set as appropriate. For example, the switching condition may be that a search using a search query has failed a predetermined number of times in a row. In this case, when the search by the search unit 105A has failed a predetermined number of times in a row, the switching unit 107A switches the response to the questioner to a response by an operator. This not only provides the effect of the information processing device 1, but also provides the effect of enabling an appropriate response by an operator in a situation where it is difficult to generate an answer based on the search results.

[0089] The conditions for success and failure of a search may be determined in advance. For example, the search may be determined to have failed if DB4A does not store data associated with a feature vector whose similarity to the feature vector of the search query is equal to or greater than a predetermined threshold.

[0090] 6. Also, for example, when the feedback from the presentation target person regarding the answer presented in S23 of Fig. 6 is negative, the switching unit 107A may switch the response to the presentation target person to a response by an operator. This makes it possible to avoid ending the call with the presentation target person without being able to present a satisfactory answer. Also, it is preferable that the switching unit 107A switch to a response by an operator when the questioner makes an utterance or operation input requesting switching to a response by an operator.

[0091] Here, when the response by the information processing device 1A is switched to the response by the operator, the presentation unit 106A may present to the operator each question that was input to the information processing device 1A before the switch. This allows the operator to understand what the questioner wants to ask and then respond to the questioner.

[0092] Furthermore, the presentation unit 106A may present a summary of each question input before switching. This allows the operator to quickly understand what the questioner wants to know and respond smoothly. The generation of the summary may be performed by the question generation unit 104A, as in S12 of FIG. 6.

[0093] [Modification] In the exemplary embodiment 2, an example has been described in which the response system 100A responds by voice to a question input by voice, but the response system 100A can also respond by text to a question input by text. In this case, the content of the processing by the information processing device 1A is the same as in the exemplary embodiment 2, except that the voice recognition device 5A and the automatic voice response system are no longer necessary.

[0094] [Software implementation example] Some or all of the functions of the information processing device 1, 1A may be realized by hardware such as an integrated circuit (IC chip), or may be realized by software.

[0095] In the latter case, the information processing devices 1 and 1A are realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Fig. 7. Fig. 7 is a block diagram showing the hardware configuration of computer C that functions as information processing device 1 or 1A.

[0096] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program (answer generation program) P for causing the computer C to operate as the information processing device 1 or 1A. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1 or 1A.

[0097] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0098] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, mouse, display, and printer.

[0099] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0100] Furthermore, each of the above functions of the information processing device 1, 1A may be realized by a single processor provided in a single computer, by multiple processors provided in a single computer working together, or by multiple processors provided in each of multiple computers working together. Furthermore, a program for causing the information processing device 1 or 1A to realize each of the above functions may be stored in a single memory provided in a single computer, or may be distributed and stored in multiple memories provided in a single computer, or may be distributed and stored in multiple memories provided in each of multiple computers.

[0101] [Additional Notes] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0102] (Appendix A1) An information processing device comprising: a query generation means that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation means that generates an answer to the target question using information detected by a search using the search query.

[0103] (Appendix A2) An information processing device as described in Appendix A1, comprising: a receiving means for receiving input of a question; and a presentation means for presenting the answer generated by the answer generation means to a person to whom the answer is to be presented, wherein the query generation means generates a search query using the question received by the receiving means as the target question.

[0104] (Appendix A3) The information processing device described in Appendix A1 or A2, wherein the generative model is a language model generated by machine learning natural language, and the answer generation means generates an answer to the target question using the generative model.

[0105] (Appendix A4) The information processing device according to any one of Appendices A1 to A3, wherein related information relating to at least one of the target question, the asker of the target question, and the person to whom an answer to the target question is to be presented is input into the generative model, and a search query is generated according to the related information.

[0106] (Appendix A5) An information processing device described in any of Appendices A1 to A4, wherein the answer generation means generates the answer using information detected by searching a database among a plurality of databases corresponding to at least one of the target question, the questioner of the target question, and the person to whom the answer to the target question is to be presented.

[0107] (Appendix A6) The information processing device described in any of Appendices A1 to A5, wherein the query generation means generates the search query using a generation model among the plurality of generation models that corresponds to at least one of the target question, the asker of the target question, and the target person to whom an answer to the target question is to be presented.

[0108] (Appendix A7) An information processing device as described in any of Appendices A1 to A6, comprising: a question generation means that generates an alternative question by paraphrasing a question input by a person asking the target question; a presentation means that presents the alternative question to the person asking the question; and a reception means that receives feedback from the person asking the question regarding the presented alternative question, wherein the answer generation means generates an answer using information detected by a search using a search query to search for information related to the alternative question if the feedback regarding the alternative question is positive.

[0109] (Appendix A8) An information processing device according to any one of appendices A1 to A7, further comprising: a switching means for switching the response to the questioner of the target question to a response by an operator when a search using the search query fails a predetermined number of times in a row.

[0110] (Appendix B1) An answer generation method including: a query generation process in which at least one processor generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation process in which an answer to the target question is generated using information detected by a search using the search query.

[0111] (Appendix B2) The answer generation method described in Appendix B1 includes a reception process in which the at least one processor receives input of a question, and a presentation process in which the at least one processor presents the answer generated in the answer generation process to a person to whom the answer is to be presented, and in the query generation process, the at least one processor generates a search query using the question received in the reception process as the target question.

[0112] (Appendix B3) The answer generation method described in Appendix B1 or B2, wherein the generative model is a language model generated by machine learning natural language, and in the answer generation process, the at least one processor generates an answer to the target question using the generative model.

[0113] (Appendix B4) The answer generation method described in any of Appendices B1 to B3, wherein in the query generation process, the at least one processor inputs related information related to at least one of the target question, the asker of the target question, and the recipient of the answer to the target question into the generation model, and generates a search query according to the related information.

[0114] (Appendix B5) An answer generation method described in any of Appendices B1 to B4, wherein in the answer generation process, the at least one processor generates the answer using information detected by searching a database among multiple databases that corresponds to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented.

[0115] (Appendix B6) An answer generation method described in any of Appendices B1 to B5, wherein in the query generation process, the at least one processor generates the search query using a generation model from among the plurality of generation models that corresponds to at least one of the target question, the asker of the target question, and the target person to whom the answer to the target question is to be presented.

[0116] (Appendix B7) The answer generation method described in any of Appendices B1 to B6, including: a question generation process in which the at least one processor paraphrases a question entered by a questioner of the target question to generate an alternative question; a presentation process in which the at least one processor presents the alternative question to the questioner; and a reception process in which the at least one processor receives feedback from the questioner on the presented alternative question, wherein if the feedback on the alternative question is positive, the at least one processor generates an answer using information detected by a search using a search query to search for information related to the alternative question in the answer generation process.

[0117] (Appendix B8) An answer generation method described in any of Appendices B1 to B7, including a switching process in which the at least one processor switches the response to the asker of the target question to a response by an operator if a search using the search query fails a predetermined number of times in a row.

[0118] (Appendix C1) An answer generation program that causes a computer to function as: a query generation means that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation means that generates an answer to the target question using information detected by a search using the search query.

[0119] (Appendix C2) The answer generation program described in Appendix C1 causes the computer to function as a receiving means for receiving input of a question and a presentation means for presenting the answer generated by the answer generation means to a person to whom the answer is to be presented, and the query generation means generates a search query using the question received by the receiving means as the target question.

[0120] (Appendix C3) The answer generation program described in Appendix C1 or C2, wherein the generative model is a language model generated by machine learning natural language, and the answer generation means generates an answer to the target question using the generative model.

[0121] (Appendix C4) The answer generation program described in any of Appendices C1 to C3, wherein the query generation means inputs related information related to at least one of the target question, the asker of the target question, and the person to whom an answer to the target question is to be presented into the generation model, and generates a search query according to the related information.

[0122] (Appendix C5) The answer generation program described in any of Appendices C1 to C4, wherein the answer generation means generates the answer using information detected by searching a database among a plurality of databases corresponding to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented.

[0123] (Appendix C6) The answer generation program according to any one of Appendices C1 to C5, wherein the query generation means generates the search query using a generation model among the plurality of generation models that corresponds to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented.

[0124] (Appendix C7) The answer generation program of any of Appendices C1 to C6, which causes the computer to function as question generation means that generates an alternative question by paraphrasing a question entered by a questioner of the target question, presentation means that presents the alternative question to the questioner, and reception means that receives feedback from the questioner on the presented alternative question, wherein the answer generation means generates an answer using information detected by a search using a search query to search for information related to the alternative question if the feedback on the alternative question is positive.

[0125] (Appendix C8) An answer generation program described in any of Appendices C1 to C7, which causes the computer to function as a switching means for switching the response to the asker of the target question to a response by an operator when a search using the search query fails a predetermined number of times in a row.

[0126] (Appendix D1) An information processing device comprising at least one processor that executes a query generation process that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation process that generates an answer to the target question using information detected by a search using the search query.

[0127] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0128] (Appendix D2) The information processing device described in Appendix D1, wherein the at least one processor executes a reception process for receiving input of a question and a presentation process for presenting the answer generated in the answer generation process to a person to whom the answer is to be presented, and in the query generation process, the at least one processor generates a search query using the question received in the reception process as the target question.

[0129] (Appendix D3) The information processing device described in Appendix D1 or D2, wherein the generative model is a language model generated by machine learning natural language, and in the answer generation process, the at least one processor generates an answer to the target question using the generative model.

[0130] (Appendix D4) An information processing device described in any of Appendices D1 to D3, wherein in the query generation process, the at least one processor inputs related information related to at least one of the target question, the questioner of the target question, and the person to whom an answer to the target question is to be presented into the generation model, and generates a search query according to the related information.

[0131] (Appendix D5) An information processing device described in any of Appendices D1 to D4, wherein in the answer generation process, the at least one processor generates the answer using information detected by searching a database among multiple databases corresponding to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented.

[0132] (Appendix D6) An information processing device described in any of Appendices D1 to D5, wherein in the query generation process, the at least one processor generates the search query using a generation model among the multiple generation models that corresponds to at least one of the target question, the asker of the target question, and the target person to whom an answer to the target question is to be presented.

[0133] (Appendix D7) The information processing device of any of Appendices D1 to D6, wherein the at least one processor executes a question generation process that paraphrases a question entered by a questioner of the target question to generate an alternative question, a presentation process that presents the alternative question to the questioner, and a reception process that receives feedback from the questioner on the presented alternative question, and if the feedback on the alternative question is positive, the at least one processor generates an answer in the answer generation process using information detected by a search using a search query to search for information related to the alternative question.

[0134] (Appendix D8) An information processing device described in any of Appendices D1 to D7, wherein the at least one processor executes a switching process to switch the response to the questioner of the target question to a response by an operator when a search using the search query fails a predetermined number of times in a row.

[0135] (Appendix E) A non-transient recording medium having recorded thereon an answer generation program that causes a computer to execute a query generation process that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question, and an answer generation process that generates an answer to the target question using information detected by a search using the search query. [Explanation of symbols]

[0136] 1. Information processing equipment 101 Query generation unit (query generation means) 102 Answer generation unit (answer generation means) 1A Information processing equipment 101A query generation unit (query generation means) 102A Answer generation unit (answer generation means) 103A Reception section (reception means) 104A Question generation unit (question generation means) 106A Presentation unit (presentation means) 107A Switching unit (switching means) 2A Generative Model 4A DB (Database)

Claims

1. a query generation means for generating a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation means for generating an answer to the target question using information detected by a search using the search query.

2. A receiving means for receiving an input of a question; a presentation means for presenting the answer generated by the answer generation means to a presentation target of the answer, The information processing device according to claim 1 , wherein the query generating means generates a search query using the question received by the receiving means as the target question.

3. the generative model is a language model generated by machine learning natural language; The information processing device according to claim 1 , wherein the answer generating means generates an answer to the target question using the generative model.

4. 3. The information processing device according to claim 1, wherein the query generation means inputs related information relating to at least one of the target question, the questioner of the target question, and the person to whom an answer to the target question is to be presented into the generation model, and generates a search query according to the related information.

5. 3. The information processing device according to claim 1, wherein the answer generation means generates the answer using information detected by searching a database corresponding to at least one of the target question, the asker of the target question, and the person to whom the answer to the target question is to be presented, among a plurality of databases.

6. 3. The information processing device according to claim 1, wherein the query generation means generates the search query using a generation model among the plurality of generation models that corresponds to at least one of the target question, the asker of the target question, and the person to whom an answer to the target question is to be presented.

7. a question generation means for generating an alternative question by paraphrasing a question input by a questioner of the target question; a presentation means for presenting the alternative question to the questioner; a receiving means for receiving feedback from the questioner regarding the presented alternative question; 3. The information processing device according to claim 1, wherein, when the feedback for the alternative question is positive, the answer generation means generates an answer using information detected by a search using a search query for searching for information related to the alternative question.

8. The information processing device according to claim 1 , further comprising: a switching means for switching a response to a questioner of the target question to a response by an operator when a search using the search query fails a predetermined number of times in succession.

9. At least one processor a query generation process that generates a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and an answer generation process for generating an answer to the target question using information detected by a search using the search query.

10. Computer, a query generation means for generating a search query for searching for information related to a target question for which an answer is to be generated, using a generative model generated by machine learning so as to be able to generate a search query according to an input question; and and an answer generation program that functions as an answer generation means that generates an answer to the target question using information detected by a search using the search query.

Images