Information processing device, information processing method, and program

Abstract

An information processing device 100 according to the present disclosure comprises: an input unit 121 that receives an input case composed of a machine-learning model including a plurality of identities, and a condition that should be satisfied by a prediction output from a prediction model by inputting the input case to the prediction model; a generation unit 122 that generates amendments for one or more identities of the input case so that the prediction, which is output for the input case by the prediction model, satisfies the condition; and a selection unit that selects, from among reference cases including the plurality of identities on the basis of the amendments, a reference case that satisfies the condition. Thus, the information processing device 100 can assist in the decision making of a user.

Description

Information processing device, information processing method, and program

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

[0002] Predictions for input data are made using machine learning models in various fields. In this context, it is also common to generate revision proposals to change predictions made by machine learning models. For example, Patent Literature 1 describes generating revision proposals for input data that will result in a desired prediction.

[0003] International Publication No. 2022 / 034685

[0004] However, in the technology of Patent Document 1, the reliability of the generated revision proposal for the case is unknown, which causes a problem that even if a revision proposal for the case is generated, it is difficult to determine whether or not to adopt the revision proposal.

[0005] Therefore, an object of the present disclosure is to solve the above-mentioned problem that even when a modification proposal for a case is generated, it is difficult to determine whether or not to adopt the modification proposal.

[0006] An information processing device according to one aspect of the present disclosure includes: an input unit that receives input examples including multiple features and a condition that a prediction output from a predictive model when the input examples are input to the predictive model must satisfy; a generation unit that generates proposed modifications to one or more features of the input examples so that the prediction output from the predictive model for the input examples satisfies the condition; and a selection unit that selects, from reference examples including the multiple features, the reference example that satisfies the condition based on the proposed modifications. Also, an information processing method according to one aspect of the present disclosure includes: receiving input examples including multiple features and a condition that a prediction output from the predictive model when the input examples are input to the predictive model must satisfy; generating proposed modifications to one or more features of the input examples so that the prediction output from the predictive model for the input examples satisfies the condition; and selecting, from reference examples including the multiple features, the reference example that satisfies the condition based on the proposed modifications. Furthermore, a program according to one embodiment of the present disclosure is configured to cause a computer to execute the following processes: receive an input example including a plurality of features and a condition that must be satisfied by a prediction output from a prediction model when the input example is input to the prediction model; generate suggested modifications to one or more of the features of the input example so that the prediction output from the prediction model for the input example satisfies the condition; and select, from reference cases including a plurality of the features, the reference case that satisfies the condition based on the suggested modifications.

[0007] With the present disclosure configured as described above, when a proposed modification for a case is generated, it becomes easy to determine whether or not to adopt the proposed modification.

[0008] FIG. 1 is a block diagram showing a configuration of an information processing device according to the present disclosure. FIG. 2 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 3 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 4 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 5 is a flowchart showing processing operations of an information processing device according to the present disclosure. FIG. 6 is a diagram illustrating a usage example of an information processing device according to the present disclosure. FIG. 7 is a diagram illustrating a usage example of an information processing device according to the present disclosure. FIG. 8 is a block diagram showing a hardware configuration of an information processing device according to the present disclosure. FIG. 9 is a block diagram showing a configuration of an information processing device according to the present disclosure.

[0009] First Embodiment The present disclosure will be described with reference to the drawings, which may be relevant to any of the embodiments.

[0010] [Configuration] The information processing device 10 of this embodiment is used to make predictions on input data using a prediction model. In particular, this embodiment makes predictions on cases containing multiple features as input data, generates feature revision proposals to improve the predictions, and outputs reference cases corresponding to the revision proposals. As an example, this embodiment describes a case in which multiple features (attributes, characteristics) of a person are used as examples, and predicts the probability of a loan, such as a mortgage, being granted for the case using a prediction model. This embodiment then outputs feature revision proposals that improve the predicted loan probability so that it satisfies preset conditions, and also outputs reference cases of other people that may qualify for the revision proposals. In this way, this embodiment can obtain reference cases that may qualify for the revision proposals in addition to revision proposals that improve the predictions for the person being predicted. As a result, the reliability of the revision proposals is increased, making it easier for the person to decide whether to adopt the revision proposals. The information processing device 10 is not limited to making predictions regarding the probability of a loan being granted to a person, but can also be applied to various predictions for various cases.

[0011] The information processing device 10 is composed of one or more information processing devices each including a calculation device and a storage device. As shown in FIG. 1 , the information processing device 10 includes an input unit 11, an amendment generation unit 12, a reference case selection unit 13, and an output unit 14. The functions of the input unit 11, amendment generation unit 12, reference case selection unit 13, and output unit 14 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The information processing device 10 also includes a model storage unit 16 and a reference case storage unit 17. The model storage unit 16 and the reference case storage unit 17 are each composed of a storage device. Each component will be described in detail below.

[0012] The model storage unit 16 stores a prediction model, which is a machine learning model that outputs prediction data in response to input case data. The prediction model is generated by machine learning using training data consisting of pre-prepared case data and result data. The case data consists of multiple pieces of background data about a person, and each piece of background data consists of the person's attributes and characteristics, such as gender, age, occupation, annual income, debt, hobbies, and hometown, as shown in Figures 2 and 3. The result data is data representing the results of a loan, such as whether or not the loan has been granted, or the probability of the loan being granted. By performing machine learning using such training data, the prediction model is generated so as to output prediction data indicating whether or not a loan will be granted and the probability of the loan being granted from the case data consisting of multiple pieces of background data about a person. Note that the prediction model in this embodiment is configured to output the probability of loan granting as prediction data, as an example.

[0013] The reference case storage unit 17 stores reference case data, which are cases in which predictions were made using the above-mentioned prediction model and loans were actually made. The reference case data is data consisting of multiple characteristics of a person, and each characteristic consists of the person's attributes and characteristics, such as gender, age, occupation, annual income, debts, hobbies, and hometown, as described above. The reference case data also includes prediction data that represents the prediction results of the prediction model for the reference case data. The prediction data includes information such as whether or not a loan will be made and the probability of the loan being made. In this embodiment, the prediction data includes the probability of the loan being made. Note that the reference case data is not necessarily limited to cases in which loans were actually made, and may also be cases in which predictions were simply made in the past.

[0014] The input unit 11 accepts input of input example data to be predicted. In this embodiment, the input example data is data consisting of multiple features of a person to be predicted for loans. Each feature consists of a person's attributes and characteristics, such as gender, age, occupation, annual income, debt, hobbies, and hometown, as described above. An example of input example data is shown as "input example" denoted by reference character D1 in FIG. 2 . In this example, the features of the "input example data" are gender "male," age "35 years old," occupation "part-time work," annual income "3 million yen," and debt "1 million yen." Note that the features of the "input example data" may include hobby "fishing" and hometown "Fukuoka," as shown by reference character D11 in FIG. 3 . Furthermore, the feature values ​​may be categorical data grouped together, such as gender "male, female" and occupation "part-time, full-time employee, ...," or may be numerical data expressed as continuous values, such as age "35 years old" and annual income "3 million yen."

[0015] Note that an importance level is set in advance for each feature of the input case data. Here, the importance level is a value corresponding to the contribution to the prediction by the prediction model. For example, a feature that has a strong influence on the prediction of the loan probability is set to a higher importance level. In this case, the importance level can be set using an existing method for evaluating the contribution of explanatory variables in a machine learning model. For example, a method such as permutation feature importance can be used to calculate importance that is independent of the content of the case, and a method such as LIME (Local Interpretable Model-agnostic Explanations) or SHAP (Shapley Additive Explanations) can be used to calculate importance that is dependent on the case. Note that the importance level may be set numerically for each feature, or may be set in stages, such as high or low, as shown in FIG. 3 .

[0016] The input unit 11 also accepts input of conditions that must be satisfied by the prediction made by the prediction model for the input case data. In this embodiment, the prediction made by the prediction model outputs the probability that a loan will be granted for the input case data, and the value of this loan probability is input as a condition. The conditional loan probability value is a value desired by the person being predicted or a set value that is generally considered high, such as 60%. The input condition may be any value, and when the prediction model predicts whether a loan will be granted, "loan is granted" may be used as the condition.

[0017] The revision proposal generation unit 12 (generation unit) first performs a prediction using the prediction model for the input case data described above. For example, as shown by reference symbol D1 in FIG. 2 , the revision proposal generation unit 12 obtains a loan probability of "15%," which is the prediction output when the input case data is input to the prediction model. Then, the revision proposal generation unit 12 generates revision proposal data including one or more feature data values ​​of the input case data so that the obtained predicted loan probability satisfies the condition entered as described above. For example, the revision proposal generation unit 12 searches for revision proposal data by changing the value of one or more feature data included in the input case data so that the loan probability of "15%," which is the prediction by the prediction model for the input case data, satisfies the condition loan probability of "60%." In this case, as shown by reference symbol D2 in FIG. 2 , suppose that when the feature values ​​of the input case data are corrected to "occupation" "full-time employee" and "annual income" "3.5 million yen or more," a loan probability of "65%" that satisfies the condition is obtained. Then, the revision proposal generation unit 12 generates revision proposal data consisting only of the feature data corrected to satisfy the condition. For example, as shown by reference numeral D3 in FIG. 2 and reference numeral D12 in FIG. 3, revision suggestion data consisting of two feature data items, namely, occupation "full-time employee" and annual income "3.5 million yen or more", is generated.

[0018] The generation of revision proposal data by the revision proposal generator 12 can be realized by, for example, the method described in Patent Literature 1 or various other known methods. For example, the revision proposal generator 12 modifies one or more feature data values ​​of the input case data and inputs them into the prediction model, checks whether the loan probability, which is the output prediction, satisfies a condition, and generates the modified feature data values ​​as revision proposal data if the predicted loan probability satisfies the condition. As another example, the revision proposal generator 12 may generate a revision proposal using the training data used to generate the prediction model. In this case, the revision proposal generator 12 may extract case data from the training data whose loan probability satisfies a condition, and use differences between the features of the case data and the input case data as revision proposal data. As described above, since importance is assigned to each feature of the input case data, the revision proposal generator 12 may generate revision proposal data based on the importance. For example, the revision proposal generator 12 may select feature data with a higher importance than others and modify the value of the selected feature data to generate revision proposal data such that the prediction satisfies a condition.

[0019] The reference case selection unit 13 (selection unit) selects reference case data stored in the reference case storage unit 17 based on the revision proposal data generated as described above. At this time, the reference case selection unit 13 selects reference case data that satisfies the above-described conditions and satisfies the values ​​of each feature data included in the revision proposal data. In the above example, reference case data that satisfies the condition of a loan probability of "60% or more" and the feature data of the revision proposal data, such as an occupation of "full-time employee" and an annual income of 3.5 million yen or more, is selected. As an example, in the example of FIG. 2, as shown by symbol D4, reference case data with a loan probability of "70%" and feature data of an occupation of "full-time employee" and an annual income of 3.6 million yen may be selected.

[0020] The reference case selection unit 13 also selects reference case data based on the input case data. For example, the reference case selection unit 13 calculates a similarity indicating the degree of similarity between the input case data and the reference case data according to a preset criterion, and selects reference case data based on this similarity. At this time, the reference case selection unit 13 particularly calculates the similarity between other feature data between the input case data and the reference case data that is different from the feature data included in the revision suggestion data, and selects reference case data that is highly similar. Furthermore, at this time, the reference case selection unit 13 calculates the similarity using feature data between the input case data and the reference case data that has a set importance higher than the others.

[0021] A method for selecting reference case data will now be described with reference to FIGS. 2 to 5. First, as described above, it is assumed that proposed revision data shown as reference numeral D12 in FIG. 3 is generated for input case data shown as reference numeral D11 in FIG. 3. Furthermore, it is assumed that importance is set for each feature data item in the input case data, as shown as reference numeral D11 in FIG. 3. In this case, the reference case selection unit 13 generates first feature group data shown as reference numeral D13 in FIG. 3, which is composed of feature data items "gender (male)," "age (35 years old)," and "liabilities (1 million yen)." This excludes "occupation" and "annual income," which are feature data items included in the proposed revision data, from the feature data items in the input case data, and further excludes "hobbies" and "place of birth," which are feature data items with lower importance than the other data items. The reference case selection unit 13 then calculates the similarity between the first feature group data and second feature group data, which is composed of feature data items in the reference case data that correspond to the first feature group data.

[0022] At this time, the reference case selection unit 13 performs data conversion on the first feature group data and the second feature group data so that the similarity between the first feature group data and the second feature group data, which are composed of a plurality of feature data, can be calculated. As an example, data conversion of the first element group data generated from the input example data will be described. For the categorical feature "gender" of the first element group data shown by reference numeral D13 in FIG. 3 , one-hot encoding is performed, and data conversion is performed as shown by reference numeral D14 in FIG. 3 , such that "gender: male = 1" and "gender: female = 0." Furthermore, for the continuous features "age" and "debt" of the first element group data shown by reference numeral D13 in FIG. 3 , standardization is performed, and data conversion is performed as shown by reference numeral D15 in FIG. 3 , such that "age = -0.14" and "debt = 0.56." At this time, standardization is performed, for example, using the average and standard deviation of all the reference case data. For example, age standardization is performed by calculating "(age of input example - μ) / σ" using the average age μ of all reference case data and its standard deviation σ. Note that the above-mentioned data conversion method is an example, and data conversion may be performed by any method.

[0023] The reference case selection unit 13 then calculates the similarity between the first feature group data of the input case data, which has been data converted as described above, and the second feature group data of each reference case data, and selects reference case data based on this similarity. For example, as shown in FIG. 4, the first feature group data of the input case data and the second feature group data of the reference case data can each be represented as a four-dimensional vector through data conversion, making it possible to calculate the similarity between such four-dimensional vectors. As an example, the similarity may be calculated using the distance between vectors (e.g., Euclidean distance or Manhattan distance), and reference case data whose distance is less than a certain value may be selected. Alternatively, the cosine similarity between vectors may be calculated as shown in Equation 1, and reference case data whose value is close to 1 may be selected. Here, <*, *> represents the Euclidean inner product, and ∥*∥ represents the Euclidean norm. Cosine similarity takes values ​​from -1 to 1, and when cosine similarity = 1, x' becomes a vector multiplied by a constant (x = kx'), meaning that the two vectors are similar.

[0024] Furthermore, the reference case selection unit 13 may calculate the similarity between the input case data and the reference case data according to the importance set for the feature data as described above. In this case, the similarity between the input case data including all feature data and the reference case data may be calculated. As an example, the similarity between the input case data including all feature data and the reference case data may be calculated by using the following formula (2): ｉｉ The similarity between the input example data x and the reference example data x′ may be calculated using a matrix incorporating the importance of each feature data indicated by the symbol x′. Here, w ｉｉ is the feature x ｉ For example, by setting the value to 1 if the importance is equal to or greater than a threshold, and 0 otherwise, only feature data that is equal to or greater than the threshold can be considered. ｉｉ is not limited to 0 or 1, but can be any value according to the importance.

[0025] In the above case, the norm and distance can be defined by the following equation 3, and the similarity can be evaluated using the distance and cosine similarity described above.

[0026] As described above, the reference case selection unit 13 selects reference case data that satisfies the conditions and the feature data values ​​of the proposed revision, and further selects reference case data that has a high similarity to the input case data according to a preset standard. At this time, the reference case selection unit 13 may select one reference case data with the highest similarity, may select multiple reference case data in descending order of similarity, or may probabilistically or randomly select reference case data with similarities equal to or greater than a threshold. The reference case selection unit 13 may also select reference case data that satisfies the conditions and the proposed revision, without necessarily using the similarity described above. Furthermore, at this time, the reference case selection unit 13 is not limited to selecting reference case data that satisfy all feature data values ​​included in the proposed revision data, and may select reference case data that satisfy some feature data values ​​included in the proposed revision data.

[0027] The output unit 14 outputs the above-mentioned input case data and the loan probability predicted by the prediction model for the input case data to a display unit of an information processing terminal operated by the person to be predicted, so as to display them. At this time, the output unit 14 also outputs amendment proposal data generated to satisfy the conditions for the input case data and the reference case data selected as described above, so as to display them together with the input case data. As an example, FIG. 5 shows an example of display of the input case data, amendment proposal data, and reference case data by the output unit 14. Note that the output unit 14 does not necessarily have to display these data simultaneously, and may display one or more of them sequentially. Furthermore, the output unit 14 does not necessarily have to display the amendment proposal data, and may display the input case data and the reference case data.

[0028] [Operation] Next, a description will be given of the operation of the above-described information processing device 10. First, it is assumed that the above-described prediction model and reference case data are stored in the information processing device 10.

[0029] The information processing device 10 receives input of input case data including multiple features that are examples of the person to be predicted, and conditions that must be satisfied by the prediction of the input case data (step S1 in FIG. 6 ). The information processing device 10 then inputs the input case data into a prediction model and outputs a loan probability that is a prediction for the input case data. The information processing device 10 then generates revision proposal data by changing the values ​​of one or more feature data of the input case data so that the loan probability predicted by the prediction model satisfies the input conditions (step S2 in FIG. 6 ). For example, for input case data such as that shown by reference numeral D1 in FIG. 2 or reference numeral D11 in FIG. 3 , the information processing device 10 generates revision proposal data such as that shown by reference numeral D3 in FIG. 2 or reference numeral S12 in FIG. 3 .

[0030] Next, the information processing device 10 selects reference case data that satisfy the conditions and the revision proposal data, and further selects, from this, reference case data that is similar to the input case data (step S3 in FIG. 6 ). At this time, the information processing device 10 calculates the similarity between the input case data and the reference case data using a feature data group that excludes the feature data of the revision proposal data and feature data with low importance from the feature data of the input case data. As an example, as shown by reference numeral D13 in FIG. 3 , the information processing device 10 generates first feature group data that excludes the feature data of the revision proposal data and feature data with low importance from the feature data of the input case data. Then, the information processing device 10 calculates the similarity between the first feature group data and second feature group data that includes feature data groups in the reference case data that correspond to the first feature group data. At this time, as shown by reference numeral D15 in FIG. 3 and in FIG. 4 , the information processing device 10 compares the feature group data of the input case data and the reference case data by standardizing them, and then calculates the similarity. Then, the information processing device 10 selects, from the reference case data that satisfy the conditions and the revision proposal data, reference case data that satisfies a preset selection criterion, such as having the highest similarity.

[0031] Thereafter, the information processing device 10 outputs the input case data, the generated revision suggestion data, and the selected reference case data to a display unit of an information processing terminal operated by the person to be predicted. For example, as shown in FIG. 5, these three data may be output to be displayed together.

[0032] As described above, in this embodiment, when a prediction is made on input example data using a prediction model, a feature revision proposal is generated that improves the prediction so that it satisfies preset conditions, and other reference examples that may correspond to the revision proposal are also output. This makes it possible to obtain, for the person being predicted, reference examples that may correspond to the revision proposal in addition to a revision proposal that improves the prediction. As a result, the reliability of the revision proposal is increased, making it easier to determine whether to adopt the revision proposal.

[0033] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, another example of use of the information processing device described in the above embodiment will be described.

[0034] The information processing device 10 of this embodiment can be used in the field of healthcare. Specifically, as shown in FIG. 7 , the information processing device 10 receives input of attribute data (e.g., gender, age) and vital data (e.g., weight, number of steps) of a subject U via a network, and predicts values ​​representing the subject U's health status, such as a health index. In this case, the attribute data and vital data of the subject U are input to the information processing device 10 as input example data consisting of the above-mentioned multiple feature data, and further, conditions that are target values ​​for the health index are input. The information processing device 10 also stores a prediction model that outputs a prediction of a health index from the input example data, and further stores reference example data that is an example of another person.

[0035] The information processing device 10 then inputs the input example data as shown in Fig. 8 into the prediction model to output a predicted value of the health index, and generates a revision proposal for feature data to be improved in the input example data so that the predicted value satisfies the condition. In the example of Fig. 8, a revision proposal is generated in which the value of "step count" is revised among the feature data of the input example data.

[0036] The information processing device 10 then selects reference case data that satisfies the conditions and the revision proposal data. At this time, it may further select reference case data that is similar to the input case data from the reference case data that satisfies the conditions and the revision proposal data. The information processing device 10 then outputs the selected reference case data for display. At this time, it may also display a prediction for the reference case data. Furthermore, as shown in FIG. 8 , the information processing device 10 may output the input case data and its prediction, the revision proposal and the prediction when the revision proposal is applied, and the prediction for the reference case data for display.

[0037] In this way, the subject U can obtain a health condition prediction based on his or her own data, as well as a corresponding revision suggestion and reference case data of others. As a result, the subject U's confidence in the revision suggestion increases, making it easier to decide whether or not to adopt the revision suggestion. In this way, providing reference case data corresponding to the revision suggestion can support the user's decision-making.

[0038] Third Embodiment Next, a third embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, an outline of the configuration of the information processing device described in the above embodiments is shown.

[0039] 9, the information processing device 100 of this embodiment is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; a storage device 105 storing the programs 104; a drive device 106 for reading and writing data from and to a storage medium 110 external to the information processing device; a communication interface 107 for connecting to a communication network 111 external to the information processing device; an input / output interface 108 for inputting and outputting data; and a bus 109 for connecting the various components.

[0040] 9 shows an example of the hardware configuration of the information processing device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with only a part of the above-described configuration, such as excluding the drive device 106. Furthermore, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof, instead of the above-described CPU.

[0041] The information processing device 100 can be equipped with the input unit 121, generation unit 122, and selection unit 123 shown in FIG. 10 by having the CPU 101 acquire and execute the program group 104. The program group 104 may be stored in advance in the storage device 105 or the ROM 102, for example, and loaded into the RAM 103 by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, with the drive device 106 reading out the program and supplying it to the CPU 101. However, the input unit 121, generation unit 122, and selection unit 123 described above may be constructed using dedicated electronic circuits for realizing such means.

[0042] The input unit 121 receives input examples including multiple features and conditions that must be satisfied by a prediction output from a prediction model when the input examples are input to the prediction model. The generation unit 122 generates suggested modifications to one or more of the features of the input examples so that the prediction output from the prediction model for the input examples satisfies the conditions. The selection unit 123 selects, from reference examples including multiple features, the reference examples that satisfy the conditions based on the suggested modifications.

[0043] With the above-described configuration, the present disclosure can obtain reference case data along with suggested corrections for input cases, thereby increasing the reliability of the suggested corrections and facilitating the decision on whether to adopt the suggested corrections.

[0044] In addition, at least one or more of the functions of the above-mentioned input unit 121, generation unit 122, and selection unit 123 may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0045] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0046] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0047] <Supplementary Notes> Some or all of the above embodiments may be described as in the following supplementary notes. Below, an outline of the configurations of an information processing device, an information processing method, and a program according to the present disclosure will be described. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) An information processing device comprising: an input unit that accepts input examples including a plurality of features and a condition that a prediction output from a prediction model when the input examples are input to the prediction model must satisfy; a generation unit that generates proposed modifications to one or more features of the input examples so that the prediction output from the prediction model for the input examples satisfies the condition; and a selection unit that selects, from reference examples including a plurality of the features, the reference example that satisfies the condition based on the proposed modifications. (Supplementary Note 2) The information processing device according to Supplementary Note 1, wherein the selection unit selects the reference example that includes the features that satisfy the proposed modifications. (Supplementary Note 3) The information processing device according to Supplementary Note 1, wherein the selection unit selects the reference example that satisfies the condition based on the proposed modifications and the input examples. (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the selection unit selects the reference case based on a similarity between the input example and the reference case according to a preset criterion. (Supplementary Note 5) The information processing device according to Supplementary Note 4, wherein the selection unit selects the reference case based on a similarity, according to a preset criterion, between a feature in the input example that is different from the feature of the revision and a feature in the reference case that corresponds to the feature. (Supplementary Note 6) The information processing device according to Supplementary Note 5, wherein the selection unit calculates the similarity between the feature in the input example and the feature in the reference case according to an importance set for the feature, and selects the reference case based on the calculated similarity. (Supplementary Note 7) The information processing device according to Supplementary Note 5, wherein the selection unit selects the reference case based on the similarity between the feature in the input example and the feature in the reference case, selected according to the importance set for the feature.(Supplementary Note 8) The information processing device according to Supplementary Note 1, comprising an output unit that outputs the selected reference case together with the proposed revision. (Supplementary Note 9) An information processing method that receives an input case including a plurality of features and a condition that a prediction output from a prediction model when the input case is input to the prediction model must satisfy, generates proposed revisions to one or more of the features of the input case so that the prediction output by the prediction model for the input case satisfies the condition, and selects the reference case that satisfies the condition based on the proposed revisions from reference cases including a plurality of the features. (Supplementary Note 10) The information processing method according to Supplementary Note 9, wherein the reference case that satisfies the condition is selected based on the proposed revision and the input case. (Supplementary Note 11) The information processing method according to Supplementary Note 10, wherein the reference case is selected based on a similarity between the input case and the reference case according to a predetermined criterion. (Supplementary Note 12) The information processing method according to Supplementary Note 11, wherein the reference case is selected based on a similarity between a feature in the input example that is different from the feature of the revision proposal and a feature in the reference case that corresponds to the feature, according to a preset criterion. (Supplementary Note 13) The information processing method according to Supplementary Note 12, wherein the similarity between the feature in the input example and the feature in the reference case is calculated according to an importance set for the feature, and the reference case is selected based on the calculated similarity. (Supplementary Note 14) The information processing method according to Supplementary Note 9, wherein the selected reference case is output together with the revision proposal. (Supplementary Note 15) A computer-readable storage medium storing a program that causes a computer to execute the following processes: accept an input example including a plurality of features and a condition that must be satisfied by a prediction output from a predictive model when the input example is input to the predictive model; generate a proposed modification of one or more of the features of the input example so that the prediction output from the predictive model for the input example satisfies the condition; and select, from reference cases including a plurality of the features, the reference case that satisfies the condition based on the proposed modification.

[0048] REFERENCE SIGNS LIST 10 Information processing device 11 Input unit 12 Revision plan generation unit 13 Reference case selection unit 14 Output unit 16 Model storage unit 17 Reference case storage unit 100 Information processing device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Input unit 122 Generation unit 123 Selection unit

Claims

1. An input unit that accepts an input example containing multiple features, and conditions that the prediction output by the prediction model should satisfy when the input example is input to the prediction model, A generation unit that generates a suggested modification for one or more of the features of the input example so that the prediction output by the prediction model for the input example satisfies the conditions, A selection unit that selects a reference example that satisfies the conditions based on the proposed modification from reference examples that include multiple such features, Equipped with an information processing device.

2. An information processing apparatus according to claim 1, The selection unit selects the reference example that includes the features that satisfy the proposed modification. Information processing device.

3. An information processing apparatus according to claim 1, The selection unit selects the reference example that satisfies the conditions based on the proposed revision and the input example. Information processing device.

4. An information processing apparatus according to claim 3, The selection unit selects the reference example based on the degree of similarity between the input example and the reference example according to a predetermined criterion. Information processing device.

5. An information processing apparatus according to claim 4, The selection unit selects the reference example based on the similarity between the feature in the input example that is different from the feature of the proposed revision and the feature in the reference example that corresponds to that feature, according to a predetermined criterion. Information processing device.

6. An information processing device according to claim 5, The selection unit calculates the similarity between the feature in the input example and the feature in the reference example according to the importance set for the feature, and selects the reference example based on the calculated similarity. Information processing device.

7. An information processing device according to claim 5, The selection unit selects the reference example based on the similarity between the feature in the input example and the feature in the reference example, which is selected according to the importance set for the feature. Information processing device.

8. An information processing apparatus according to claim 1, The system includes an output unit that outputs the selected reference example along with the aforementioned revised proposal. Information processing device.

9. The information processing device is The system accepts an input example containing multiple features, and the conditions that the prediction output by the prediction model should satisfy when the input example is input into the prediction model. To ensure that the prediction output by the prediction model for the input example satisfies the conditions, one or more suggested modifications to the features of the input example are generated. From reference examples containing multiple such features, select the reference example that satisfies the conditions based on the proposed modification. Information processing methods.

10. The system accepts an input example containing multiple features, and the conditions that the prediction output by the prediction model should satisfy when the input example is input into the prediction model. To ensure that the prediction output by the prediction model for the input example satisfies the conditions, one or more suggested modifications to the features of the input example are generated. From reference examples containing multiple such features, select the reference example that satisfies the conditions based on the proposed modification. A program that instructs a computer to perform a process.