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

Abstract

To provide an information processing device, an information processing method, and an information processing program that present an improvement proposal that approaches a request.SOLUTION: An information processing device, which includes a processor that executes a program and a storage device that stores the program, is capable of accessing evaluation information that stores, for each generation method for generating an object to be generated, a group of factors constituting the generation method, evaluation data related to the object to be generated, and a target similarity between target evaluation data and the evaluation data. The processor performs, on an inference model which outputs a similarity indicating how similar the evaluation data is to the target evaluation data when a group of factors that constitute the generation method are input, inference processing for inputting a group of factors constituting an improvement target generation method into the inference model and acquiring an inferred similarity output from the inference model, identification processing for identifying, on the basis of the inferred similarity acquired by the inference processing and the target similarity, a group of factors constituting an identified generation method from the evaluation information, and output processing for outputting a group of factors constituting the identified generation method by the identification processing.SELECTED DRAWING: Figure 1

Description

Technical Field

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

Background Art

[0002] In product development, customer feedback is important. By collecting customer feedback, product developers can understand customer desires and needs for products. Information such as customer desires and needs can be used to improve products, maintain competitiveness, and build customer loyalty. However, asking people for their feedback poses problems in terms of time and cost. Therefore, it is conceivable to use automatically generated feedback on products for product development.

[0003] As background art in the technical field, there is a text generation system of Patent Document 1 below. This text generation system uses items, users, reviews of items by users, and evaluations of items by users as learning data. When a user and an item are input, it generates a review of the item by the user and learns a generation model to generate an evaluation of the item by the user. It also includes a generation unit that inputs the item and the user into the generation model and outputs a review and an evaluation of the item by the user. The generation model includes an aspect layer, a preference layer, a topic component, a gate layer, and a text generation evaluation prediction layer.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] It is difficult to satisfy customers' subjective desires during product development. In Patent Document 1, while it is possible to generate impressions of products, it does not consider the aspect of examining measures for products that are closer to the desires.

[0006] The present invention aims to present improvement proposals that are closer to the desires.

Means for Solving the Problems

[0007] An information processing apparatus according to an aspect of the invention disclosed in the present application is an information processing apparatus having a processor that executes a program and a storage device that stores the program, and is capable of accessing evaluation information that stores, for each generation method for generating a generation target, a factor group that constitutes the generation method, evaluation data regarding the generation target, a target evaluation data, and a target similarity between the target evaluation data and the evaluation data. The processor inputs a factor group that constitutes an improvement target generation method to an inference model that outputs a similarity indicating how similar the evaluation data is to the target evaluation data when the factor group that constitutes the generation method is input, and obtains an estimated similarity output from the inference model; a specifying process of specifying a factor group that constitutes a specific generation method from the evaluation information based on the estimated similarity obtained by the inference process and the target similarity; and an output process of outputting the factor group that constitutes the specific generation method by the specifying process.

[0008] An information processing apparatus, which is another aspect of the invention disclosed in this application, is an information processing apparatus having a processor that executes a program and a storage device that stores the program. For each generation method that generates a generation target, it can access evaluation information that stores a factor group that constitutes the generation method, evaluation data related to the generation target, target evaluation data, and a target similarity between the target evaluation data and the evaluation data. The processor uses the factor group as an explanatory variable group and the target similarity as an objective variable to learn, thereby generating a generation process for generating a regression model that outputs a similarity indicating how similar the evaluation data is to the target evaluation data, and an output process for outputting information related to the regression model generated by the generation process.

Effect of the Invention

[0009] According to a typical embodiment of the present invention, it is possible to present improvement plans closer to requirements. Problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

DETAILED DESCRIPTION OF THE INVENTION

[0011] <FIG. 1 Example of Recipe Improvement> FIG. 1 is an explanatory diagram showing an example of recipe improvement. In FIG. 1, the generation target is curry rice 100, and the generation method will be described by taking the recipe of curry rice 100 (hereinafter, the recipe to be improved 101) as an example. The recipe to be improved 101 stipulates the quantities of cinnamon, garlic, chili pepper, lemon, and onion. Cinnamon and garlic are defined by two values: included (1) or not included (0). Chili pepper, lemon, and onion are each defined as a continuous quantity in the normalized range of 0 to 1.

[0012] The output response sentence 110 is a sentence indicating the reaction to the curry rice 100 cooked according to the recipe to be improved 101, and is data for evaluating the curry rice 100 cooked according to the recipe to be improved 101. The output response sentence 110 may be a sentence in which a person who actually ate the curry rice 100 stated their impression, or it may be a sentence generated by the generation AI.

[0013] When using the generation AI, the smell sensor detects the smell of the curry rice 100 and inputs it to the generation AI. When the detection data of the smell sensor is input to the generation AI, the generation AI generates a descriptive text (hereinafter, taste index) regarding the smell of the curry rice 100 and the taste (for example, spiciness) evoked from the smell.

[0014] Also, when using generative AI, the camera captures the curry rice 100 and inputs the image data into the generative AI. When the image data is input into the generative AI, the generative AI generates a description (hereinafter referred to as a caption) regarding the appearance of the captured curry rice 100 and the deliciousness reminiscent of the appearance.

[0015] In addition, a reaction format is set for the generative AI. The reaction format is information for setting the reaction subject and the reaction method. The reaction subject is information for setting who will react to the generative AI, such as a critic or a 20-year-old woman. The reaction method is information for setting how to react to the generative AI, such as an SNS post or a news flash.

[0016] When the reaction format is set for the generative AI, the generative AI generates a request sentence for a review regarding the curry rice 100 expressed in taste indicators and captions according to the reaction format. Then, the generative AI generates an output reaction sentence 110 indicating the reaction to the review request sentence.

[0017] The desired reaction sentence 120 is a sentence indicating an ideal reaction and is the target evaluation data that would be obtained from the curry rice 100 cooked with the improved recipe 102 that improves the recipe to be improved 101. The desired reaction sentence 120 may be an artificially created sentence or a sentence generated by the generative AI.

[0018] This embodiment presents improvement plans on how to improve the recipe to be improved 101 and what the recipe after improvement (improved recipe) 102 will be in order for the output reaction sentence 110 to approach the desired reaction sentence 120.

[0019] <Example of the hardware configuration of the information processing device in FIG. 2> FIG. 2 is a block diagram showing an example of the hardware configuration of an information processing apparatus. The information processing apparatus 200 includes a processor 201, a storage device 202, an input device 203, an output device 204, and a communication interface (communication IF) 205. The processor 201, the storage device 202, the input device 203, the output device 204, and the communication IF 205 are connected by a bus 206. The processor 201 controls the information processing apparatus 200. The storage device 202 serves as a working area for the processor 201. Also, the storage device 202 is a non-temporary or temporary recording medium that stores various programs and data. Examples of the storage device 202 include a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and a flash memory. The input device 203 inputs data. Examples of the input device 203 include a keyboard, a mouse, a touch panel, a numeric keypad, a scanner, a microphone, and a sensor. The output device 204 outputs data. Examples of the output device 204 include a display, a printer, and a speaker. The communication IF 205 connects to a network and transmits and receives data.

[0020] Also, the storage device 202 stores a recipe table 221 and a review table 222. The recipe table 221 is evaluation information having an entry that defines the quantity of ingredients for each recipe. The review table 222 is evaluation information that defines, as evaluation data, a review regarding a dish cooked based on the recipe in the recipe table 221. The recipe table 221 and the review table 222 exist for each dish. Note that at least one of the recipe table 221 and the review table 222 may be stored in another computer that can communicate with the information processing apparatus 200.

[0021] <FIG. 3 Recipe Table 221> FIG. 3 is an explanatory diagram showing an example of the recipe table 221. The recipe table 221 defines the recipe for the curry rice 100. The recipe table 221 has, as fields, a recipe ID 301 and a factor group 302. The factor group 302 includes, for example, cinnamon 321, garlic 322, chili pepper 323, lemon 324, and onion 325. Cinnamon 321 and garlic 322 are defined with two values of put in (1) and not put in (0). The chili pepper 323, lemon 324, and onion 325 are each defined as a continuous amount in the range of 0 to 1 obtained by normalizing the amounts of chili pepper, lemon, and onion. A combination of the values of the fields in the same row becomes an entry that constitutes one recipe.

[0022] Note that the factor group 302 defined in the recipe table 221 may be some of the ingredients used for cooking the curry rice 100. Also, in FIG. 3, although the recipe is defined by the factor group 302, cooking methods such as heating time and the presence or absence of preliminary preparation of specific ingredients may be included.

[0023] <FIG. 4 Review Table 222> FIG. 4 is an explanatory diagram showing an example of the review table 222. The review table 222 defines, as evaluation data, the reviews for the curry rice 100 cooked according to the recipe of the recipe table 221. The review table 222 has, as fields, a recipe ID 301, a review 400, an improvement similarity 401, and a target similarity 402.

[0024] The review 400 is a sentence showing the content that has been criticized in the past for the curry rice 100 cooked according to the recipe defined by the recipe ID 301. In this example, for the sake of convenience of explanation, one review 400 is defined for one recipe, but there may be a plurality of reviews 400 for one recipe.

[0025] The improvement similarity 401 is the similarity between the review 400 and the output response sentence 110 for the curry rice 100 cooked according to the recipe of the recipe ID 301.

[0026] The target similarity 402 is the similarity between the review 400 and the desired response sentence 120 for the curry rice 100 cooked according to the recipe of the recipe ID 301. In this example, the range of the similarity value is 0.0 to 1.0, and the larger the value, the more similar it indicates. Note that the output response sentence 110, the desired response sentence 120, and the review 400 may be a single sentence or a set of two or more sentences.

[0027] <Functional configuration example of the information processing apparatus 200 in FIG. 5> FIG. 5 is a block diagram showing a functional configuration example of the information processing apparatus 200. The information processing apparatus 200 includes an acquisition unit 501, a calculation unit 502, a setting unit 503, a generation unit 504, an inference unit 505, a specification unit 506, and an output unit 507. Specifically, for example, the acquisition unit 501, the calculation unit 502, the setting unit 503, the generation unit 504, the inference unit 505, the specification unit 506, and the output unit 507 are realized by causing the processor 201 to execute a program stored in the storage device 202 shown in FIG. 2.

[0028] The acquisition unit 501 acquires the output response sentence 110 and the desired response sentence 120. Specifically, for example, the acquisition unit 501 acquires the output response sentence 110 and the desired response sentence 120 by operation input of the input device 203, by reception from the communication IF 205, or by generation by the generation AI. In addition, the acquisition unit 501 acquires the review 400 from the review table 222.

[0029] The calculation unit 502 calculates the improvement similarity 401 and the target similarity 402. The improvement similarity 401 and the target similarity 402 are calculated, for example, by vectorizing the input sentence in a distributed representation. Specifically, for example, a method of decomposing a sentence into words, obtaining a word distributed representation using word2vec, and calculating the similarity by obtaining the average of the word distributed representations as the distributed representation of the sentence may be used, or a method of directly generating the distributed representation of the sentence and calculating the similarity, such as Doc2Vec, may be used. The calculation unit 502 stores the improvement similarity 401 and the target similarity 402 in the review table 222.

[0030] In addition, when there are multiple reviews 400 for one recipe, the calculation unit 502 calculates the improvement similarity 401 and the target similarity 402 for each review 400.

[0031] The setting unit 503 sets the review 400 as an exclusion target for the calculation of the target similarity 402. For the review 400 set as the exclusion target by the setting unit 503, the calculation unit 502 does not calculate the target similarity 402. The review 400 set as the exclusion target is, for example, a review 400 in which the value of the improvement similarity 401 is equal to or less than a threshold value. Also, the review 400 set as the exclusion target may be a review 400 with a relatively low value of the improvement similarity 401. Specifically, for example, when there are 100 reviews 400, the reviews 400 ranked 51st or lower in descending order of the improvement similarity 401 are set as the exclusion targets.

[0032] In addition, when there are multiple reviews 400 for one recipe, the setting unit 503 applies the statistical value of the improvement similarity 401 of each of the multiple reviews 400. The statistical value of the improvement similarity 401 is, for example, the average value, maximum value, minimum value, median value, mode value, or a randomly selected value of the improvement similarity 401 of each of the multiple reviews 400.

[0033] In addition, when the target similarity 402 has been calculated, the setting unit 503 sets the target similarity 402 as not for use by the specifying unit 506.

[0034] The generation unit 504 generates an inference model 500 using the review table 222 and the recipe table 221. The inference model 500 is a model that estimates the second similarity between the review 400 and the desired response sentence 120 for the curry rice 100 cooked with the recipe to be improved 101 when the recipe to be improved 101 is input. The inference model 500 may be a regression model that can be expressed by a regression equation, or may be a neural network, decision tree, random forest, or support vector machine that cannot be expressed by a regression equation.

[0035] The generation unit 504 is configured to be able to generate at least one type of the above-described inference model 500. When the generation unit 504 can generate a plurality of types of inference models 500, the type of the inference model 500 can be set in advance.

[0036] The inference unit 505 inputs the recipe 101 to be improved into the inference model 500 and outputs an estimated value of the second similarity (hereinafter, estimated similarity), which is the inference result from the inference model 500.

[0037] The specifying unit 506 specifies the improvement points of the recipe 101 to be improved based on the estimated similarity. Specifically, for example, the specifying unit 506 specifies a recipe whose target similarity 402 is higher than the estimated similarity, or specifies the difference between the recipe and the recipe 101 to be improved. The difference between the specified recipe and the recipe 101 to be improved is referred to as an improvement plan 510. Note that the target similarity 402 set as not to be used by the setting unit 503 is not used by the specifying unit 506.

[0038] The output unit 507 outputs the improvement plan 510 from the specifying unit 506. Specifically, for example, the output unit 507 displays the improvement plan 510 from the specifying unit 506 on a display, which is an example of the output device 204, outputs it as audio from a speaker, which is an example of the output device 204, prints it out from a printer, which is an example of the output device 204, stores it in the storage device 202, or transmits it to another computer via the communication IF 205.

[0039] <Exclusion setting process for FIGS. 6 and 7> Next, the exclusion setting process will be described. The exclusion setting process is a process for setting the review 400 as an exclusion target for calculating the target similarity 402. Thereby, for the review 400 set as the exclusion target, the target similarity 402 is not calculated, so that the calculation load can be reduced. Whether the exclusion setting process can be executed is set in advance.

[0040] FIG. 6 is a flowchart showing a detailed processing procedure example of the exclusion setting process. FIG. 7 is an explanatory diagram showing an exclusion setting process example.

[0041] (Step S601) The information processing apparatus 200 acquires the output response sentence 110 and the review 400 by the acquisition unit 501.

[0042] (Step S602) The information processing apparatus 200 calculates the improvement similarity 401 by the calculation unit 502. The improvement similarity 401 is calculated for each review 400. The improvement similarity 401 is stored in the review table 222.

[0043] (Step S603) The information processing apparatus 200 sets the review 400 to be excluded by the setting unit 503. Specifically, for example, the setting unit 503 sets the review 400 whose improvement similarity 401 is equal to or less than a threshold value (for example, 0.5) as the exclusion target. In the example of FIG. 7, the review 400 with the recipe ID 301 being "3" is set as the exclusion target.

[0044] <FIGS. 8 to 11 Improvement Plan Presentation Process> Next, the improvement plan presentation process will be described with reference to FIGS. 8 to 11. The improvement plan presentation process is a process of presenting the improvement plan 510 for the improvement target recipe 101. In the improvement plan presentation process, the review table 222 is used, and the review table 222 may or may not have been applied with the exclusion setting process.

[0045] FIG. 8 is a flowchart showing a detailed processing procedure example of the improvement plan presentation process.

[0046] (Step S801) The information processing apparatus 200 acquires the desired response sentence 120 by the acquisition unit 501.

[0047] (Step S802) The information processing apparatus 200 calculates a target similarity 402 by a calculation unit 502. The improvement similarity 401 is calculated for each review 400. The target similarity 402 is stored in a review table 222.

[0048] (Step S803) The information processing apparatus 200 generates an inference model 500 by a generation unit 504. Here, the generation of the inference model 500 will be specifically described with reference to FIG. 9.

[0049] [Example of generating the inference model 500] FIG. 9 is an explanatory diagram showing an example of generating the inference model 500. The generation unit 504 machine-learns the inference model 500 using the factor group 302 (cinnamon 321, garlic 322, pepper 323, lemon 324, onion 325) of the recipe table 221 as an explanatory variable group and the target similarity 402 of the review table 222 as an objective variable.

[0050] For example, when the inference model 500 is a regression model, the inference model 500 is expressed by, for example, the following formula (1).

[0051] y = Ax1 + Bx2 + Cx3 + Dx4 + Ex5 + F ··· (1)

[0052] In the above formula (1), y represents the target similarity 402, x1 represents the quantity of cinnamon 321, x2 represents the quantity of garlic 322, x3 represents the quantity of pepper 323, x4 represents the quantity of lemon 324, and x5 represents the quantity of onion 325. A to E are regression coefficients, and F is an intercept.

[0053] (Step S804) The information processing apparatus 200 obtains an estimated similarity calculated by the inference model 500 by inputting the values of cinnamon 321, garlic 322, pepper 323, lemon 324, and onion 325 of the recipe 101 to be improved into the inference model 500 by an inference unit 505.

[0054] (Step S805) The information processing device 200 specifies, by means of the specifying unit 506, a recipe for which the target similarity 402 is higher than the estimated similarity from the inference model 500. Here, the inference by the inference model 500 and the recipe specification will be specifically described with reference to FIG. 10.

[0055] [Example of Inference by Inference Model 500 and Recipe Specification] FIG. 10 is an explanatory diagram showing an example of inference by the inference model 500 and recipe specification. In FIG. 10, the inference model 500 calculates "0.8" as the estimated similarity for the recipe to be improved 101. The specifying unit 506 specifies, as a recipe for which the target similarity 402 is higher than the estimated similarity "0.8", the recipe with a target similarity 402 of "0.9" and a recipe ID 301 = "7". The specified recipe is referred to as the comparison target recipe.

[0056] (Step S806) The information processing device 200 compares, by means of the specifying unit 506, the comparison target recipe and the recipe to be improved 101, and specifies which factor group 302 is controllable to bring the output response sentence 110 closer to the desired response sentence 120. Here, the specification of the improvement plan 510 will be specifically described with reference to FIG. 11.

[0057] [Example of Specification of Improvement Plan 510] FIG. 11 is an explanatory diagram showing an example of the specification of the improvement plan 510. In FIG. 11, the case where the inference model 500 is a regression model expressed by the following formula (2) will be described as an example. Note that the recipe with a recipe ID 301 = "7" is used as the comparison target recipe 1100.

[0058] y = 0.4x1 + 0.2x2 - 0.1x3 + 0.3x4 + 0.9x5 + 0.7 ··· (2)

[0059] The specific part 506 compares the comparison target recipe 1100 and the improvement target recipe 101. In the comparison target recipe 1100 and the improvement target recipe 101, since the cinnamon 321 has the same value, it is specified as a factor not requiring control. Note that if cinnamon is added in the improvement target recipe 101, that is, if "1" is substituted for x1 in the above formula (2), the estimated similarity y becomes low and moves away from the target similarity 402 (= 0.9) for the comparison target recipe 1100. Therefore, cinnamon 321 can also be said to be a factor whose increase cannot be controlled.

[0060] The garlic 321 is "1" in the comparison target recipe 1100 and "0" in the improvement target recipe 101. By adding garlic to the improvement target recipe 101, that is, substituting "1" for x2 in the above formula (2), the estimated similarity y becomes even higher and approaches the target similarity 402 (= 0.9) for the comparison target recipe 1100. Therefore, garlic 321 is specified as a controllable factor.

[0061] The red pepper 323 is "0.5" in the comparison target recipe 1100 and "0.7" in the improvement target recipe 101. If the amount of red pepper is further increased in the improvement target recipe 101, that is, if a value greater than "0.7" is substituted for x3 in the above formula (2), the estimated similarity y becomes low and moves away from the target similarity 402 (= 0.9) for the comparison target recipe 1100. On the other hand, if the amount of red pepper is decreased in the improvement target recipe 101, that is, if a value smaller than "0.7" is substituted for x3 in the above formula (2), the estimated similarity y becomes high and approaches the target similarity 402 (= 0.9) for the comparison target recipe 1100. Therefore, red pepper 323 is specified as a factor whose increase cannot be controlled and a factor whose decrease can be controlled.

[0062] The lemon 324 is "0.4" in the comparison target recipe 1100 and "0.2" in the improvement target recipe 101. If the lemon is further increased in the improvement target recipe 101, that is, a value greater than "0.2" is substituted for x4 in the above formula (2), the estimated similarity y will increase and approach the target similarity 402 (=0.9) for the comparison target recipe 1100. On the other hand, if the lemon is decreased in the improvement target recipe 101, that is, a value smaller than "0.2" is substituted for x4 in the above formula (2), the estimated similarity y will decrease and move away from the target similarity 402 (=0.9) for the comparison target recipe 1100. Therefore, the lemon 324 is identified as an increase controllable factor and a decrease uncontrollable factor.

[0063] The onion 325 is "0.8" in the comparison target recipe 1100 and "1" in the improvement target recipe 101. The onion in the improvement target recipe 101 cannot be increased further. On the other hand, if the onion is decreased in the improvement target recipe 101, that is, a value smaller than "1" is substituted for x5 in the above formula (2), the estimated similarity y will decrease and move away from the target similarity 402 (=0.9) for the comparison target recipe 1100. Therefore, the onion 325 is identified as an increase uncontrollable factor and a decrease uncontrollable factor.

[0064] In steps S805 and S806, when the inference model 500 is a model that cannot be expressed by a regression formula such as a neural network, decision tree, random forest, or support vector machine, the specifying unit 506 specifies the improvement plan 510 as follows.

[0065] The specifying unit 506 changes the improvement target recipe 101 so that the difference becomes small for at least one of the factor groups 302 whose values differ between the comparison target recipe 1100 and the improvement target recipe 101. The specifying unit 506 generates the changed improvement target recipe 101 by brute force.

[0066] The inference unit 505 inputs the improved recipe 101 after the change into the inference model 500 and recalculates the estimated similarity. If the recalculated estimated similarity is higher than the estimated similarity before the change of the recipe 101 to be improved, the specifying unit 506 specifies at least one of the improved recipe 101 after the change, the factor name whose value has been changed, and the value of the factor after the change as the improvement plan 510.

[0067] (Step S807) The information processing apparatus 200 outputs the improvement plan 510 by the output unit 507. Here, a specific example of the improvement plan 510 will be described.

[0068] The output unit 507 outputs a regression equation as shown in the above formula (2) as the improvement plan 510. Thereby, the user who sees the regression equation can confirm which factor to increase or decrease to increase the estimated similarity y. In this case, the information processing apparatus 200 may not execute steps S805 and S806.

[0069] The output unit 507 outputs the comparison target recipe 1100 as the improvement plan 510. Thereby, the user can confirm the comparison target recipe 1100. In this case, the information processing apparatus 200 may not execute step S806.

[0070] The output unit 507 outputs the comparison target recipe 1100 and the recipe 101 to be improved as the improvement plan 510. Thereby, the user can confirm the comparison target recipe 1100 while comparing it with the recipe 101 to be improved. In this case, the information processing apparatus 200 may not execute step S806.

[0071] The output unit 507 outputs the specification result of the factor group 302 in step S806 as the improvement plan 510. The specification result is information indicating which factor is a non-control factor, an increase non-control factor, an increase control factor, a decrease non-control factor, or a decrease control factor. Thereby, the user can confirm which factor group 302 should be improved.

[0072] Specifically, for example, the output unit 507 outputs, as an improvement plan 510, information indicating that cinnamon 321 is a factor that does not require control. This information may include information indicating the reason why cinnamon 321 is a factor that does not require control (the values are the same between the comparison target recipe 1100 and the recipe to be improved 101).

[0073] Also, the output unit 507 outputs, as a specific result, information indicating that cinnamon 321 is a factor that cannot be controlled for increase. This information may include information indicating the reason why cinnamon 321 is a factor that cannot be controlled for increase (the estimated similarity decreases when cinnamon 321 is set to "1").

[0074] Also, the output unit 507 outputs, as a specific result, information indicating that garlic 322 is a factor that can be controlled. This information may include information indicating the reason why garlic 322 is a factor that can be controlled (the estimated similarity increases when garlic 322 is set to "1").

[0075] Also, the output unit 507 outputs, as a specific result, information indicating that chili pepper 323 is a factor that cannot be controlled for increase. This information may include information indicating the reason why chili pepper 323 is a factor that cannot be controlled for increase (the estimated similarity y decreases when the amount of chili pepper 323 is increased). Also, the output unit 507 outputs, as a specific result, information indicating that chili pepper 323 is a factor that can be controlled for decrease. This information may include information indicating the reason why chili pepper 323 is a factor that can be controlled for decrease (it increases when the amount of chili pepper 323 is decreased). Also, the specific result may include the difference in the quantity of chili pepper 323 between the comparison target recipe 1100 and the recipe to be improved 101.

[0076] In addition, as a specific result, the output unit 507 outputs information indicating that lemon 324 is an increase-controllable factor. This information may include information indicating the reason why lemon 324 is an increase-controllable factor (the estimated similarity y increases when the amount of lemon 324 is increased). In addition, as a specific result, the output unit 507 outputs information indicating that lemon 324 is a decrease-uncontrollable factor. This information may include information indicating the reason why lemon 324 is a decrease-uncontrollable factor (it decreases when the amount of lemon 324 is decreased). Further, the specific result may include the difference in the quantity of lemon 324 between the comparison target recipe 1100 and the improvement target recipe 101.

[0077] In addition, as a specific result, the output unit 507 outputs information indicating that onion 325 is an increase-uncontrollable factor. This information may include information indicating the reason why onion 325 is an increase-uncontrollable factor (onion 325 has reached the upper limit and cannot be increased further). In addition, as a specific result, the output unit 507 outputs information indicating that onion 325 is a decrease-uncontrollable factor. This information may include information indicating the reason why onion 325 is a decrease-uncontrollable factor (the estimated similarity decreases when the amount of onion 325 is decreased). Further, the specific result may include the difference in the quantity of onion 325 between the comparison target recipe 1100 and the improvement target recipe 101.

[0078] In addition, when the inference model 500 is a model that cannot be expressed by a regression formula such as a neural network, a decision tree, a random forest, or a support vector machine, if the recalculated estimated similarity is higher than the estimated similarity before the change of the improvement target recipe 101, the output unit 507 outputs at least one of the changed improvement target recipe 101, the factor name whose value has been changed, and the value of the changed factor as the improvement plan 510.

[0079] <Another configuration example of the review table 222> FIG. 12 is an explanatory diagram showing another configuration example of the review table 222. The review table 1222 has a review factor group 1200 instead of the review 400. The review factor group 1200 includes deliciousness 1201, spiciness 1202, and sourness 1203. The deliciousness 1201, spiciness 1202, and sourness 1203 are an example of the review factor group 1200, and other factors may be included. Each factor (deliciousness 1201, spiciness 1202, sourness 1203) of the review factor group 1200 becomes vector information regarding the review quantified for each recipe.

[0080] In this case, in order to calculate the improvement similarity 401, instead of the output response sentence 110, vector information regarding the output response consisting of deliciousness 1201, spiciness 1202, and sourness 1203 for the curry rice 100 cooked with the recipe to be improved 101 is used. In this case, the improvement similarity 401 is calculated by the distance between vectors.

[0081] Similarly, in order to calculate the target similarity 402, instead of the desired response sentence 120, vector information regarding the desired response consisting of deliciousness 1201, spiciness 1202, and sourness 1203 is used. In this case, the target similarity 402 is also calculated by the distance between vectors.

[0082] As described above, according to this embodiment, in order for the output response data (output response sentence 110 or vector information regarding the output response) to approach the desired response (desired response sentence 120 or vector information regarding the desired response), it is possible to present an improvement plan 510 such as how the recipe to be improved 101 should be improved and what the recipe (improved recipe) 102 after improvement will be.

[0083] Also, in the above-described embodiment, the recipe of the review 400 having a target similarity 402 higher than the estimated similarity y is specified, but the recipe of the review 400 having a target similarity 402 lower than the estimated similarity y may be specified as the comparison target recipe. In this case, the improvement plan 510 is not a plan to improve the recipe to be improved 101, but a plan to improve the comparison target recipe.

[0084] In the above-described embodiments, the generation target was described as curry rice (the improvement target was curry rice 100), and the generation method was described as the recipe for curry rice (the generation method of the improvement target was the improvement target recipe 101). However, the generation target and the improvement target may be dishes other than curry rice as long as their generation methods can be quantified. Further, the generation target and the improvement target may be foods other than dishes (for example, processed foods such as wieners and kamaboko) or beverages (for example, soft drinks and alcoholic beverages). Further, the generation target and the improvement target may be articles other than foods and beverages.

[0085] Further, the information processing apparatus 200 according to the above-described embodiment can also be configured as follows (1) to (16). Note that the names in parentheses are examples.

[0086] (1) An information processing apparatus 200 having a processor 201 that executes a program and a storage device 202 that stores the program can access evaluation information (recipe table 221 and review table 222) that stores, for each generation method (recipe) for generating a generation target (curry rice), a factor group 302 that constitutes the generation method, evaluation data (review 400) related to the generation target, target evaluation data (desired response sentence 120), and a target similarity 402 between the evaluation data (review 400) and the target evaluation data. When the factor group 302 that constitutes the generation method is input, the processor 201 obtains an estimated similarity y output from the inference model 500 as a result of inputting the factor group 302 that constitutes the improvement target generation method (improvement target recipe 101) for the improvement target (curry rice 100) to the inference model 500 that outputs a similarity indicating how similar the evaluation data (review 400) is to the target evaluation data (desired response sentence 120) (step S804). Based on the estimated similarity y obtained by the inference process and the target similarity 402, a specifying process (step S805) for specifying a factor group (comparison target recipe 1100) that constitutes a specific generation method from the evaluation information, and an output process (step S807) for outputting the factor group (comparison target recipe 1100) that constitutes the specific generation method by the specifying process are executed.

[0087] (2) In the above (1), in the specific processing, the processor 201 identifies a specific generation method (comparison target recipe 1100) in which the target similarity 402 is higher than the estimated similarity y obtained by the inference processing from the evaluation information.

[0088] (3) In the above (1), in the output processing, the processor 201 outputs a factor group (comparison target recipe 1100) constituting the specific generation method and a factor group (improvement target recipe 101) constituting the generation method to be improved.

[0089] (4) In the above (1), in the output processing, the processor 201 outputs a comparison result between a factor group (comparison target recipe 1100) constituting the specific generation method and a factor group (improvement target recipe 101) constituting the generation method to be improved.

[0090] (5) In the above (4), in the specific processing, the processor 201 identifies a specific factor having the same value between a factor group (comparison target recipe 1100) constituting the specific generation method and a factor group (improvement target recipe 101) constituting the generation method to be improved as a factor not requiring control, and in the output processing, the processor 201 outputs information regarding the specific factor.

[0091] (6) In the above (4), in the specific processing, the processor 201 identifies a specific factor in which the estimated similarity decreases as the value increases between a factor group (comparison target recipe 1100) constituting the specific generation method and a factor group (improvement target recipe 101) constituting the generation method to be improved as a factor whose increase cannot be controlled, and in the output processing, the processor 201 outputs information regarding the specific factor.

[0092] (7) In the above (4), in the specific process, when a specific factor whose estimated similarity increases as the value increases between the factor group (comparison target recipe 1100) constituting the specific generation method and the factor group (recipe to be improved 101) constituting the generation method to be improved is specified as an increase controllable factor, in the output process, the processor 201 outputs information regarding the specific factor.

[0093] (8) In the above (4), in the specific process, when a specific factor whose estimated similarity y decreases as the value decreases between the factor group (comparison target recipe 1100) constituting the specific generation method and the factor group (recipe to be improved 101) constituting the generation method to be improved is specified as a decrease uncontrollable factor, in the output process, the processor 201 outputs information regarding the specific factor.

[0094] (9) In the above (4), in the specific process, when a specific factor whose estimated similarity y increases as the value decreases between the specific generation method (comparison target recipe 1100) and the generation method to be improved (recipe to be improved 101) is specified as a decrease controllable factor, in the output process, the processor outputs information regarding the specific factor.

[0095] (10) In the above (4), the evaluation information stores, for each generation method (recipe), the factor group 302, the evaluation data (review 400), the improvement similarity 401 between the improvement target evaluation data (output response sentence 110) regarding the improvement target (curry rice 100) and the evaluation data (review 400), and the processor 201 executes a setting process of setting the target similarity 402 as not to be used in the specific process based on the improvement similarity 401.

[0096] (11) In the above (10), in the setting process, the processor 201 sets the target similarity 402 of the generation method whose improvement similarity 401 is equal to or less than a predetermined threshold as not to be used in the specific process.

[0097] (12) In the above (10), in the setting process, when there are a predetermined number of the target similarities 402 higher than the improvement similarity 401, the processor 201 sets the target similarity 402 of the generation method that is equal to or lower than the improvement similarity 401 as not being an object to be used in the specifying process.

[0098] (13) In the above (1), the processor 201 executes a generation process of generating the inference model 500 by using the factor group 302 as an explanatory variable group and the target similarity 402 as an objective variable for learning (step S803).

[0099] (14) In the above (13), the processor 201 executes a first acquisition process of acquiring the target evaluation data (desired response sentence 120) (step S601), and a first calculation process of calculating the target similarity 402 based on the target evaluation data (desired response sentence 120) acquired by the first acquisition process and the evaluation data (review 400), and storing the result in the evaluation information (step S602).

[0100] (15) In the above (14), the processor 201 executes a second acquisition process of acquiring improvement target evaluation data (output response sentence 110) regarding the improvement target (curry rice 100) (step S601), a second calculation process of calculating the improvement similarity (401) between the improvement target evaluation data (output response sentence 110) acquired by the second acquisition process and the evaluation data (review 400), and storing the result in the evaluation information (step S602), and a setting process of setting the evaluation data (review 400) as not being an object for calculating the target similarity 402 by the first calculation process based on the improvement similarity 401 calculated by the second calculation process (step S603).

[0101] An information processing apparatus 200 having a processor 201 that executes a program and a storage device 202 that stores the program can access evaluation information (recipe table 221 and review table 222) that stores, for each generation method (recipe) for generating a generation target (curry rice), a factor group 302 that constitutes the generation method, evaluation data (review 400) regarding the generation target, target evaluation data (desired response sentence 120), and a target similarity 402 between the evaluation data (review 400) and the target evaluation data (desired response sentence 120). The processor 201 generates a regression model that outputs a similarity indicating how similar the evaluation data (review 400) is to the target evaluation data (desired response sentence 120) by learning using the factor group 302 as an explanatory variable group and the target similarity 402 as an objective variable (step S803), and executes an output process of outputting information regarding the regression model generated by the generation process (step S807).

[0102] Note that the present invention is not limited to the above-described embodiments, and various modifications and equivalent configurations within the scope of the appended claims are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the configurations described. Also, a part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Further, the configuration of another embodiment may be added to the configuration of one embodiment. Also, for a part of the configuration of each embodiment, addition, deletion, or replacement with another configuration may be made.

[0103] In addition, some or all of the above-described configurations, functions, processing units, processing means, etc. may be realized in hardware by, for example, designing them in an integrated circuit, or may be realized in software by a processor interpreting and executing a program that realizes each function.

[0104] Information such as programs, tables, and files that implement each function can be stored in a storage device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC (Integrated Circuit) card, an SD card, or a DVD (Digital Versatile Disc).

[0105] In addition, the control lines and information lines show those considered necessary for explanation, and do not necessarily show all the control lines and information lines required for implementation. In practice, it can be considered that almost all components are interconnected.

Explanation of Signs

[0106] 101 Recipe to be improved 110 Output response sentence 120 Desired response sentence 200 Information processing apparatus 201 Processor 202 Storage device 221 Recipe table 222 Review table 302 Factor group 400 Review 401 Improvement similarity 402 Target similarity 500 Inference model 501 Acquisition unit 502 Calculation unit 503 Setting unit 504 Generation unit 505 Inference unit 506 Identification unit 507 Output unit 510 Improvement plan 601 Acquisition unit 1100 Recipe for comparison y Estimated similarity

Claims

1. An information processing apparatus having a processor that executes a program and a storage device that stores the program, is capable of accessing evaluation information that stores, for each generation method for generating a generation target, a factor group that constitutes the generation method, evaluation data related to the generation target, target evaluation data, and a target similarity between the target evaluation data and the evaluation data, wherein the processor performs an inference process of obtaining an estimated similarity output from the inference model by inputting a factor group that constitutes an improvement target generation method, which is a method for generating an improvement target, into an inference model that outputs a similarity indicating how similar the evaluation data is to the target evaluation data when a factor group that constitutes the generation method is input, a specifying process of specifying a factor group that constitutes a specific generation method from the evaluation information based on the estimated similarity obtained by the inference process and the target similarity, and an output process of outputting a factor group that constitutes the specific generation method obtained by the specifying process, and is characterized by executing the above.

2. The information processing apparatus according to Claim 1, wherein in the specifying process, the processor specifies, from the evaluation information, a factor group that constitutes a generation method having a target similarity higher than the estimated similarity obtained by the inference process as the factor group that constitutes the specific generation method. and is characterized by the above.

3. The information processing apparatus according to Claim 1, wherein in the output process, the processor outputs a factor group that constitutes the specific generation method and a factor group that constitutes the improvement target generation method. and is characterized by the above.

4. The information processing apparatus according to Claim 1, wherein in the output process, the processor outputs a comparison result between a factor group that constitutes the specific generation method and a factor group that constitutes the improvement target generation method. and is characterized by the above.

5. The information processing apparatus according to Claim 4, wherein in the specifying process, the processor specifies a specific factor having the same value between a factor group that constitutes the specific generation method and a factor group that constitutes the improvement target generation method as a factor not requiring control, and in the output process, the processor outputs information related to the specific factor. and is characterized by the above.

6. The information processing apparatus according to Claim 4, In the specific process, the processor identifies a specific factor whose estimated similarity decreases as the value increases between the specific generation method and the generation method to be improved as a factor that cannot be controlled for increase, In the output process, the processor outputs information regarding the specific factor. An information processing apparatus characterized by the above. **Claim 7** An information processing apparatus according to claim 4, In the specific process, the processor identifies a specific factor whose estimated similarity increases as the value increases between the group of factors constituting the specific generation method and the group of factors constituting the generation method to be improved as a factor that can be controlled for increase, In the output process, the processor outputs information regarding the specific factor. An information processing apparatus characterized by the above. **Claim 8** An information processing apparatus according to claim 4, In the specific process, the processor identifies a specific factor whose estimated similarity decreases as the value decreases between the group of factors constituting the specific generation method and the group of factors constituting the generation method to be improved as a factor that cannot be controlled for decrease, In the output process, the processor outputs information regarding the specific factor. An information processing apparatus characterized by the above. **Claim 9** An information processing apparatus according to claim 4, In the specific process, the processor identifies a specific factor whose estimated similarity increases as the value decreases between the group of factors constituting the specific generation method and the group of factors constituting the generation method to be improved as a factor that can be controlled for decrease, In the output process, the processor outputs information regarding the specific factor. An information processing apparatus characterized by the above. **Claim 10** An information processing apparatus according to claim 1, The evaluation information stores, for each generation method, the group of factors, the evaluation data, the improvement similarity between the improvement target evaluation data regarding the improvement target and the evaluation data, The processor, A setting process of setting the target similarity to be excluded from use in the specific process based on the improvement similarity, An information processing apparatus characterized by executing the above. **Claim 11** An information processing apparatus according to claim 10, In the setting process, the processor sets the target similarity of a generation method whose improvement similarity is equal to or less than a predetermined threshold value to be excluded from use in the specific process. An information processing apparatus characterized by the above. **Claim 12** An information processing apparatus according to claim 10, In the setting process, when there are a predetermined number of the target similarities higher than the improvement similarity, the processor sets the target similarity of the generation method that is equal to or lower than the improvement similarity as not being for use in the specific process. An information processing apparatus characterized by the above.

13. An information processing apparatus according to claim 1, wherein the processor generates the inference model by learning with a group of factors constituting the generation method as an explanatory variable group and the target similarity as an objective variable; and An information processing apparatus characterized by executing the above.

14. An information processing method executed by an information processing apparatus having a processor that executes a program and a storage device that stores the program, wherein for each generation method for generating a generation target, it is possible to access evaluation information that stores a group of factors constituting the generation method, evaluation data regarding the generation target, target evaluation data, and a target similarity between the evaluation data and the target evaluation data; wherein the processor performs an inference process of obtaining an estimated similarity output from the inference model as a result of inputting a group of factors constituting an improvement target generation method to the inference model that outputs a similarity indicating how similar the evaluation data is to the target evaluation data when the group of factors constituting the generation method is input; a specific process of specifying a group of factors constituting a specific generation method from the evaluation information based on the estimated similarity obtained by the inference process and the target similarity; an output process of outputting a group of factors constituting the specific generation method by the specific process; An information processing method characterized by executing the above.

15. In a processor capable of accessing evaluation information that stores a group of factors constituting the generation method, evaluation data regarding the generation target, target evaluation data, and a target similarity between the evaluation data and the target evaluation data for each generation method for generating a generation target, when a group of factors constituting the generation method is input, an inference process of obtaining an estimated similarity output from the inference model as a result of inputting a group of factors constituting an improvement target generation method to the inference model that outputs a similarity indicating how similar the evaluation data is to the target evaluation data; a specific process of specifying a group of factors constituting a specific generation method from the evaluation information based on the estimated similarity obtained by the inference process and the target similarity; an output process of outputting a group of factors constituting the specific generation method by the specific process; An information processing program characterized by causing [operation] to be executed.

Images