Information processing device, support method, and support program

Abstract

Making it easier to find insurance products that meet your desired conditions. [Solution] The information processing device comprises an acquisition unit that acquires condition information indicating the subject's desired conditions for insurance, and an extraction unit that uses a machine learning extraction model, which takes a set of documents and data as input and outputs the portion of the document related to the data, to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to recommend to the subject. Furthermore, this information processing device can also support the subject's decision-making when selecting an insurance product.

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, a support method, and a support program.

Background Art

[0002] With the diversification of needs for insurance products, a wide variety of insurance products are being offered. As a technology for assisting in finding an insurance product suitable for oneself from among a large number of insurance products, for example, there is an insurance product recommendation system described in Patent Document 1. The insurance product recommendation system described in Patent Document 1 outputs an insurance product suitable for a user by using various attribute information such as the gender, age, address, and preferences of the user for whom the recommendation is intended.

[0003] More specifically, in the insurance product recommendation system described in Patent Document 1, a large number of insurance products classified into main contracts, special contracts, and services in advance are each extracted from a database, and then all patterns of combinations of main contracts, special contracts, and services are generated. And in this insurance product recommendation system, a virtual product suitable for the user is extracted and recommended from among a virtual product group composed of the insurance product group generated as described above, using the user's attribute information and recommendation information.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] When applying the insurance product recommendation system described in Patent Document 1, it is necessary to predetermine which insurance products to recommend to users with what kind of attribute information. However, predetermining appropriate insurance products according to user attributes is not easy. In particular, user attributes, including desired conditions for insurance, have become more diverse recently, making it difficult to cover such diverse user attributes and propose appropriate insurance products.

[0006] Therefore, currently, many people considering purchasing insurance have to go through the cumbersome process of searching for insurance that meets their desired conditions from various insurance product brochures and other materials. The illustrative purpose of this disclosure is to provide a technology that simplifies the process of finding insurance products that meet the desired conditions for the target person. [Means for solving the problem]

[0007] An information processing device relating to an illustrative aspect of this disclosure includes an acquisition means for acquiring condition information indicating the desired conditions of an insurance policyholder, and an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the policyholder, using an extraction model trained on machine learning to take a set of documents and data as input and output the portion of the document related to the data.

[0008] Other information processing devices relating to illustrative aspects of this disclosure include acquisition means for acquiring condition information indicating the desired conditions of an insured person, and determination means for determining whether an insurance product meets the said desired conditions using a document describing the insurance product and a language model trained on natural language.

[0009] In the exemplary aspects of this disclosure, the support method involves at least one processor performing an acquisition process to acquire condition information indicating the subject's desired conditions for insurance, and an extraction process to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the subject, using an extraction model trained on machine learning to take a document and data pair as input and output the portion of the document related to the data.

[0010] The support program relating to an illustrative aspect of this disclosure uses a computer as an acquisition means for acquiring condition information indicating the desired conditions of an insurance policyholder, and as an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the policyholder, using an extraction model that has been trained to take a set of documents and data as input and output the portion of the document related to the data. [Effects of the Invention]

[0011] One illustrative aspect of this disclosure is that it can facilitate the process for respondents to find insurance products that meet their desired conditions. [Brief explanation of the drawing]

[0012] [Figure 1] This is a block diagram showing the configuration of the information processing device related to this disclosure. [Figure 2] This flowchart shows the flow of support methods related to this disclosure. [Figure 3] This is a block diagram showing the configuration of other information processing devices related to this disclosure. [Figure 4] This figure shows an example of extracting related information. [Figure 5] This figure shows an example of determining whether an insurance product corresponding to a related description meets the desired conditions indicated in the conditions information. [Figure 6] This figure shows an example of a UI (User Interface) screen that accepts correction requests. [Figure 7]This figure shows an example of a UI screen that allows users to add or change their desired conditions. [Figure 8] This figure shows an example of a UI screen that accepts input for a description of the history information. [Figure 9] Figure 3 is a flowchart showing the processing flow executed by the information processing device. [Figure 10] This is a block diagram showing the configuration of an information processing device related to a reference example. [Figure 11] This is a block diagram showing the configuration of a computer that functions as an information processing device related to this disclosure. [Modes for carrying out the invention]

[0013] 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 some or all of the technologies (things or methods) employed in each of 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 each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, the effects mentioned in each of the exemplary embodiments shown below are examples of effects that can be expected in that exemplary embodiment and do not define the scope of the present invention. That is, embodiments that do not produce the effects mentioned in each of the exemplary embodiments shown below may also be included in the scope of the present invention.

[0014] [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 later. Note that the scope of application of each technology adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology adopted in this exemplary embodiment can also be adopted in other exemplary embodiments included in the present disclosure as long as there are no particular technical obstacles. Further, each technology shown in the drawings referred to for explaining this exemplary embodiment can also be adopted in other exemplary embodiments included in the present disclosure as long as there are no particular technical obstacles.

[0015] (Configuration of Information Processing Apparatus 1) The configuration of the information processing apparatus 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 apparatus 1. As shown in FIG. 1, the information processing apparatus 1 includes an acquisition unit 101 and an extraction unit 102.

[0016] The acquisition unit 101 acquires condition information indicating the desired conditions of the person insured for insurance. This "condition information" may be anything that indicates at least one desired condition. For example, the desired conditions may indicate the compensation content desired by the person insured or the upper limit of the premium to be paid by the person insured. Further, the above "insurance" may be any insurance, for example, insurance related to the life and health of the insured such as life insurance, pension insurance, medical insurance, etc., or insurance related to the property owned such as automobile insurance, fire insurance, etc. For example, the acquisition unit 101 may acquire text data in which the person insured has listed the desired conditions as the condition information.

[0017] Further, for example, the acquisition unit 101 may acquire historical information regarding the desired conditions of the target person for insurance as condition information. For example, in addition to the medical history of the target person (which may include the current health status), the acquisition unit 101 may acquire historical information indicating the presence or absence of regular hospital visits, the history of prescribed medications, the results of medical check-ups, the results of doctor consultations, the results of medical examinations, and the receipt history of public assistance, etc. Such historical information can be said to indicate the desire of the target person to receive recommendations for insurance that can be subscribed even in the physical and mental state as shown in the information. In addition to this, for example, the acquisition unit 101 may acquire historical information indicating the insurance that has been contracted in the past, or purchase history information of products and services that reflects the hobbies and preferences of the target person, as condition information, and may also acquire attribute information such as the name and date of birth of the target person as condition information.

[0018] Note that the method of acquiring the condition information is arbitrary. Also, the acquisition unit 101 may acquire multiple types of condition information. For example, the acquisition unit 101 may acquire both the text data input by the target person and the historical information recorded in a predetermined database (for example, a database that records various information related to medical insurance, etc., in association with the personal identification information assigned to each citizen) as condition information.

[0019] The extraction unit 102 uses an extraction model that is machine-learned to output the location related to the data in the document with a set of document and data as input, and extracts relevant descriptions related to the desired conditions indicated by the condition information acquired by the acquisition unit 101 from the document explaining the insurance products that are candidates for recommendation to the target person.

[0020] The above "insurance product" may be any insurance product sold by an insurance company or the like. Also, the above insurance product may be a product with only a main contract, or a product consisting of a combination of a main contract and a special contract.

[0021] The "extraction model" described above can be any model that can be used to extract relevant descriptions related to desired conditions from a document describing an insurance product. For example, a model that divides a document describing an insurance product into predetermined units of text (e.g., each sentence or each desired condition), calculates a score for each of the resulting texts indicating the similarity (or relevance) of the content to the desired conditions, and outputs texts whose calculated score is above a predetermined threshold can be used as the extraction model described above. Such a model can be generated, for example, by machine learning using training data that associates the above score in a text-and-data pair with the correct answer data. The data paired with the text is typically text data, but other forms of data such as image data can also be used.

[0022] Furthermore, the "extraction model" described above may be a general-purpose model usable for purposes other than the extraction described in the related section, or it may be a general-purpose model that has been fine-tuned for the extraction described in the related section. Also, the extraction model may be provided by the information processing device 1 or by another device. In the latter case, the extraction unit 102 utilizes the extraction model via the other device that provides the extraction model.

[0023] Furthermore, the "document explaining the insurance product" mentioned above can be any document that describes information about the insurance in question. For example, an application form for the insurance, a Q&A collection, an explanatory document, a webpage introducing the insurance, a review by a policyholder, or the terms and conditions of the insurance can be used as a document explaining the insurance product. Also, a single document may describe multiple insurance products. In that case, the extraction unit 102 can extract relevant descriptions from the sections describing each of the multiple insurance products. Also, a single document may describe only one insurance product. In that case, the extraction unit 102 can extract relevant descriptions from each of the multiple documents.

[0024] As described above, the information processing device 1 according to this exemplary embodiment is configured to include an acquisition unit 101 that acquires condition information indicating the desired conditions of the subject for insurance, and an extraction unit 102 that uses an extraction model trained on machine learning to take a set of documents and data as input and output the part of the document related to the data, to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the subject.

[0025] The relevant information extracted as described above is related to the desired conditions, and therefore can be said to be a highly important part of the document describing the insurance product when searching for an insurance product that meets those conditions. Thus, the information processing device 1 has the effect of making it possible for the subject to find an insurance product that meets their desired conditions. Furthermore, the information processing device 1 can also support the subject's decision-making when selecting an insurance product.

[0026] It is optional how the extracted relevant information is used to facilitate the search for insurance products. For example, the information processing device 1 may present the extracted relevant information to the user as reference information when searching for insurance products. Alternatively, as described in the exemplary embodiment 2 below, the extracted relevant information may be used to identify and present insurance products that meet the user's desired conditions.

[0027] (Support Program) The functions of the information processing device 1 described above can also be implemented by a program. The support program according to this exemplary embodiment is a support program for finding insurance, and the computer functions as an acquisition means for acquiring condition information indicating the subject's desired conditions for insurance, and an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the subject, using an extraction model that has been trained to take a set of documents and data as input and output the portion of the document related to the data. This support program has the effect of making it possible for the subject to find an insurance product that meets their desired conditions.

[0028] (Flow of support methods) The flow of the support method according to this exemplary embodiment will be explained with reference to Figure 2. Figure 2 is a flowchart showing the flow of the support method. Note that the entity executing each step in this support method may be a processor provided in the information processing device 1, a processor provided in another device, or the entity executing each step may be a processor provided in a different device.

[0029] In S1 (acquisition process), at least one processor acquires condition information indicating the desired conditions of the insured person.

[0030] In S2 (extraction process), at least one processor uses an extraction model trained on machine learning to take a document and data pair as input and output the portion of the document that is relevant to the data in question. This model extracts relevant descriptions related to the desired conditions indicated in the condition information obtained in S1 from documents describing candidate insurance products to recommend to the target audience.

[0031] As described above, the support method according to this exemplary embodiment is a support method for finding insurance, and employs a configuration in which at least one processor performs an acquisition process to acquire condition information indicating the subject's desired conditions for insurance, and an extraction process to extract relevant descriptions related to the above desired conditions from a document describing candidate insurance products to be recommended to the subject, using an extraction model that has been trained to take a pair of documents and data as input and output the portion of the document related to the data. This support method has the effect of making it possible for the subject to find an insurance product that meets their desired conditions.

[0032] [Second exemplary embodiment] A second exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same function as those described in the above-described exemplary embodiment are denoted by the same reference numerals, and their descriptions are omitted as appropriate. The scope of application of each technology adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology adopted in this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure, to the extent that no particular technical problems arise. Furthermore, each technology shown in the drawings referenced to describe this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure, to the extent that no particular technical problems arise.

[0033] (Configuration of Information Processing Device 1A) The configuration of the information processing device 1A according to this exemplary embodiment will be described with reference to Figure 3. Figure 3 is a block diagram showing the configuration of the information processing device 1A. The information processing device 1A is a device equipped with functions to support insurance searching. The information processing device 1A may be a local device used by individual users, or it may be a server that provides insurance search support services to multiple users.

[0034] As shown in the figure, the information processing device 1A includes a control unit 10A that controls all parts 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 for the information processing device 1A to communicate with other devices, an input unit 13A that receives input to the information processing device 1A, and an output unit 14A for the information processing device 1A to output data. The control unit 10A includes an acquisition unit 101A, an extraction unit 102A, a determination unit 103A, a presentation control unit 104A, and a reception unit 105A.

[0035] The acquisition unit 101A acquires condition information indicating the insured person's desired conditions for insurance, similar to the acquisition unit 101 in the exemplary embodiment 1. Below, we will mainly describe an example in which the insured person inputs condition information in text format, describing their desired conditions in natural language, into the information processing device 1A, and the acquisition unit 101A acquires the input condition information.

[0036] Similar to the extraction unit 102 in Exemplary Embodiment 1, the extraction unit 102A uses an extraction model trained on machine learning to take a document and data pair as input and output the portion of the document related to the data, to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to recommend to the target person. Hereinafter, the extraction model used by the extraction unit 102A will be referred to as extraction model M1.

[0037] The determination unit 103A determines whether the insurance product described by the relevant description extracted by the extraction unit 102A meets the subject's desired conditions, using a language model trained on natural language. Here, machine learning on natural language means, more specifically, learning the arrangement of its constituent elements (words, etc.) in natural language sentences and the arrangement of sentences in texts. Examples of language models trained on natural language include BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), and ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately). Hereafter, the language model used by the determination unit 103A will be referred to as language model M2.

[0038] In this exemplary embodiment, we describe an example in which the language model M2 accepts input in the form of a text prompt written in natural language and outputs a response in natural language. However, the language model M2 may also be a model that can accept input in the form of data other than text data, such as images.

[0039] The language model M2 may be a general-purpose language model that can be used for purposes other than inferring whether an insurance product meets the desired conditions, or it may be a general-purpose language model that has been fine-tuned for inferring whether an insurance product meets the desired conditions. Furthermore, the language model M2 may be provided by the information processing device 1A or by another device. In the latter case, the determination unit 103A utilizes the language model M2 via another device that is equipped with the language model M2.

[0040] Furthermore, the extraction of relevant descriptions by the extraction unit 102A is not a mandatory prerequisite for the determination by the determination unit 103A. For example, the determination unit 103A may determine whether each sentence in the document describing the insurance product satisfies the desired conditions indicated in the condition information acquired by the acquisition unit 101A. Although this increases the number of determinations made by the determination unit 103A compared to the case where relevant descriptions are extracted, it is still possible to recommend insurance products that meet the desired conditions of the target person even when this type of processing is adopted.

[0041] The presentation control unit 104A presents various information related to supporting the search for insurance. For example, the presentation control unit 104A presents insurance products that the determination unit 103A has determined to meet the desired conditions as recommended insurance products to the target person. Also, for example, the presentation control unit 104A presents the inference results output by the language model M2. As will be described in detail later, the presentation control unit 104A may present the above inference results along with the basis for the inference. The method of presenting information is arbitrary. For example, the presentation control unit 104A may present the information by having the output unit 14A output the information, or by having the communication unit 12A output the information to another device. Furthermore, the information can be presented in any manner, such as display, printing, voice, or a combination thereof.

[0042] The reception unit 105A receives various instructions related to assisting in the search for insurance. For example, the reception unit 105A receives instructions to modify the results of inference by the language model M2. The method of receiving instructions is arbitrary. For example, the reception unit 105A may receive instructions input via the input unit 13A, or it may receive instructions from other devices via the communication unit 12A.

[0043] As described above, the information processing device 1A includes an acquisition unit 101A that acquires condition information indicating the subject's desired conditions for insurance, and an extraction unit 102A that uses an extraction model M1, which has been trained to take a document and data set as input and output the portion of the document related to the data, to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to recommend to the subject. Therefore, the information processing device 1A has the effect of making it possible for the subject to find an insurance product that meets their desired conditions, similar to the information processing device 1.

[0044] Furthermore, as described above, the information processing device 1A includes a determination unit 103A that uses a language model M2 trained on natural language to determine whether the insurance product described by the relevant description extracted by the extraction unit 102A meets the desired conditions, and a presentation control unit 104A that presents the insurance product that the determination unit 103A has determined to meet the desired conditions as a recommended insurance product to the target person. As a result, in addition to the effects of the information processing device 1, the effect of being able to recommend insurance products that are highly likely to meet the desired conditions to the target person can be obtained.

[0045] Furthermore, the determination by the determination unit 103A can be omitted. If the determination by the determination unit 103A is omitted, the presentation control unit 104A only needs to present the relevant descriptions extracted by the extraction unit 102A. In this case, the subject only needs to compare the presented relevant descriptions with their desired conditions and decide whether or not to select the insurance product corresponding to those descriptions. This eliminates the need for the subject to search for relevant descriptions related to their desired conditions within the documents describing the insurance products, thereby simplifying the process of finding an insurance product.

[0046] Furthermore, as described above, the information processing device 1A includes an acquisition unit 101A that acquires condition information indicating the desired conditions of the person concerned regarding insurance, and a determination unit 103A that determines whether or not an insurance product meets the above desired conditions using a document describing the insurance product and a language model M2 that has been trained on natural language. With this information processing device 1A, an objective determination result regarding whether or not an insurance product meets the desired conditions can be obtained, which has the effect of making it possible for the person concerned to find an insurance product that meets their desired conditions.

[0047] (Example of extracting related information) An example of extracting relevant information using the information processing device 1A will be explained with reference to Figure 4. Figure 4 is a diagram showing an example of extracting relevant information. In the example in Figure 4, a person considering purchasing insurance inputs condition information 401 into the information processing device 1A. The input condition information 401 is acquired by the acquisition unit 101A provided in the information processing device 1A.

[0048] Condition information 401 indicates the insured person's desired conditions for insurance, and specifically, it is a text in natural language describing the risks the insured person wants to be covered by insurance. Since the extraction unit 102A extracts relevant descriptions using the extraction model M1, it can handle condition information 401 that the insured person has freely written in natural language. Although the condition information 401 shown in Figure 4 shows only one desired condition, it is also possible to input condition information that lists multiple desired conditions.

[0049] The user may, for example, input the condition information 401 via the input unit 13A, or input the condition information 401 via the communication unit 12A using their own terminal device. The user may also input the condition information 401 as text data or as voice data. In the latter case, the acquisition unit 101A can acquire the condition information 401 in text format by having the input voice data recognized by the information processing device 1A or other voice recognition device.

[0050] In the example shown in Figure 4, the information processing unit 1A (more specifically, the acquisition unit 101A) acquires a document 402 from database D that describes candidate insurance products to recommend to the target person. If the system is configured to acquire document 402 from database D, then documents describing various types of insurance should be recorded in database D. This allows the acquisition unit 101A to acquire document 402 from database D that describes candidate insurance products to recommend to the target person. Although Figure 4 shows an example of acquiring document 402 describing one insurance product, documents describing multiple insurance products may be acquired, or multiple documents describing one or more insurance products may be acquired. Furthermore, the method of acquiring document 402 is arbitrary and is not limited to the example described above. For example, the target person may input document 402 along with condition information 401 into the information processing unit 1A, or document 402 may be stored in the information processing unit 1A in advance.

[0051] Next, the information processing device 1A (more specifically, the extraction unit 102A) inputs the condition information 401 and document 402 acquired by the acquisition unit 101A as described above into the extraction model M1. As a result, the related description 403 is output from the extraction model M1.

[0052] Related description 403 is a description of the portion of the insurance product description in document 402 that relates to the desired conditions that meet the desired conditions shown in condition information 401. Specifically, related description 403 is a description of the skydiving coverage rider, which relates to the desired condition shown in condition information 401, that the insurance is recommended for people whose hobby is skydiving.

[0053] As will be explained in detail below, the determination unit 103A determines whether the insurance product corresponding to the related description 403 extracted in this manner (i.e., the insurance product described by the related description 403) satisfies the desired conditions indicated in the condition information 401. As mentioned above, the presentation control unit 104A may also present the extracted related description 403 to the target person. In this case, the presentation control unit 104A may extract and present the portion of the related description 403 from document 402, or it may present document 402 with the portion of the related description 403 highlighted. The highlighting should be done in a manner that allows the highlighted portion and the unhighlighted portion to be distinguished. For example, the presentation control unit 104A may highlight the portion of the related description 403 in document 402 by changing the background color of the highlighted portion, changing the font of the text in the highlighted portion, etc.

[0054] (Example of determining whether an insurance product meets the desired conditions) Figure 5 shows an example of determining whether an insurance product corresponding to related description 403 meets the desired conditions shown in condition information 401. As described above, this determination is performed by the determination unit 103A. The language model M2 is also used for this determination. Therefore, the determination unit 103A generates a prompt describing the content of the instructions to the language model M2 and inputs it to the language model M2.

[0055] In the example shown in Figure 5, the information processing device 1A (more specifically, the determination unit 103A) inputs prompt 501 to the language model M2. Prompt 501 includes the related description 403 shown in Figure 4 and the condition information 401 shown in Figure 4, and instructs the device to infer the relationship between the desired condition shown in the condition information 401 and the related description 403.

[0056] More specifically, condition information 401 is described in the "Desired Conditions" section of prompt 501. Related description 403 is described in the "Insurance Product Description" section of prompt 501. Prompt 501 also includes the statement, "You are a financial planner and are verifying whether the insurance product meets the client's desired conditions." While including such a statement is not mandatory, it can improve inference accuracy.

[0057] Furthermore, prompt 501 instructs the system to check whether it can be inferred that the insurance product meets the client's desired conditions. The wording of the prompt can be changed as appropriate within the range to obtain the desired inference result. For example, the determination unit 103A may generate prompts with different inference instructions depending on the desired conditions, type of insurance, language model used, etc.

[0058] Furthermore, prompt 501 includes an "answer format" and instructs the user to answer using this format. By specifying the answer format in this way, it becomes possible to obtain inference results in the desired format. This answer format also includes an item called "basis for inference." By including such an item, the language model M2 can output the basis for the inference along with the inference result. Note that the instruction to output the basis for the inference may also be given by including a sentence such as "Please answer with the basis along with the inference result" in the prompt.

[0059] Furthermore, prompt 501 may include sentences specifying output conditions. For example, prompt 501 may include sentences such as, "If multiple desired conditions are listed, it is necessary to infer whether the insurance product satisfies all of those desired conditions," or "The number of elements included in the response format must match the number of desired conditions." This makes it possible to improve the inference accuracy of the language model M2.

[0060] In prompt 501, everything except the content of "desired conditions" and "insurance product description" is standard. For this reason, the parts of prompt 501 other than the content of "desired conditions" and "insurance product description" may be stored as a standard template in the storage unit 11A or the like. This allows the determination unit 103A to generate prompt 501 by inputting the condition information 401 acquired by the acquisition unit 101A and the related description 403 extracted by the extraction unit 102A into the above template.

[0061] The inference result 502 shown in Figure 5 is an example of an inference result obtained by inputting prompt 501 into language model M2. Inference result 502 shows the result of inferring the relationship between the desired conditions and related description 403 in the format of the response shown in prompt 501. Specifically, inference result 502 shows the inference result that the insurance product corresponding to related description 403 "meets the desired conditions". In addition to the inference result, the basis for the inference is also shown in inference result 502.

[0062] As mentioned above, the condition information 401 may indicate multiple desired conditions. In that case, a prompt can be used to instruct the system to perform an inference for each of the multiple desired conditions regarding whether the related description 403 satisfies the desired conditions. The extraction unit 102A may also extract multiple related descriptions. In that case, a prompt can be used to instruct the system to perform an inference for each of the multiple related descriptions regarding whether the related description satisfies the desired conditions.

[0063] Furthermore, the format in which the inference results are output can be specified by a prompt. For example, the determination unit 103A may generate a prompt instructing the user to answer with one of three options: meets the desired conditions, neutral, or does not meet the desired conditions. In this case, the determination unit 103A should determine that the insurance product corresponding to a certain related description meets the desired conditions when the response is that the insurance product corresponding to that related description meets the desired conditions. The processing to be performed when the response is neutral can be predetermined. For example, if the presentation control unit 104A receives a neutral response to a prompt asking whether an insurance product corresponding to a certain related description meets certain desired conditions, it may present the user with combinations of related descriptions and desired conditions and ask them to input whether the insurance product corresponding to that related description meets the desired conditions. Alternatively, for example, the determination unit 103A may generate a prompt instructing the user to output a numerical value (for example, a number between 0 and 1) indicating the likelihood of meeting the desired conditions. In this case, the determination unit 103A only needs to determine that an insurance product corresponding to a given related description satisfies the desired condition if the numerical value output for a combination of a given desired condition and a given related description is greater than or equal to a predetermined threshold.

[0064] Furthermore, Figure 5 shows an example screen 503 for presenting the inference result 502. In example screen 503, the insurance product corresponding to the related description 403, for which an inference result of satisfying the desired conditions was obtained, is shown as a recommended insurance product, and the basis for the inference included in the inference result 502 is also shown. The presentation control unit 104A can generate and present a screen like example screen 503 using the various information shown in the inference result 502. In this way, the presentation control unit 104A may present the insurance product that the determination unit 103A has determined to satisfy the desired conditions as a recommended insurance product to the target person. In addition, the presentation control unit 104A may present the basis for the inference output by the language model M2 to the target person as a basis for judging whether the inference result is appropriate or not.

[0065] Furthermore, if the output unit 14A has a function to display and output an image, the presentation control unit 104A may cause the output unit 14A to display a screen like the example screen 503. Alternatively, the presentation control unit 104A may, via the communication unit 12A, cause an external display device of the information processing device 1A (for example, a display device on the terminal device used by the target person) to display a screen like the example screen 503.

[0066] As described above, the determination unit 103A may generate a prompt 501 that includes the relevant description and the desired conditions, instructing it to infer the relationship between the relevant description and the desired conditions, and then determine whether the insurance product described by the relevant description satisfies the desired conditions based on the inference result 502, which is the output obtained by inputting the generated prompt 501 to the language model M2. This makes it possible to appropriately determine whether the insurance product described by the relevant description satisfies the desired conditions.

[0067] (Regarding the presentation of inference results and re-inference) The reception unit 105A may accept correction instructions for the inference results of the language model M2. This will be explained with reference to Figure 6. Figure 6 is a diagram showing an example of a UI screen for accepting correction instructions. In addition to the example screen 601, which is an example of a UI screen for accepting correction instructions, Figure 6 also shows a prompt 602 that instructs re-inference based on the correction instructions, and a re-inference result 603 obtained by inputting prompt 602 to the language model M2.

[0068] Screen example 601 displays the inference results of the language model M2, along with the desired conditions and insurance product descriptions (related descriptions extracted by the extraction unit 102A) that were the subject of the inference. Screen example 601 also displays text prompting the user to check for errors in the inference results and to input details of any errors to instruct re-inference. Screen example 601 also displays a text box for inputting corrections and a button (software key) to instruct re-inference.

[0069] The presentation control unit 104A can allow the subject to confirm the inference results by presenting a UI screen, such as the example screen 601. The reception unit 105A can also accept corrections to the inference results via a UI screen, such as the example screen 601.

[0070] In the example screen 601, when the content of the correction is entered into the text box and the button to instruct re-inference is operated, the reception unit 105A receives the entered content of the correction as a correction instruction. For example, the reception unit 105A may accept a correction instruction that describes in natural language why the inference result is incorrect. Alternatively, for example, the reception unit 105A may accept a correction instruction that describes in natural language whether the insurance product corresponding to the related description that was the subject of the inference meets the desired conditions.

[0071] Then, the determination unit 103A re-determines whether the insurance product corresponding to the relevant description meets the desired conditions based on the correction instruction received by the reception unit 105A. For example, the determination unit 103A may generate the prompt 602 shown in Figure 6 and input it to the language model M2, and then re-determine whether the insurance product corresponding to the relevant description meets the desired conditions based on the re-inference result 603 output from the language model M2.

[0072] Prompt 602 includes the text of the correction instruction received by the reception unit 105A and instructs the system to infer the relationship between the relevant description (specifically, the description of the insurance product) and the desired conditions based on the correction instruction. Prompt 602 also includes the sentence, "However, do not output any words that are not included in the response format." Thus, even in prompts that instruct re-inference, sentences specifying the output conditions may be included. This prevents unnecessary content (for example, sentences such as "Apologies. I will re-output." or "The content has been updated") from being output from the language model M2, thus preventing any disruption to the recommendation of insurance products, etc. Prompt 602, like prompt 501 shown in Figure 5, can be generated using a predetermined template. In addition, like prompt 501, desired conditions, relevant descriptions, and response formats may also be described in prompt 602.

[0073] In the reinference result 603 shown in Figure 6, the inference result regarding the relationship between the desired conditions and related descriptions (specifically, the description of the insurance product) has changed from that shown on UI screen 601, reflecting the content of the "comment," or correction instruction, shown in prompt 602. Specifically, in the reinference result 603, the related description, "This medical insurance is available to those aged 65 and over. However, those who fall under the following categories are not eligible to join: those who have been hospitalized within the last 5 years due to cerebral hemorrhage, myocardial infarction, or heart failure…" has changed the inference result regarding whether it meets the desired condition, "Please tell me about medical insurance that I can join even if I am aged 65 and over," to "Does not meet the desired condition." Furthermore, the reasoning for this inference has changed to the sentence, "Because subarachnoid hemorrhage is a type of cerebral hemorrhage and falls under hospitalization within the last 5 years." In this way, by performing reinference using prompt 602, which includes the natural language sentence received as a correction instruction, it is possible to output a reinference result 603 that reflects the content of the correction instruction.

[0074] As described above, the determination unit 103A may generate a prompt (for example, prompt 501 in Figure 5) that instructs the system to output the reasoning result of the relationship between the relevant description and the desired conditions, along with the basis for that reasoning. The presentation control unit 104A may then present the reasoning result output by the language model M2 along with the basis for that reasoning. This provides the effect that, in addition to the effects performed by the information processing device 1, the system can consider whether the reasoning result of the language model M2 is appropriate or not, using the basis for that reasoning as a basis for judgment. The presentation control unit 104A may, instead of presenting the reasoning result output by the language model M2, present the determination result of the determination unit 103A, or an insurance product that the determination unit 103A has determined to satisfy the desired conditions (i.e., an insurance product recommended to the target person), along with the basis for that reasoning.

[0075] Furthermore, as described above, the information processing device 1A is equipped with a receiving unit 105A that receives instructions for correction of the inference results. The determination unit 103A then re-determines whether the insurance product described in the related description meets the desired conditions based on the correction instructions received by the receiving unit 105A. This provides the additional benefit of correcting errors in the inference results so that a suitable insurance product is recommended, in addition to the effects achieved by the information processing device 1.

[0076] Furthermore, as described above, the reception unit 105A may accept correction instructions expressed in natural language. In this case, the determination unit 103A generates a prompt (for example, prompt 602 in Figure 6) that includes the correction instructions received by the reception unit 105A and instructs the determination unit 103A to infer the relationship between the relevant description and the desired conditions based on the correction instructions. Based on the output obtained by inputting the generated prompt into the language model M2, the determination unit 103A re-determines whether the insurance product described by the relevant description meets the desired conditions. This provides the effect of enabling appropriate re-determination by understanding the intent of the correction instructions, in addition to the effects of the information processing device 1.

[0077] An example of how the intent behind a correction instruction can be understood is the absorption of variations in wording. For example, even if a correction instruction such as "I was hospitalized for a cerebral hemorrhage" or "I was hospitalized for a subarachnoid hemorrhage, so I don't think I can enroll" is entered instead of the comment "I was hospitalized for a subarachnoid hemorrhage three years ago," as in the example in Figure 6, it is still possible to determine through re-inference that the desired conditions are not met.

[0078] Furthermore, accepting correction instructions expressed in natural language has the advantage of allowing the content of the correction instructions to be reflected in other inferences. For example, suppose the inference result that the related description "We will compensate for accidents caused by bicycles" satisfies the desired condition "Cycling is my hobby" is entered as a correction instruction, and the instruction is "I have separate insurance for bicycles, so I do not need compensation for bicycles." In this case, the judgment unit 103A can not only correct this inference result to "Does not meet the desired condition," but can also update other inference results based on the fact that the subject already has insurance for bicycles.

[0079] Note that correction instructions do not necessarily have to be entered in natural language. For example, the presentation control unit 104A may present options for selecting whether the inference content is appropriate or not, and the reception unit 105A may notify the judgment unit 103A when an option indicating that the inference content is inappropriate is selected. In this case, the judgment unit 103A, upon receiving this notification, will change the judgment result for the relevant related description. That is, if the judgment unit 103A receives input indicating that the inference content is inappropriate for an inference result of "meets the desired conditions" for a certain related description, it will change the judgment result for that related description to "does not meet the desired conditions". Similarly, if the judgment unit 103A receives input indicating that the inference content is inappropriate for an inference result of "does not meet the desired conditions" for a certain related description, it will change the judgment result for that related description to "meets the desired conditions".

[0080] (Reuse of correction instructions) The content of the correction instructions described above may be recorded in the memory unit 11A or an external database, etc., and used for subsequent inferences. For example, the comment of the correction instruction, "I have separate insurance for my bicycle, so I do not need coverage for my bicycle," may be recorded, and this comment may be used for inferences regarding other related descriptions. This allows the determination unit 103A to make inferences in subsequent inferences that the subject already has insurance for their bicycle. Furthermore, if the content of the correction instructions is relevant to persons other than the specific subject, it can be reused for inferences regarding other persons.

[0081] Furthermore, the presentation control unit 104A may present the recorded content of the correction instructions to the user, and the reception unit 105A may accept corrections, deletions, and additions to the content of the correction instructions. This allows the content of the correction instructions, which align with the user's intentions, to be reflected in subsequent inferences without requiring retraining of the language model M2. The correction instructions can be reused in the same way as when they are used for the first time, by including them in the prompt. In other words, the determination unit 103A generates a prompt that includes the recorded correction instructions and instructs the language model M2 to infer the relationship between the relevant description and the desired conditions based on the correction instructions, and then inputs the generated prompt to the language model M2 to output an inference result indicating whether the relevant description satisfies the desired conditions.

[0082] (Regarding iterative reasoning) Since the language model M2 is a probabilistic model, even if the exact same prompt is entered, the inference results in multiple inferences may differ. In addition, it has been found that inference results that differ from the facts tend not to be output repeatedly. For this reason, the determination unit 103A may perform the process of inputting a prompt to the language model M2 and outputting an inference result multiple times. In this case, the determination unit 103A may determine that insurance products corresponding to related descriptions with large variations in inference results do not meet the desired conditions. For example, the determination unit 103A may calculate a score indicating the magnitude of the variation in inference results (for example, the proportion of inference results that differ in content from other inference results out of all inference results), and determine that insurance products corresponding to related descriptions whose calculated score exceeds a predetermined threshold do not meet the desired conditions.

[0083] (Regarding additions or changes to desired conditions) The information processing device 1A may accept additions or changes to desired conditions. This will be explained with reference to Figure 7. Figure 7 shows an example of a UI screen that accepts additions and changes to desired conditions. The example screen 701 shown in Figure 7 displays the recommended insurance product, i.e., the insurance product that the determination unit 103A has determined to meet the desired conditions, and the reason for suggesting that insurance product. As mentioned above, this reason for suggestion is the basis for the inference of whether or not the insurance product corresponding to the related description, output by the language model M2, meets the desired conditions.

[0084] Furthermore, screen example 701 displays a button (software key) to proceed with the application for the recommended insurance product. Displaying such a button allows for a smoother application process for the insurance product. When the button is operated, the information processing device 1A performs a predetermined process to assist the user in proceeding with the application for the insurance product. For example, when the button is operated, the presentation control unit 104A may perform a process to display contact information for the insurance company or agency selling the insurance product, or detailed information such as the terms and conditions of the insurance product.

[0085] Furthermore, screen example 701 also displays a text box for entering and modifying desired conditions, and a button (software key) for instructing the system to re-propose insurance products based on the desired conditions entered in the text box. In the example in Figure 7, this text box displays the desired conditions already entered by the user, as well as additional desired conditions newly entered by the user. Thus, when accepting modifications or additions to desired conditions, the presentation control unit 104A may present the desired conditions already entered by the user. The reception unit 105A may then accept modifications to the desired conditions that were not presented.

[0086] When additional desired conditions are entered in the text box above, and the button to instruct the user to request a revised insurance product is pressed, the acquisition unit 101A acquires the newly entered desired conditions as new condition information. Subsequently, as in the case where the condition information was acquired earlier, the extraction unit 102A extracts relevant information, and the determination unit 103A makes a determination. Based on the determination result, recommended insurance products are presented again.

[0087] (Regarding the use of historical information) The condition information acquired by the acquisition unit 101A may be the desired insurance conditions entered by the subject, or it may be historical information regarding the subject's desired insurance conditions, or it may be both. In this case, if the acquisition unit 101A acquires both the information entered by the subject and the historical information as condition information, there may be items that are not shown in the information entered by the subject but are shown in the historical information that are useful for recommending the most suitable insurance to the subject.

[0088] Therefore, the acquisition unit 101A may perform a process of acquiring desired conditions from the information entered by the subject, as well as from the history information, and comparing these desired conditions. Then, if the presentation control unit 104A detects desired conditions that were not acquired from the information entered by the subject but were acquired from the history information through the above process, it may present those desired conditions to the subject and prompt them to input whether or not to use them for recommending insurance products. This makes it possible to supplement the information entered by the subject and recommend insurance products that are suitable for the subject.

[0089] Furthermore, the historical information acquired by the acquisition unit 101A may not contain sufficient information to be used for recommending insurance products. For this reason, the presentation control unit 104A may present the historical information to the subject and prompt them to input an explanatory text describing that historical information. This will be explained with reference to Figure 8.

[0090] Figure 8 shows an example of a UI screen that accepts input for a description of historical information. In the example screen 801 shown in Figure 8, a message is displayed indicating that historical information that should be considered when selecting an insurance product has been detected, and the detected historical information is displayed. Specifically, this historical information indicates that the patient was hospitalized at Y Hospital between September 1st and 5th, 2021.

[0091] Furthermore, screen example 801 displays text prompting the user to enter the reason for hospitalization and press the "Output Recommended Insurance Products" button. In addition, screen example 801 also displays a text box for entering the reason for hospitalization and the aforementioned "Output Recommended Insurance Products" button. The "Output Recommended Insurance Products" button is a software key that instructs the system to output insurance products considering the entered reason. The reception unit 105A can accept input of the reason for hospitalization, or in other words, a description of the history information, through such a UI screen.

[0092] In example screen 801, when the reason for hospitalization is entered in the text box and the "Output Recommended Insurance Products" button is operated, the extraction unit 102A extracts relevant information. In this case, the extraction unit 102A uses the pair of detected historical information and the entered reason for hospitalization as a single desired condition and extracts relevant information related to that desired condition. For example, if the entered reason for hospitalization is "heart failure," the extraction unit 102A inputs the sentence "2021 / 9 / 1-5: Hospitalized at Y Hospital, Reason for hospitalization: heart failure" as a desired condition into the extraction model M1 and extracts relevant information. Subsequently, the judgment unit 103A performs a judgment and presents a result indicating whether the extracted relevant information satisfies the desired condition shown in the historical information, or in other words, whether it satisfies the enrollment conditions for the insurance product corresponding to the relevant information.

[0093] In the example shown in Figure 8, the detected history information indicates a hospitalization history. Therefore, the reception unit 105A accepts input for the reason for hospitalization. Thus, the reception unit 105A only needs to accept input for additional information corresponding to the content of the history information, in other words, for explanatory text that explains the history information. For example, if history information indicating a drug prescription history is detected, the reception unit 105A only needs to accept input for explanatory text about the prescribed drug.

[0094] As described above, the information processing device 1A includes a presentation control unit 104A that presents historical information regarding the desired conditions of the insured person, and a reception unit 105A that receives input of an explanatory text describing the presented historical information. The extraction unit 102A then extracts related descriptions associated with the desired conditions, treating the historical information and the explanatory text as a single desired condition. This provides the effect of being able to extract appropriate related descriptions that take into account the input explanatory text, in addition to the effects performed by the information processing device 1. Furthermore, the determination unit 103A may use the language model M2 to determine whether or not the insurance product satisfies the desired conditions, including the historical information and the explanatory text. This provides the effect of being able to obtain an appropriate determination result that takes into account the input explanatory text.

[0095] (Process flow) The processing flow performed by the information processing device 1A will be explained with reference to Figure 9. Figure 9 is a flowchart showing the processing flow performed by the information processing device 1A. The flowchart in Figure 9 includes each process of the support method according to this exemplary embodiment.

[0096] In S11 (acquisition process), the acquisition unit 101A acquires condition information indicating the subject's desired conditions for insurance. For example, the acquisition unit 101A may acquire condition information that the subject inputs to the information processing device 1A.

[0097] In S12 (extraction process), the extraction unit 102A uses an extraction model M1, which has been trained to take a document and data pair as input and output the relevant portion of the document related to the data, to extract relevant descriptions related to the desired conditions shown in the condition information obtained in S11 from a document describing candidate insurance products to recommend to the target person. As mentioned above, the condition information may show multiple desired conditions, and multiple sets of condition information may be obtained. In this case, if multiple desired conditions are identified, in S12 the extraction unit 102A extracts relevant descriptions related to each desired condition. The extraction unit 102A may also extract multiple relevant descriptions for a single desired condition. The document describing candidate insurance products to recommend to the target person may be input along with the condition information in S11, or it may be obtained from a predetermined database as shown in the example in Figure 4.

[0098] In S13, the determination unit 103A generates a prompt to be input to the language model M2. Specifically, the determination unit 103A generates a prompt that includes the relevant description extracted in S12 and the desired conditions shown in the condition information obtained in S11, and instructs the system to infer the relationship between the relevant description and the desired conditions, in other words, whether or not the insurance product corresponding to the relevant description satisfies the desired conditions.

[0099] In S14, the determination unit 103A inputs the prompt generated in S13 into the language model M2 to infer the relationship between the relevant description and the desired conditions, in other words, whether the insurance product corresponding to the relevant description meets the desired conditions. In S15, the presentation control unit 104A presents the inference result from S14 to the subject. In S16, the reception unit 105A determines whether there is a request for correction to the inference result presented in S15. If the result in S16 is YES, the process proceeds to S17; if the result in S16 is NO, the process proceeds to S19.

[0100] In S15, the presentation control unit 104A may present the inference result by displaying a UI screen such as the example screen 503 in Figure 5 or the example screen 601 in Figure 6. If a UI screen such as the example screen 601 is displayed, the reception unit 105A may receive a correction instruction via that UI screen. The process of presenting the inference result may be omitted. In that case, the process proceeds from S14 to S19.

[0101] In S17, the reception unit 105A records the contents of the received correction instruction in the storage unit 11A or an external database, etc. The timing of recording the contents of the correction instruction is arbitrary. For example, the reception unit 105A may record the contents of the correction instruction after the determination described later has been completed, or after the insurance product has been presented.

[0102] In S18, the determination unit 103A generates a prompt that reflects the content of the correction instruction received by the reception unit 105A, specifically a prompt that includes the correction instruction and instructs the system to infer the relationship between the desired conditions and related descriptions based on the correction instruction. After this, the process returns to S14, where the determination unit 103A inputs the newly generated prompt into the language model M2 to infer the relationship between the desired conditions and related descriptions.

[0103] In S19, the determination unit 103A determines, based on the inference results of the language model M2, whether the insurance product described by the related description extracted in S12 satisfies the desired conditions indicated in the condition information obtained in S11. If the language model M2 performs inference multiple times, the processing in S19 is performed based on the most recent inference result among the multiple inference results.

[0104] In S20, the presentation control unit 104A presents the insurance products that were determined to meet the desired conditions in S19 as recommended insurance products to the subject. For example, the presentation control unit 104A may present insurance products by displaying a UI screen such as the example screen 701 in Figure 7. If there are multiple insurance products that were determined to meet the desired conditions in S19, the presentation control unit 104A may present all of those insurance products or some of them. For example, if multiple desired conditions are specified, the presentation control unit 104A may prioritize presenting insurance products that meet the most desired conditions.

[0105] In S21, the reception unit 105A determines whether or not it has received a request for a resubmission. For example, the reception unit 105A may determine that it has received a request for a resubmission (YES in S21) if the "Resubmit" button in the example screen 701 of Figure 7 is operated, or that it has not received a request for a resubmission (NO in S21) if the "Apply" button is operated. If NO is determined in S21, the process in Figure 9 ends. If NO is determined in S21, the presentation control unit 104A may end the presentation of the insurance product, or it may provide support to proceed with the application for the insurance product (for example, presenting more detailed information about the insurance product, presenting the application destination, etc.).

[0106] On the other hand, if the result in S21 is YES, the process returns to S11. In S11, which is accessed after transitioning from S21, the acquisition unit 101A acquires new condition information. For example, if a UI screen like the example screen 701 in Figure 7 is displayed in S20, the acquisition unit 101A only needs to acquire the new condition information entered via that UI screen.

[0107] [Reference example 1] Figure 10 is a block diagram showing the configuration of the information processing device 1B according to this reference example. As shown in the figure, the information processing device 1B includes an acquisition unit 101B and a determination unit 103B.

[0108] The acquisition unit 101B acquires condition information indicating the subject's desired conditions for insurance, similar to the acquisition unit 101A in the exemplary embodiment 2.

[0109] Similar to the determination unit 103A in the exemplary embodiment 2, the determination unit 103B determines whether the insurance product satisfies the desired conditions indicated in the condition information acquired by the acquisition unit 101B, using a document describing the insurance product and a language model trained on natural language.

[0110] The above document may be extracted using an extraction model from a document describing candidate insurance products to be recommended to the target person, as in exemplary embodiment 1 or 2 (i.e., related descriptions), or it may not be extracted using an extraction model. For example, the determination unit 103B can have the target person input a part of a document describing candidate insurance products to be recommended to the target person (for example, a coherent part such as a sentence or paragraph), or extract it by analyzing the document, and perform the above determination using that part.

[0111] As described above, the information processing device 1B includes an acquisition unit 101B that acquires condition information indicating the desired conditions of the insured person for insurance, and a determination unit 103B that determines whether or not an insurance product satisfies the desired conditions indicated in the condition information acquired by the acquisition unit 101B, using a document describing the insurance product and a language model that has been trained on natural language. With this information processing device 1B, an objective determination result can be obtained as to whether or not an insurance product satisfies the desired conditions. Therefore, the effect is obtained that it becomes possible to simplify the process of finding an insurance product that satisfies the desired conditions of the insured person.

[0112] Furthermore, how the determination result of the determination unit 103B is used to facilitate the search for insurance products is optional. For example, the information processing device 1B may be equipped with a presentation control unit 104A similar to the information processing device 1A in the exemplary embodiment 2. In this case, the presentation control unit 104A can be made to present the determination result of the determination unit 103B, or insurance products that the determination unit 103B has determined to meet the desired conditions. This allows the user to consider insurance products that suit them by referring to the presented determination results or insurance products.

[0113] (Support Program) The functions of the information processing device 1B described above can also be implemented by a program. The support program in this reference example functions as an acquisition means for acquiring condition information indicating the desired conditions of the person concerned regarding insurance, and as a determination means for determining whether an insurance product meets the above desired conditions using a document describing the insurance product and a language model that has been trained on natural language. This support program has the effect of making it possible for the person concerned to find an insurance product that meets their desired conditions.

[0114] (Support method) The support method described in this reference example is a method for supporting the application for an insurance contract, in which at least one processor performs an acquisition process to acquire condition information indicating the applicant's desired conditions for insurance, and a determination process to determine whether an insurance product meets the above desired conditions, using a document describing the insurance product and a language model that has been trained on natural language. This support method has the effect of making it possible for the applicant to find an insurance product that meets their desired conditions.

[0115] [Reference example 2] In the exemplary embodiment described above, information processing devices 1 and 1A were described that acquire condition information indicating the subject's desired conditions for insurance, and extract relevant descriptions related to the above desired conditions from a document describing candidate insurance products to be recommended to the subject.

[0116] Furthermore, the above-mentioned reference example describes an information processing device 1B that acquires condition information indicating the desired conditions of the insured person, and determines whether or not the insurance product meets the above desired conditions using a document describing the insurance product and a language model that has been trained on natural language.

[0117] These information processing devices 1, 1A, and 1B can be used not only to assist in finding insurance but also to assist in searching for any other object. For example, information processing devices 1, 1A, and 1B can be used when searching for products or services other than insurance products.

[0118] In this process, the user simply inputs conditional information indicating their desired goods or services into information processing device 1 or 1A, and allows information processing device 1 or 1A to refer to documents describing those goods or services. This allows the information processing device 1 or 1A to extract relevant information related to the user's desired conditions from the documents. By having information processing device 1 or 1A present the extracted relevant information, the user can easily determine whether the goods or services described by that information meet their desired conditions and efficiently consider purchasing those goods or services.

[0119] Furthermore, the user may input condition information indicating their desired conditions for the products or services they wish to purchase into the information processing device 1B, and also have the information processing device 1B refer to documents describing those products or services. This allows the information processing device 1B to output a determination result indicating whether or not the products or services described in the documents meet the desired conditions. The user can then use the presented determination result to efficiently consider purchasing the products or services.

[0120] In addition, information processing devices 1, 1A, and 1B can also be used to search for countries or regions that meet desired conditions, individuals that meet desired conditions, product specifications that meet desired conditions, etc.

[0121] [Variation] The entities executing each process described in the above-described exemplary embodiments and reference examples are arbitrary and not limited to the examples given above. For example, a system having the same functions as information processing devices 1, 1A, and 1B can be constructed using multiple devices that can communicate with each other. Furthermore, the entities executing each process shown in the flowchart in Figure 9 may be a single device (which can also be called a processor) or multiple devices (which can also be called processors).

[0122] [Examples of implementation using software] Some or all of the functions of the information processing devices 1, 1A, and 1B (hereinafter also referred to as "the above devices") may be implemented by hardware such as integrated circuits (IC chips) or by software.

[0123] In the latter case, each of the above devices is implemented, for example, by a computer that executes instructions for a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Figure 11. Figure 11 is a block diagram showing the hardware configuration of computer C, which functions as each of the above devices.

[0124] Computer C comprises at least one processor C1 and at least one memory C2. Memory C2 stores a program (support program) P that causes computer C to operate as each of the above-mentioned devices. In computer C, processor C1 reads program P from memory C2 and executes it, thereby realizing each of the above-mentioned devices.

[0125] For processor C1, for example, a CPU (Central Processing Unit), GPU (Graphic Processing Unit), DSP (Digital Signal Processor), MPU (Micro Processing Unit), FPU (Floating Point Number Processing Unit), PPU (Physics Processing Unit), TPU (Tensor Processing Unit), quantum processor, microcontroller, or a combination thereof can be used. For memory C2, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), or a combination thereof can be used.

[0126] Computer C may also be equipped with RAM (Random Access Memory) for loading program P at runtime and for temporarily storing various data. Furthermore, computer C may be equipped with communication interfaces for sending and receiving data with other devices. Additionally, computer C may be equipped with input / output interfaces for connecting input / output devices such as keyboards, mice, displays, and printers.

[0127] Furthermore, program P can be recorded on a non-temporary, tangible recording medium M that is readable by computer C. Such a recording medium M could be, for example, tape, disk, card, semiconductor memory, or programmable logic circuitry. Computer C can acquire program P via such a recording medium M. Program P can also be transmitted via a transmission medium. Such a transmission medium could be, for example, a communication network or broadcast waves. Computer C can also acquire program P via such a transmission medium.

[0128] Furthermore, each of the above functions of each of the above devices may be implemented by a single processor in a single computer, by multiple processors in a single computer working together, or by multiple processors in each of multiple computers working together. In addition, the programs for implementing each of the above functions in each of the above devices may be stored in a single memory in a single computer, distributed and stored in multiple memories in a single computer, or distributed and stored in multiple memories in each of multiple computers.

[0129] [Additional Notes] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0130] (Note A1) An information processing device comprising: an acquisition means for acquiring condition information indicating the desired conditions of an insurance policyholder; and an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the policyholder, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data.

[0131] (Appendix A2) The information processing device described in Appendix A1 comprises: a determination means for determining whether the insurance product described in the above related description meets the desired conditions using a language model that has been trained on natural language; and a presentation control means for presenting the insurance product that the determination means has determined to meet the desired conditions as a recommended insurance product to the subject.

[0132] (Note A3) The information processing device according to Appendix A2, wherein the determination means includes the related description and the desired conditions, generates a prompt instructing the language model to infer the relationship between the related description and the desired conditions, and determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

[0133] (Note A4) The information processing device according to Appendix A3, wherein the determination means generates a prompt instructing the determination means to output the basis for the inference along with the inference result of the relationship between the related description and the desired conditions, and the presentation control means presents the inference result output by the language model, the determination result of the determination means, or an insurance product that the determination means has determined to satisfy the desired conditions, along with the basis.

[0134] (Note A5) The information processing device described in Appendix A4, comprising a receiving means for receiving instructions to modify the aforementioned inference result, wherein the determination means re-determines whether the insurance product described by the related description satisfies the aforementioned desired conditions based on the modification instructions received by the receiving means.

[0135] (Note A6) The information processing device described in Appendix A5, wherein the receiving means receives a modification instruction expressed in natural language, the determination means generates a prompt that includes the modification instruction received by the receiving means and instructs the determination means to infer the relationship between the desired conditions and the related description based on the modification instruction, and based on the output obtained by inputting the generated prompt into the language model, the determination means re-determines whether the insurance product described by the related description satisfies the desired conditions.

[0136] (Note A7) An information processing device according to any one of Appendix A1 to A6, comprising: a presentation control means for presenting historical information relating to the subject's desired conditions for insurance; and a reception means for receiving input of an explanatory text describing the presented historical information, wherein the extraction means extracts related descriptions relating to the desired conditions, with the set of the historical information and the explanatory text being treated as a single desired condition.

[0137] (Note A8) An information processing device comprising: an acquisition means for acquiring condition information indicating the desired conditions of an insured person; and a determination means for determining whether an insurance product meets the said desired conditions, using a document describing the insurance product and a language model that has been trained on natural language.

[0138] (Note B1) A support method that performs, at least one processor, an acquisition process to acquire condition information indicating the desired conditions of an insurance policyholder, and an extraction process to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the policyholder, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data.

[0139] (Note B2) The support method according to Appendix B1, which includes a determination process in which at least one processor determines whether the insurance product described in the related description satisfies the desired conditions using a language model that has been trained on natural language, and a presentation control process in which the at least one processor presents the insurance product that it has determined to satisfy the desired conditions in the determination process as a recommended insurance product to the target person.

[0140] (Note B3) The support method according to Appendix B2, wherein in the determination process, the at least one processor generates a prompt that includes the related description and the desired conditions and instructs the processor to infer the relationship between the related description and the desired conditions, and determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

[0141] (Note B4) The support method according to Appendix B3, wherein in the determination process, the at least one processor generates a prompt instructing the system to output the basis for the inference along with the inference result of the relationship between the related description and the desired conditions, and the at least one processor presents the inference result output by the language model, the determination result of the determination process, or an insurance product determined to satisfy the desired conditions in the determination process, along with the basis for the inference.

[0142] (Note B5) The support method according to Appendix B4, wherein the at least one processor includes an acceptance process for receiving instructions to modify the inference result, and the at least one processor re-determines whether the insurance product described in the related description meets the desired conditions based on the instructions received in the acceptance process.

[0143] (Note B6) The support method described in Appendix B5, wherein in the reception process, the at least one processor receives a correction instruction expressing the content of the correction in natural language, the at least one processor generates a prompt that includes the correction instruction received in the reception process and instructs the system to infer the relationship between the desired conditions and the related description based on the correction instruction, and based on the output obtained by inputting the generated prompt into the language model, the system re-determines whether the insurance product described by the related description satisfies the desired conditions.

[0144] (Note B7) The support method according to any one of Appendix B1 to B6, comprising: a presentation control process in which at least one processor presents historical information relating to the subject's desired conditions for insurance; and a reception process in which at least one processor accepts input of an explanatory text describing the presented historical information, wherein in the extraction process, the at least one processor extracts related descriptions relating to the desired conditions, with the pair of historical information and the explanatory text being treated as one desired condition.

[0145] (Note B8) A support method in which at least one processor performs an acquisition process to acquire condition information indicating the desired conditions of an insured person, and a determination process to determine whether an insurance product meets the said desired conditions, using a document describing the insurance product and a language model trained on natural language.

[0146] (Note C1) A support program that enables a computer to function as an acquisition means for acquiring condition information indicating the desired conditions of an insurance policyholder, and an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the policyholder, using an extraction model trained on machine learning to take a document and data set as input and output the portion of the document related to the data.

[0147] (Note C2) The support program described in Appendix C1 causes the computer to function as a determination means for determining whether the insurance product described in the related description meets the desired conditions using a language model that has learned natural language through machine learning, and a presentation control means for presenting the insurance product that the determination means has determined to meet the desired conditions to the target person as a recommended insurance product.

[0148] (Note C3) The support program described in Appendix C2 includes the relevant description and the desired conditions, generates a prompt instructing the language model to infer the relationship between the relevant description and the desired conditions, and determines whether the insurance product described by the relevant description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

[0149] (Note C4) The support program described in Appendix C3, wherein the determination means generates a prompt instructing the determination means to output the basis for the inference along with the inference result of the relationship between the related description and the desired conditions, and the presentation control means presents the inference result output by the language model, the determination result of the determination means, or an insurance product that the determination means has determined to satisfy the desired conditions, along with the basis.

[0150] (Note C5) The support program described in Appendix C4, wherein the computer functions as a receiving means for receiving instructions to correct the inference result, and the determination means re-determines whether the insurance product described in the related description meets the desired conditions based on the instructions for correction received by the receiving means.

[0151] (Appendix C6) The support program described in Appendix C5, wherein the receiving means receives a modification instruction expressed in natural language, the determination means generates a prompt that includes the modification instruction received by the receiving means and instructs the determination means to infer the relationship between the desired conditions and the related description based on the modification instruction, and the determination means re-determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

[0152] (Note C7) The computer functions as a presentation control means for presenting historical information relating to the subject's desired conditions for insurance, and a reception means for receiving input of an explanatory text describing the presented historical information, and the extraction means is a support program as described in any of Appendix C1 to C6, which takes the pair of the historical information and the explanatory text as a single desired condition and extracts related descriptions associated with that desired condition.

[0153] (Note C8) A support program that enables a computer to function as an acquisition means for obtaining condition information indicating the desired conditions of an insured person, and as a determination means for determining whether an insurance product meets the said desired conditions, using a document describing the insurance product and a language model trained on natural language.

[0154] (Note D1) An information processing device comprising at least one processor, the at least one processor performing an acquisition process to acquire condition information indicating the desired conditions of an insured person, and an extraction process to extract relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the insured person, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data.

[0155] The information processing device may also include memory. Furthermore, the memory may store a program that causes at least one processor to execute each of the aforementioned processes.

[0156] (Note D2) The information processing device described in Appendix D1, wherein the at least one processor performs a determination process that determines whether the insurance product described in the related description satisfies the desired conditions using a language model that has been trained on natural language, and a presentation control process that presents the insurance product determined to satisfy the desired conditions in the determination process as a recommended insurance product to the subject.

[0157] (Note D3) The information processing device according to Appendix D2, wherein in the determination process, the at least one processor generates a prompt that includes the related description and the desired conditions and instructs the system to infer the relationship between the related description and the desired conditions, and determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt to the language model.

[0158] (Note D4) The information processing device according to Appendix D3, wherein in the determination process, the at least one processor generates a prompt instructing to output the basis for the inference along with the inference result of the relationship between the related description and the desired conditions, and the at least one processor presents the inference result output by the language model, the determination result in the determination process, or an insurance product determined to satisfy the desired conditions in the determination process, along with the basis.

[0159] (Note D5) An information processing device according to Appendix D4, wherein it performs an acceptance process to receive instructions for correction to the inference result, and in the determination process, at least one processor re-determines whether the insurance product described in the related description satisfies the desired conditions based on the instructions for correction received in the acceptance process.

[0160] (Note D6) The information processing device according to Appendix D5, wherein in the reception process, the at least one processor receives a modification instruction expressing the modification in natural language, generates a prompt that includes the modification instruction received in the reception process and instructs the system to infer the relationship between the desired conditions and the related description based on the modification instruction, and inputs the generated prompt into the language model to re-determine whether the insurance product described by the related description satisfies the desired conditions based on the output obtained.

[0161] (Note D7) An information processing device according to any one of the appendices D1 to D6, which performs a presentation control process for presenting historical information regarding the subject's desired conditions for insurance, and a reception process for receiving input of an explanatory text describing the presented historical information, wherein in the extraction process, at least one processor extracts related descriptions relating to the desired conditions, with the pair of the historical information and the explanatory text being treated as one desired condition.

[0162] (Note D8) An information processing device comprising at least one processor, wherein the at least one processor performs an acquisition process to acquire condition information indicating the desired conditions of an insured person, and a determination process to determine whether an insurance product meets the said desired conditions, using a document describing the insurance product and a language model trained on natural language.

[0163] (Note E1) A non-temporary recording medium that records a support program that causes a computer to perform an acquisition process to acquire condition information indicating the desired conditions of an insured person, and an extraction process that uses a machine learning extraction model that takes a set of documents and data as input and outputs the portion of the document related to the data to be extracted from a document describing candidate insurance products to be recommended to the insured person, and extracts relevant descriptions related to the desired conditions.

[0164] (Note E2) A non-temporary recording medium that records a support program that causes a computer to perform an acquisition process to acquire condition information indicating the desired conditions of an insured person, and a determination process to determine whether an insurance product meets the said desired conditions, using a document describing the insurance product and a language model trained on natural language. [Explanation of Symbols]

[0165] 1. Information Processing Device 101 Acquisition unit (acquisition means) 102 Extraction part (extraction means) 1A Information Processing Device 101A Acquisition part (acquisition means) 102A Extraction part (extraction means) 103A Judgment unit (judgment means) 104A Presentation Control Unit (Presentation Control Means) 105A Reception area (reception method) M1 Extraction Model M2 Language Model

Claims

1. A means of obtaining conditional information that indicates the desired conditions of the person covered by insurance, An information processing device comprising: an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the subject, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data in the document.

2. A determination means for determining whether the insurance product described in the above related description meets the above desired conditions, using a language model that has learned natural language through machine learning, The information processing apparatus according to claim 1, further comprising: a presentation control means for presenting to the subject a target insurance product that the determination means has determined to satisfy the desired conditions as a recommended insurance product.

3. The information processing apparatus according to claim 2, wherein the determination means includes the related description and the desired conditions, generates a prompt instructing the language model to infer the relationship between the related description and the desired conditions, and determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

4. The determination means generates a prompt that instructs the system to output the basis for the inference, along with the inference result regarding the relationship between the related description and the desired conditions. The information processing device according to claim 3, wherein the presentation control means presents the inference result output by the language model, the determination result of the determination means, or an insurance product that the determination means has determined to satisfy the desired conditions, along with the basis for that determination.

5. The system includes a receiving means for receiving instructions to modify the aforementioned inference results, The information processing device according to claim 4, wherein the determination means re-determines whether the insurance product described by the related description satisfies the desired conditions based on the modification instructions received by the receiving means.

6. The aforementioned receiving means receives correction instructions that express the content of the correction in natural language, The information processing device according to claim 5, wherein the determination means includes a correction instruction received by the receiving means, generates a prompt instructing the determination means to infer the relationship between the desired conditions and the related description based on the correction instruction, and re-determines whether the insurance product described by the related description satisfies the desired conditions based on the output obtained by inputting the generated prompt into the language model.

7. A presentation control means that presents historical information regarding the desired conditions of the aforementioned person for insurance, The system includes a receiving means for receiving input of an explanatory text describing the presented historical information, The information processing apparatus according to any one of claims 1 to 6, wherein the extraction means extracts related descriptions associated with the desired condition, with the set of the historical information and the explanatory text being treated as a single desired condition.

8. A means of obtaining conditional information that indicates the desired conditions of the person covered by insurance, An information processing device comprising: determination means for determining whether an insurance product meets the aforementioned desired conditions, using a document describing the insurance product and a language model that has been trained on natural language.

9. At least one processor, A process to obtain condition information indicating the desired conditions of the person covered by the insurance, A support method for performing an extraction process that extracts relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the target person, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data.

10. Computers, A means for obtaining condition information that indicates the desired conditions of the person covered by insurance, and A support program that functions as an extraction means for extracting relevant descriptions related to the desired conditions from a document describing candidate insurance products to be recommended to the target person, using an extraction model trained on machine learning to take a pair of documents and data as input and output the portion of the document related to the data in question.

Images