Information processing device, method, and program

Abstract

To increase credibility of output of AI.SOLUTION: An information processing device comprises a control section for inputting multiple variables 51 showing various kinds of features that an evaluated object has to a first learned model 11 for evaluating individual objects, inputting an evaluation value 52 outputted from the first learned model 11, multiple explanation values showing an influence of the respective variables included in the multiple variables 51 on the valuation value 52, and a fill-in-the-blank question to a second learned model 12 explaining each output from the first learned model 11, and outputting an explanation sentence 55 that is generated in a fill-in-the blank format by the second learned model 12 and explains an influence of at least one feature included in various kinds of features on the evaluation value 52.SELECTED DRAWING: Figure 2

Description

[Technical Field]

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

[0002] Patent document 1 discloses an information processing device that presents, in a graph, a first importance level relating to the effect common to multiple treatment options, a second importance level relating to the difference in effect between multiple treatment options, or both, for each of one or more features that affect the treatment effect. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2023-103039 Summary of the Invention [Problem to be solved by the invention]

[0004] When AI determines that a certain feature has the greatest influence on the output, simply presenting the importance of each feature as disclosed in Patent Document 1 may result in the user being unable to trust the AI's output, since the user will not know on what basis the AI ​​determined that that one feature has the greatest influence on the output. "AI" is an abbreviation for artificial intelligence.

[0005] The purpose of this disclosure is to increase the credibility of AI output. [Means for solving the problem]

[0006] An information processing device according to an embodiment of the present disclosure includes: Multiple variables that indicate the various characteristics of the object to be evaluated are input into the first trained model that evaluates individual objects, An evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a hole-punched sentence are input into a second trained model that explains each output from the first trained model; outputting an explanatory sentence, generated in a fill-in-the-blank format by the second trained model, explaining the influence of at least one feature included in the various features on the evaluation value; It has a control unit.

[0007] According to one embodiment of the present disclosure, a method comprises: an information processing device inputting a plurality of variables indicating various characteristics of the object to be evaluated into a first trained model that evaluates each individual object; The information processing device inputs an evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a punched sentence into a second trained model that explains each output from the first trained model; the information processing device outputs an explanatory text that is generated in a fill-in-the-blank format by the second trained model and that explains the influence that at least one feature included in the various features has on the evaluation value; and Includes:

[0008] A program according to an embodiment of the present disclosure includes: Inputting a plurality of variables indicating various characteristics of the object to be evaluated into a first trained model that evaluates each individual object; Inputting an evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a punched sentence into a second trained model that explains each output from the first trained model; outputting an explanatory sentence that explains the influence of at least one feature included in the various features on the evaluation value, the explanatory sentence being generated in a fill-in-the-blank format by the second trained model; and The computer is caused to perform operations including: [Effects of the Invention]

[0009] This disclosure increases the credibility of AI output. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a block diagram illustrating a configuration of a system according to an embodiment of the present disclosure. [Figure 2] FIG. 1 is a diagram illustrating an overview of an embodiment of the present disclosure. [Figure 3] 10 is a graph illustrating an example of multiple explanatory values ​​according to an embodiment of the present disclosure. [Figure 4] 1A and 1B are diagrams illustrating examples of hole-punched sentences according to embodiments of the present disclosure. [Figure 5] FIG. 10 is a diagram illustrating an example of an output including an evaluation value and an explanatory text according to an embodiment of the present disclosure. [Figure 6] 10 is a flowchart illustrating an operation of an information processing device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings.

[0012] In each drawing, the same or corresponding parts are denoted by the same reference numerals. In the description of this embodiment, the description of the same or corresponding parts will be omitted or simplified as appropriate.

[0013] The configuration of a system 10 according to this embodiment will be described with reference to FIG.

[0014] The system 10 according to this embodiment includes an information processing device 20 and a terminal device 30. The information processing device 20 is capable of communicating with the terminal device 30 via a network 40.

[0015] The information processing device 20 is a computer such as a server that is installed in a facility such as a data center and that belongs to a cloud computing system or other computing system.

[0016] The terminal device 30 is a mobile device used by a user, such as a PC, a mobile phone, a smartphone, or a tablet. "PC" is an abbreviation for personal computer.

[0017] Network 40 may include the Internet, at least one WAN, at least one MAN, or any combination thereof. "WAN" is an abbreviation for wide area network. "MAN" is an abbreviation for metropolitan area network. Network 40 may include at least one wireless network, at least one optical network, or any combination thereof. A wireless network may be, for example, an ad-hoc network, a cellular network, a wireless LAN, a satellite communication network, or a terrestrial microwave network. "LAN" is an abbreviation for local area network.

[0018] The present embodiment will be outlined with reference to FIGS. 1 to 5. FIG.

[0019] The information processing device 20 inputs a plurality of variables 51 indicating various characteristics of an object to be evaluated to a first trained model 11 that evaluates each object. The information processing device 20 inputs an evaluation value 52 output from the first trained model 11, a plurality of explanatory values ​​53 indicating the influence of each variable included in the plurality of variables 51 on the evaluation value 52, and a blank statement 54 to a second trained model 12 that explains each output from the first trained model 11. The information processing device 20 outputs, via the network 40, an explanatory statement 55 that is generated by the second trained model 12 in a fill-in-the-blank format and that explains the influence of at least one feature included in the various characteristics on the evaluation value 52. The terminal device 30 presents the explanatory statement 55 output from the information processing device 20 to a user. The information processing device 20 may output not only the explanatory statement 55 but also the evaluation value 52. That is, the evaluation value 52 may not only be input to the second trained model 12 but also be output directly from the second trained model 12. Then, the terminal device 30 may present the evaluation value 52 and the explanation 55 output from the information processing device 20 to the user.

[0020] In this embodiment, for example, when the AI ​​determines that a certain feature has the greatest influence on the output, an explanation is presented to the user as to the basis on which the AI ​​determined that the certain feature has the greatest influence on the output. Therefore, according to this embodiment, the credibility of the AI ​​output is increased.

[0021] In this embodiment, the first trained model 11 is a model for evaluating used cars, but it may also be a model for evaluating other used durable goods, such as used bicycles, used furniture, used homes, or used equipment. Alternatively, the first trained model 11 may be a model for evaluating durable goods other than used goods, such as custom-made furniture or newly built homes. Alternatively, the first trained model 11 may be a model for evaluating goods other than durable goods, such as land or real estate including homes and land.

[0022] In this embodiment, the evaluation value 52 is a predicted value of price, but it may be a value obtained by evaluation based on an index other than price, such as a predicted value of lifespan. For example, in this embodiment, the evaluation value 52 is a predicted value of the purchase price of a used car, but it may also be a predicted value of the price of real estate, or a predicted value of the time when equipment will fail.

[0023] The punctured sentence 54 may be, for example, input by a user into the terminal device 30 and transmitted from the terminal device 30 to the information processing device 20. Alternatively, the punctured sentence 54 may be input by an administrator into a terminal device for the administrator that is separate from the terminal device 30 and transmitted from the terminal device for the administrator to the information processing device 20. Alternatively, the punctured sentence 54 may be generated by a third trained model that is different from the first trained model 11 and the second trained model 12 and acquired by the information processing device 20. The third trained model may, for example, generate the punctured sentence 54 in response to a prompt input by a user into the terminal device 30, or may generate the punctured sentence 54 in response to a prompt input by an administrator into the terminal device for the administrator.

[0024] The punched sentence 54 includes blank spaces for entering an explanation of the influence of the above-mentioned "at least one feature" on the evaluation value 52. It is desirable that the punched sentence 54 further includes blank spaces for entering the name of the "at least one feature" and blank spaces for entering an explanatory value corresponding to the "at least one feature."

[0025] If "at least one feature" includes two or more features, it is desirable that the perforated sentence 54 include a sentence for each feature included in the "two or more features."

[0026] The "two or more features" desirably include one or more features corresponding to positive explanatory values ​​if the multiple explanatory values ​​53 include positive explanatory values, and desirably include one or more features corresponding to negative explanatory values ​​if the multiple explanatory values ​​53 include negative explanatory values. That is, the "two or more features" desirably include M features corresponding to positive explanatory values ​​and N features corresponding to negative explanatory values, where M and N are natural numbers. For example, among the positive explanatory values, the top M features in descending order of absolute value are selected, and among the negative explanatory values, the top N features in descending order of absolute value are selected. However, if the number of positive explanatory values ​​included in the multiple explanatory values ​​53 is less than M, fewer features than M may be selected as features corresponding to positive explanatory values. If the number of positive explanatory values ​​included in the multiple explanatory values ​​53 is zero, i.e., if the multiple explanatory values ​​53 include only negative explanatory values, no features corresponding to positive explanatory values ​​are selected. If the number of negative explanatory values ​​included in the multiple explanatory values ​​53 is less than N, fewer features than N may be selected as features corresponding to negative explanatory values. If the number of negative explanatory values ​​included in the multiple explanatory values ​​53 is 0, i.e., if the multiple explanatory values ​​53 include only positive explanatory values, no features corresponding to negative explanatory values ​​are selected. M may be any natural number, but is 2 in this embodiment. N may also be any natural number, but is 1 in this embodiment. That is, in this embodiment, the gap sentence 54 includes a sentence for each of the three features, namely, two features corresponding to positive explanatory values ​​and one feature corresponding to a negative explanatory value. When the corresponding feature is defined as feature Fi, each sentence includes a blank space for entering an explanation of the effect that feature Fi had on the evaluation value 52, a blank space for entering the name of feature Fi, and a blank space for entering the explanatory value corresponding to feature Fi.

[0027] The "two or more features" may be selected based on the relationship between the absolute value of the explanatory value and a threshold. For example, when L is a natural number, the top L features with the largest absolute values ​​may be selected from the multiple explanatory values ​​53. L may be any natural number, e.g., 3.

[0028] The configuration of an information processing device 20 according to this embodiment will be described with reference to FIG.

[0029] The information processing device 20 includes a control unit 21, a storage unit 22, and a communication unit 23.

[0030] The control unit 21 includes at least one processor, at least one programmable circuit, at least one dedicated circuit, or any combination thereof. The processor is a general-purpose processor such as a CPU or GPU, or a dedicated processor specialized for specific processing. "CPU" is an abbreviation for central processing unit. "GPU" is an abbreviation for graphics processing unit. An example of the programmable circuit is an FPGA. "FPGA" is an abbreviation for field-programmable gate array. An example of the dedicated circuit is an ASIC. "ASIC" is an abbreviation for application specific integrated circuit. The control unit 21 controls each unit of the information processing device 20 and executes processing related to the operation of the information processing device 20.

[0031] The storage unit 22 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or any combination thereof. The semiconductor memory is, for example, a RAM, a ROM, or a flash memory. "RAM" is an abbreviation for random access memory. "ROM" is an abbreviation for read only memory. RAM is, for example, an SRAM or a DRAM. "SRAM" is an abbreviation for static random access memory. "DRAM" is an abbreviation for dynamic random access memory. ROM is, for example, an EEPROM. "EEPROM" is an abbreviation for electrically erasable programmable read only memory. Flash memory is, for example, an SSD. "SSD" is an abbreviation for solid-state drive. Magnetic memory is, for example, an HDD. "HDD" is an abbreviation for hard disk drive. The storage unit 22 functions, for example, as a main storage device, an auxiliary storage device, or a cache memory. The storage unit 22 stores information used in the operation of the information processing device 20 and information obtained by the operation of the information processing device 20. For example, the first trained model 11, the second trained model 12, and the above-mentioned third trained model may be stored in the storage unit 22. The hole-punched sentence 54 may also be stored in the storage unit 22.

[0032] The communication unit 23 includes at least one communication module. The communication module is, for example, a module compatible with a wired LAN communication standard such as Ethernet (registered trademark) or a wireless LAN communication standard such as IEEE802.11. "IEEE" is an abbreviation for Institute of Electrical and Electronics Engineers. The communication unit 23 communicates with the terminal device 30. The communication unit 23 receives information used in the operation of the information processing device 20 and transmits information obtained by the operation of the information processing device 20. For example, the first trained model 11, the second trained model 12, and the above-mentioned third trained model may be stored in cloud storage accessible via the communication unit 23. The hole-punched sentence 54 may also be stored in cloud storage accessible via the communication unit 23.

[0033] The functions of the information processing device 20 are realized by executing a program according to this embodiment on a processor serving as the control unit 21. That is, the functions of the information processing device 20 are realized by software. The program causes a computer to execute the operations of the information processing device 20, thereby causing the computer to function as the information processing device 20. That is, the computer functions as the information processing device 20 by executing the operations of the information processing device 20 in accordance with the program.

[0034] The program can be stored on a non-transitory computer-readable medium. Examples of non-transitory computer-readable media include flash memory, magnetic recording devices, optical disks, magneto-optical recording media, and ROMs. The program can be distributed by selling, transferring, or lending portable media such as SD cards, DVDs, or CD-ROMs that store the program. "SD" is an abbreviation for Secure Digital. "DVD" is an abbreviation for digital versatile disc. "CD-ROM" is an abbreviation for compact disc read only memory. The program can also be distributed by storing it in the storage of a server and transferring it from the server to another computer. The program can also be provided as a program product.

[0035] A computer temporarily stores a program stored on a portable medium or transferred from a server in its main storage device. The computer then reads the program stored in the main storage device using a processor and executes processing in accordance with the read program. The computer may also read the program directly from a portable medium and execute processing in accordance with the program. The computer may also execute processing in accordance with the received program each time a program is transferred from a server to the computer. Processing may also be executed through a so-called ASP-type service that achieves its functions by issuing execution instructions and obtaining results without transferring the program from the server to the computer. "ASP" is an abbreviation for application service provider. A program is information used for processing by a computer and includes something equivalent to a program. For example, data that is not a direct instruction to a computer but has properties that specify computer processing falls under the category of "something equivalent to a program."

[0036] Some or all of the functions of the information processing device 20 may be realized by a programmable circuit or a dedicated circuit as the control unit 21. In other words, some or all of the functions of the information processing device 20 may be realized by hardware.

[0037] The operation of the information processing device 20 according to this embodiment will be described with reference to Fig. 6. The operation described below corresponds to the method according to this embodiment. That is, the method according to this embodiment includes steps S1 to S3 shown in Fig. 6.

[0038] In S1, the control unit 21 inputs a plurality of variables 51 to the first trained model 11. The plurality of variables 51 are, for example, input by a user to the terminal device 30 and transmitted from the terminal device 30 to the information processing device 20. Alternatively, the plurality of variables 51 may be stored in advance in the storage unit 22.

[0039] As shown in Fig. 2, in this embodiment, the first trained model 11 is a model generated by machine learning, such as a decision tree or a neural network, that identifies the purchase price of a used car. In this embodiment, the multiple variables 51 are information about used cars, and include each feature of the used car as a variable. In the example shown in Fig. 2, eight feature values ​​are input: car model, year, exterior condition, repair history, equipment, breakdown state, interior, and mileage.

[0040] In S2, the control unit 21 inputs the evaluation value 52, the multiple explanation values ​​53, and the punched sentence 54 to the second trained model 12.

[0041] As shown in FIG. 2, in this embodiment, the second trained model 12 is a model such as an LLM that explains the basis for the purchase price of a used car output from the first trained model 11. "LLM" is an abbreviation for large language model. In this embodiment, the evaluation value 52 is a predicted value of the purchase price of a used car output from the first trained model 11 in S1. As an example, the evaluation value 52 is assumed to be "2 million yen."

[0042] As shown in FIG. 3, in this embodiment, the multiple explanatory values ​​53 include, as explanatory values, the SHAP values ​​of each feature input to the first trained model 11 in S1. "SHAP" is an abbreviation for SHapley Additive exPlanations. The multiple explanatory values ​​53 may be obtained using other methods, such as LIME, instead of SHAP. "LIME" is an abbreviation for locally interpretable model-agnostic explanations. In the example shown in FIG. 3, the SHAP value of the vehicle model is "+1 million yen." The SHAP value of the model year is "+400,000 yen." The SHAP value of the exterior condition is "+300,000 yen." The SHAP value of the repair history is "+300,000 yen." The SHAP value of the equipment is "+200,000 yen." The SHAP value of the malfunction condition is "+200,000 yen." The SHAP value of the interior is "+100,000 yen." The SHAP value of the mileage is "-500,000 yen." In this example, the sum of the SHAP values ​​is "2 million yen" which is the same as the evaluation value 52, but since the SHAP values ​​are approximate values, the sum of the SHAP values ​​does not necessarily have to be the same as the evaluation value 52.

[0043] As shown in FIG. 4, in this embodiment, the hole-punched sentences 54 include sentences corresponding to the output of the AI, sentences corresponding to M feature quantities that bring about a positive effect, sentences corresponding to N feature quantities that bring about a negative effect, and other sentences.

[0044] The sentence corresponding to the AI ​​output includes a blank [XXXX] for entering an evaluation value of 52. In the example shown in Figure 4, the sentence corresponding to the AI ​​output is as follows: The purchase price for the used vehicle is [XXXX] million yen.

[0045] When the corresponding feature is defined as feature Fj, the sentences corresponding to M features that bring about a positive effect include a blank space [2-3 lines of explanation of the positive effect] for entering an explanation of the effect that feature Fj has on the evaluation value 52, a blank space [XXXX] for entering the name of feature Fj, and a blank space [XXX] for entering the SHAP value corresponding to feature Fj. In the example shown in FIG. 4, since M=2, there are two sets of sentences corresponding to M features that bring about a positive effect, each of which is as follows: The explanatory variable that has a positive impact on the price is [XXXX]. [2-3 lines of explanation for the positive impact] Due to the explanatory variable: [XXXX], the AI ​​model estimates the price to be worth + [XXX] million yen.

[0046] When the corresponding feature is defined as feature Fk, the sentences corresponding to the N features that bring about a negative effect include a blank space [2-3 lines of explanation of the negative effect] for entering an explanation of the effect that feature Fk has on the evaluation value 52, a blank space [XXXX] for entering the name of feature Fk, and a blank space [XXX] for entering the SHAP value corresponding to feature Fk. In the example shown in FIG. 4, since N=1, there is one set of sentences corresponding to the N features that bring about a negative effect, which are as follows: The explanatory variable that has a negative impact on the price is [XXXX]. [2-3 lines of explanation for the negative impact] Due to the explanatory variable: [XXXX], the AI ​​model values ​​the price at -[XXX] million yen.

[0047] Other sentences include a blank space [comments, suggestions, etc.] for entering an overall evaluation that takes into account all features.

[0048] An instruction such as "The user is not very knowledgeable about machine learning, so explanations should be given in simple terms," ​​a specific example of an explanation, or both may be further input to the second trained model 12. A specific example of an overall evaluation may be further input to the second trained model 12. A definition of the format of the output using symbols, tags, or a markup language may be further input to the second trained model 12.

[0049] In S3, the control unit 21 outputs an explanatory sentence 55. The control unit 21 also simultaneously outputs a sentence corresponding to the output of the AI, which is equivalent to the sentence reporting the evaluation value 52. The control unit 21 also simultaneously outputs other sentences. The control unit 21 transmits the sentence, including the sentence corresponding to the output of the AI, the explanatory sentence 55, and other sentences, to the terminal device 30 via the communication unit 23. The terminal device 30 receives the sentence transmitted from the information processing device 20. The terminal device 30 presents the received sentence to the user. Specifically, the terminal device 30 displays the received sentence on a display. Alternatively, the terminal device 30 may output the received sentence as audio from a speaker.

[0050] As shown in Figure 5, in this embodiment, the sentences corresponding to the output of the AI, explanatory sentences 55, and other sentences are generated in a format in which each blank in the gapped sentence 54 input into the second trained model 12 in S2 is filled in.

[0051] The sentence corresponding to the AI ​​output is the following, in the example shown in Figure 5, with the blank filled in with an evaluation value of 52: The purchase price of the used car is 2 million yen.

[0052] In the example shown in Figure 5, the sentence corresponding to the first feature that brings about a positive effect in the explanatory sentence 55 is as follows after each blank is filled in: The explanatory variable that has a positive effect on price is the car model. The car model "..." is popular among all generations, which is a major attraction. Explanatory variable: Depending on the car model, the AI ​​model estimates a value of +1 million yen in price.

[0053] In the example shown in Figure 5, the sentence corresponding to the second feature that brings about a positive effect in explanatory sentence 55 is as follows, after each blank has been filled in: The explanatory variable that has a positive effect on the price is the model year. The model year is new, and even though it is a used car, it is almost like new, which is also attractive. Explanatory variable: Due to the model year, the AI ​​model estimates a value of +400,000 yen in price.

[0054] In the example shown in FIG. 5, the sentences corresponding to the features that bring about a negative effect in the explanatory sentence 55 are as follows after the blanks are filled in: The explanatory variable that has a negative impact on price is mileage. It is also important to consider that the mileage is over 100,000 km. When the mileage exceeds a certain distance, there is a concern that the vehicle itself or its equipment may deteriorate. Explanatory variable: Due to the mileage, the AI ​​model values ​​the price at -500,000 yen.

[0055] The remaining sentences are filled in with the blanks, resulting in the following in the example shown in Figure 5: Considering the overall appeal of used cars, the prices are attractive. As they are popular models for all generations and are newer models, they are ideal for a variety of uses, such as commuting, going out, or traveling.

[0056] If the user requests a more detailed explanation after explanatory text 55 has been presented to the user, the control unit 21 may generate such a detailed explanation using the second trained model 12 and output it to the user in the same manner as explanatory text 55. For example, if the user asks, "Why do popular used cars regardless of age have higher purchase prices?", the control unit 21 inputs the question into the second trained model 12. If the second trained model 12 determines that a detailed explanation is necessary and outputs a background explanation such as, "The purchase price of a used car depends on the size of the target group wishing to purchase it. Therefore, the fact that it is popular among all age groups has a significant impact on the purchase price," the control unit 21 outputs the background explanation to the user.

[0057] As described above, in this embodiment, the control unit 21 of the information processing device 20 prepares, as the blank sentence 54, a sentence including a blank portion corresponding to the feature name, a blank portion corresponding to the feature SHAP value, and a blank portion for entering an explanation corresponding to the feature name and SHAP value. The control unit 21 of the information processing device 20 creates, as the explanation sentence 55, a sentence in which the feature name, SHAP value, and explanation are entered in each blank portion. Therefore, according to this embodiment, the stability of the AI ​​output can be ensured.

[0058] As a variation of this embodiment, when calculating the SHAP value, instead of calculating the SHAP values ​​of all feature quantities included in the information about used cars, one or several feature quantities may be limited and the SHAP values ​​of the remaining feature quantities may be calculated. For example, if it is known that a specific brand of vehicle has a positive effect, the SHAP values ​​of the remaining feature quantities may be calculated using the average price of used cars of that brand as a starting condition. A possible method for implementing this variation is to modify the SHAP value calculation library used in the first trained model 11 so that limited SHAP values ​​can be calculated, assuming that the first trained model 11 calculates the SHAP values ​​of all feature quantities.

[0059] When a human appraises a used car, it is conceivable that the price is determined to a certain extent based on the model of the car and then adjusted based on the year of manufacture and mileage. In this case, when the order in which the features are viewed is determined, the above-described modified example may be further modified to narrow down the order in which the features are limited when calculating the SHAP value.

[0060] The present disclosure is not limited to the above-described embodiments. For example, two or more blocks shown in the block diagrams may be integrated, or one block may be divided. Two or more steps shown in the flowcharts may be executed in parallel or in a different order, instead of being executed in chronological order as described, depending on the processing capabilities of the device executing each step, or as needed. Other modifications are possible within the scope of the present disclosure. [Explanation of symbols]

[0061] 10 Systems 11 First trained model 12 Second trained model 20 Information processing equipment 21 Control section 22 Memory section 23 Communications Department 30 Terminal Equipment 40 Network 51 variables 52 rating value 53 Explanation Values 54 Hole-punched sentence 55 Description

Claims

1. A first trained model evaluates individual objects by inputting multiple variables that indicate various characteristics of the object to be evaluated, An evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a hole-punched sentence are input into a second trained model that explains each output from the first trained model; outputting an explanatory sentence that explains the influence of at least one feature included in the various features on the evaluation value, the explanatory sentence being generated in a fill-in-the-blank format by the second trained model; An information processing device including a control unit.

2. The information processing device according to claim 1 , wherein, when the at least one feature includes two or more features, the perforated sentence includes a sentence for each feature included in the two or more features.

3. 3. The information processing device according to claim 2, wherein, if the plurality of explanatory values ​​includes a positive explanatory value, the two or more features include one or more features corresponding to the positive explanatory value, and if the plurality of explanatory values ​​includes a negative explanatory value, the two or more features include one or more features corresponding to the negative explanatory value.

4. 2. The information processing device according to claim 1, wherein the perforated sentence includes a blank space for entering an explanation of the effect that the at least one feature has on the evaluation value, a blank space for entering a name of the at least one feature, and a blank space for entering an explanatory value corresponding to the at least one feature.

5. The information processing device according to claim 1 , wherein the evaluation value is a predicted value of price.

6. an information processing device inputting a plurality of variables indicating various characteristics of the object to be evaluated into a first trained model that evaluates each individual object; The information processing device inputs an evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a hole-punched sentence into a second trained model that explains each output from the first trained model; the information processing device outputs an explanatory sentence that is generated in a fill-in-the-blank format by the second trained model and that explains an influence that at least one feature included in the various features has on the evaluation value; A method comprising:

7. Inputting a plurality of variables indicating various characteristics of the object to be evaluated into a first trained model that evaluates each individual object; Inputting an evaluation value output from the first trained model, a plurality of explanatory values ​​indicating the influence of each variable included in the plurality of variables on the evaluation value, and a hole-punched sentence into a second trained model that explains each output from the first trained model; outputting an explanatory sentence that explains the influence of at least one feature included in the various features on the evaluation value, the explanatory sentence being generated in a fill-in-the-blank format by the second trained model; and A program that causes a computer to perform operations including:

Images