Information processing system and information processing method

Abstract

To provide an information processing system with the function to adjust a content of the reason explanation depending on a degree of satisfaction of a person to be explained the reason.SOLUTION: An information processing system has a control unit and a storage unit connected to the control unit. The storage unit holds prediction models for outputting a prediction result based on an input value and multiple reason explanations for explaining a reason the prediction result was predicted, and each of the multiple reason explanations includes one or more descriptive elements. The control unit calculates weight indicating a degree of contribution of each explanation element included in each reason explanation to the satisfaction degree on the basis of the satisfaction degree to the reason explanation indicating a degree of satisfaction of the person who received each reason explanation. On the basis of the weight, an extended explanation that is a new reason explanation that includes one or more explanatory elements is created, and on the basis of, whether the explained person is satisfied with the created extended explanation, the control unit determines whether to create a new extended description or to select any of the one or more extended descriptions that are already created as a final extended description.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to explaining reasons for prediction judgments made by artificial intelligence. [Background technology]

[0002] JP 2020-129295 A (Patent Document 1) describes a method in which: (1) a computer analyzes data to be analyzed based on an analytical model including multiple discrimination conditions for discriminating abnormal data, the multiple discrimination conditions being generated by a learning process for multiple data; (2) when the data to be analyzed satisfies a predetermined discrimination condition among the multiple discrimination conditions, the computer identifies an abnormality reason associated with the predetermined discrimination condition based on discrimination information including an abnormality reason associated with each of the multiple discrimination conditions; and (3) the computer outputs information indicating the abnormality reason associated with the predetermined discrimination condition. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-129295 A Summary of the Invention [Problem to be solved by the invention]

[0004] When explaining the reasons for a prediction made using machine learning, if a uniform explanation is given about the basis for the judgment, the person receiving the explanation may not be able to understand it.

[0005] Patent Document 1 describes a technology for outputting information indicating the reason for an anomaly, but this technology is not intended to adjust the content of the explanation of the reason depending on the level of understanding or satisfaction of the person receiving the explanation.

[0006] The present invention has been made in consideration of the above points, and aims to provide a function for adjusting the content of the explanation of the reason depending on the level of understanding and satisfaction of the person who receives the explanation of the reason for a predictive judgment based on machine learning. [Means for solving the problem]

[0007] In order to solve the above problem, an information processing system according to one embodiment of the present invention is an information processing system having a control unit and a memory unit connected to the control unit, wherein the memory unit holds a prediction model that outputs a prediction result based on an input value and a plurality of reason explanations for explaining why the prediction result was predicted, each of the plurality of reason explanations including one or more explanatory elements, and the control unit calculates a weight indicating the magnitude of contribution of each of the explanatory elements included in each of the reason explanations to the degree of persuasion based on the persuasion degree of each of the reason explanations indicating the degree of persuasion of the person receiving each of the reason explanations, creates an extended explanation that is a new reason explanation including one or more of the explanatory elements based on the weight, and determines whether to create the new extended explanation or select one of the one or more extended explanations already created as the final extended explanation based on whether the person receiving the explanation is convinced by the created extended explanation. Effect of the Invention

[0008] According to one aspect of the present invention, it is possible to search for and adjust an explanation for a certain prediction judgment that is convincing to the person being explained about.

[0009] Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram showing an example of an outline of a flow of using an information processing system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating an example of functional blocks of the information processing system according to the embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram showing an example of an overall processing flow according to an embodiment of the present invention. [Figure 4] FIG. 11 is a diagram showing an example of a flow of adjusting the reason explanation content in the embodiment of the present invention. [Diagram 5] FIG. 13 is a diagram showing an example of a screen display of an initial reason explanation content by the information processing system according to the embodiment of the present invention. [Figure 6] 11 is a diagram showing an example of a screen display for inputting a relative degree of satisfaction regarding a plurality of reason explanations, according to the information processing system in the embodiment of the present invention. FIG. [Figure 7] 13 is an example of a screen display of a final reason explanation content by the information processing system according to the embodiment of the present invention. [Figure 8] 1 is a diagram illustrating an example of a table with a data structure regarding whether or not a correction is made and whether or not an update is applied for a target case, which is held by an information processing system according to an embodiment of the present invention. [Figure 9] 1 is a diagram illustrating an example of a table with a data structure related to updating of item values ​​included in a target case held by an information processing system according to an embodiment of the present invention. [Figure 10] 1 is a diagram showing an example of a table with a data structure relating to explanations of reasons for prediction judgments, explanatory elements, and extended explanations for items included in target cases held by an information processing system in an embodiment of the present invention. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, examples of embodiments of the present invention will be described with reference to the drawings.

[0012] FIG. 1 is a diagram showing an example of an outline of a flow of use of an information processing system 4 according to an embodiment of the present invention.

[0013] The information processing system 4 includes an information processing device 1 and a client terminal 2. First, the information processing device 1 creates a machine learning model 5 using data constituting past examination documents as training data (step S1).

[0014] Next, when data constituting the document to be inspected is input as evaluation data (step S2), the information processing device 1 performs a prediction judgment based on the machine learning model 5 and outputs the prediction judgment result (step S3). At this time, the information processing device 1 also creates a judgment reason that led to this prediction judgment result (step S4). The information processing device 1 transmits the prediction judgment result and the judgment reason to the client terminal 2.

[0015] The document review screen display unit 3 of the client terminal 2 displays the predicted judgment results and the judgment reasons on the document review screen display unit 3 (step S5), and then acquires an input of the relative satisfaction level by the person being explained (step S6). The client terminal 2 transmits the input value of the relative satisfaction level to the information processing device 1.

[0016] Thereafter, the information processing device 1 calculates which explanatory element is recognized as important among the explanatory elements constituting the judgment reason based on the input value of the relative degree of satisfaction (step S7). After identifying the most important explanatory element, the information processing device 1 creates multiple extended explanations by adding other explanatory elements to that element as a base (step S8). The information processing device 1 transmits the multiple created extended explanations to the client terminal 2.

[0017] The client terminal 2 displays the extended explanation on the document review screen display unit 3, acquires an input value of the degree of satisfaction regarding the extended explanation, and transmits it to the information processing device 1. Based on the input value of the degree of satisfaction regarding the extended explanation, the information processing device 1 determines whether to finalize the content of the final reason explanation or to create another extended explanation.

[0018] FIG. 2 is a diagram showing an example of functional blocks of the information processing system 4 according to the embodiment of the present invention.

[0019] The information processing device 1 is an information processing device consisting of a control unit 101, a memory unit 102, a model creation unit 103, a prediction judgment creation unit 104, a judgment condition extraction unit 105, a judgment reason creation unit 106, an individual satisfaction level calculation unit 107, an important explanatory element extraction unit 108, an extended explanation creation unit 109, a final explanation content creation unit 110 and a prediction result application unit 111, and is, for example, a server computer.

[0020] The client terminal 2 is an information processing device including a display output unit 201, a relative satisfaction level input unit 202, and an extended explanation satisfaction level input unit 203, and is, for example, a client PC (Personal Computer) that operates in cooperation with the information processing device 1.

[0021] The control unit 101 is hardware that controls the execution of various processes in the information processing device 1 and the client terminal 2, and is, for example, a central processing unit (CPU).

[0022] The memory unit 102 is a data storage medium that stores training data, evaluation data, the created machine learning model, etc., and refers to data storage such as a memory, HDD (Hard Disk Drive), and SSD (Solid State Drive).

[0023] The model creation unit 103 creates a machine learning model using the training data stored in the storage unit 102 as an input.

[0024] The prediction judgment creation unit 104 inputs the evaluation data to the machine learning model created by the model creation unit 103, and creates a prediction judgment result regarding the evaluation data.

[0025] The judgment condition extraction unit 105 extracts judgment conditions for deriving the above-mentioned predicted judgment results and combinations of the judgment results as explanatory elements.

[0026] The judgment reason generating unit 106 generates a plurality of judgment reasons by combining the explanatory elements extracted by the judgment condition extracting unit 105 .

[0027] The individual satisfaction degree calculation unit 107 calculates individual satisfaction degrees for individual explanations of the reason for determination from the relative satisfaction degrees for a plurality of reasons for determination inputted via a relative satisfaction degree input unit 202 described later.

[0028] The important explanatory element extraction unit 108 assigns the individual satisfaction levels calculated by the individual satisfaction level calculation unit 107 as coefficients of the explanatory elements constituting the individual determination reason explanations, and identifies the explanatory element with the largest coefficient as an important explanatory element.

[0029] The extended explanation creating unit 109 creates a plurality of extended explanations of the reason for determination by adding other explanatory elements to the important explanatory elements identified by the important explanatory element extracting unit 108 .

[0030] When the degree of satisfaction input via the extended explanation satisfaction input unit 203 described later is smaller than a predetermined threshold, the final explanation content creation unit 110 determines that the extended explanation is not satisfactory and excludes it from the candidates for the final reason explanation.It extracts extended explanations with relatively good judgment accuracy from extended explanations with a degree of satisfaction greater than the predetermined threshold, extracts them as candidates for the final reason explanation, and displays them via the display output unit 201.

[0031] The prediction result application unit 111 updates the evaluation data stored in the storage unit 102 using the prediction judgment result regarding the final reason explanation made in the final explanation content creation unit 110 via the display output unit 201. At this time, in addition to the data before and after application, the presented reason explanation, the explanation elements constituting the explanation element, the input value of the satisfaction level for the reason explanation, the ID of the explanation target person, the explanation date and time, etc. may be stored in the storage unit 102 so that the history before and after application can be traced later.

[0032] The display output unit 201 is a display device or the like that displays and outputs a plurality of judgment reasons created by the judgment reason creating unit 106 and also displays and outputs the extended judgment reason explanation created by the extended explanation creating unit 109 .

[0033] The relative degree of convincing input unit 202 is a device for inputting the relative degree of convincing between multiple explanations regarding the multiple reasons for judgment displayed and output via the display output unit 201, and is, for example, a keyboard, a mouse, or a touch-type display device.

[0034] The extended explanation satisfaction level input unit 203 is a device for inputting the satisfaction level of each of a plurality of extended explanations displayed and output via the display output unit 201, and may be the same as the above-mentioned keyboard or the like.

[0035] The initial explanation satisfaction level input unit 204 is a device for inputting a satisfaction level regarding the initial reason explanation displayed and output via the display output unit 201, and may be the same as the other input units described above.

[0036] 2 shows one control unit 101 and one storage unit 102, each of the information processing device 1 and the client terminal 2 may actually include the control unit 101 and the storage unit 102. The model creation unit 103, the prediction judgment creation unit 104, the judgment condition extraction unit 105, the judgment reason creation unit 106, the individual satisfaction level calculation unit 107, the important explanatory element extraction unit 108, the extended explanation creation unit 109, the final explanation content creation unit 110, and the prediction result application unit 111 may be realized by the control unit 101 of the information processing device 1 executing a program stored in the storage unit 102.

[0037] FIG. 3 is a diagram showing an example of an overall processing flow in the embodiment of the present invention.

[0038] In step S301, the model creation unit 103 performs machine learning using the training data stored in the storage unit 102 as an input, creates a machine learning model for prediction judgment using a predetermined learning algorithm, and stores the model in the storage unit 102.

[0039] In step S302, the prediction judgment creation unit 104 inputs the data constituting the document to be examined stored in the storage unit 102 as evaluation data for the machine learning model created in step S301, creates a prediction judgment result for each evaluation data, and outputs it to the storage unit. The prediction judgment result may be, for example, a classification prediction or numerical prediction result using the machine learning model, or a prediction result of correction contents.

[0040] In step S303, the judgment condition extraction unit 105 extracts judgment conditions for deriving each predicted judgment result for each evaluation data in the machine learning model created by the learning algorithm used in step S301. There may be one judgment condition, or multiple judgment conditions may be used. An example of a judgment condition is height 170 cm or more. The judgment condition extraction unit 105 makes a judgment based on the judgment condition, creates an explanation element that combines the judgment result, such as the degree of conformity or similarity to a certain classification, not limited to a binary value of Yes or No, with the judgment condition, and stores the explanation element in the storage unit.

[0041] In step S304, when x explanatory elements are extracted in step S303, the judgment reason creation unit 106 creates judgment reasons for the number of combinations (ΣxCr) of r (1≦r≦x) elements selected from the x different explanatory elements. Note that although the maximum number of combinations to be created as judgment reasons is shown here, the number of judgment reasons used hereafter does not need to be the maximum number. Note that, when an explanatory element that contributes to improving prediction accuracy is known in advance among the explanatory elements constituting the judgment reason, the judgment reasons to be created may be limited to only those including that explanatory element.

[0042] In step S305, the control unit 101 displays the reason for determination created in step S304 via the display output unit 201. This reason for determination is not an adjusted extended explanation, which will be described later, and therefore will be called an initial reason for determination.

[0043] In step S306, if an input indicating that the person is satisfied with the displayed and output reason for judgment is received via the initial explanation satisfaction level input unit 204, or if the satisfaction level is greater than a certain threshold, it is determined that the person being explained with the displayed and output reason for judgment is satisfied, and the process proceeds to step S307. At this time, in order to search for a reason explanation aimed at a more accurate prediction judgment, the process may proceed to a process of adjusting the content of the reason explanation (step S308) described later. At this time, the display output unit 201 may display a confirmation screen that displays other content of the reason explanation, etc.

[0044] In step S307, the prediction result application unit 111 updates history information related to the prediction judgment result using the prediction judgment result based on the explanatory element constituting the judgment reason in step S305. At this time, a confirmation screen related to the application of the prediction judgment result may be displayed via the display output unit 201. In addition, the input history up to now, the explanation of the reason performed, the explanatory element constituting the explanation of the reason, the extended explanation described later, the degree of satisfaction related to them, and the record related to the adoption or rejection of the prediction result may be stored as history in the storage unit 102. Examples of the history to be stored will be described later with reference to Figs. 8 to 10.

[0045] In step S308, if the degree of satisfaction with the reason explanation in step S305 is lower than a predetermined threshold, or if there is an input indicating that the user is not satisfied, adjustment processing of the content of the reason explanation is performed. Details of this processing will be described later in the description of FIG. 4.

[0046] In step S309, the person to whom the explanation has been given inputs, via the input unit, whether or not he / she is satisfied with the final explanation of the reason after the content of the explanation of the reason has been adjusted, or inputs the degree of satisfaction. If there is an input indicating satisfaction, or if the input degree of satisfaction is greater than a predetermined threshold, the process proceeds to step S306. On the other hand, if there is no input indicating satisfaction, or if the input degree of satisfaction is less than the predetermined threshold, the process proceeds to step S310.

[0047] In step S310, the prediction result application unit 111 does not apply correction using the result of classification prediction by machine learning or the result of correction prediction, and maintains the state before application. At this time, the input history up to that point, the explanation of the reason given, the explanatory elements constituting the explanation of the reason, the extended explanation described later, the degree of satisfaction therewith, and records regarding the adoption or rejection of the prediction result may be stored in the storage unit 102 as history.

[0048] In step S311, if there is another prediction judgment for the same evaluation data, or if the other evaluation data includes a prediction judgment, the process returns to step S302 and a prediction judgment is made. On the other hand, if there is no other such judgment, the process for this case ends.

[0049] FIG. 4 is a diagram showing an example of a flow of adjusting the reason explanation content in the embodiment of the present invention.

[0050] In step S401, the judgment reason generating section 106 displays the generated judgment reasons on the display output section 201.

[0051] In step S402, the recipient checks the displayed contents of a plurality of judgment reasons, and then inputs the relative degree of satisfaction for each combination of judgment reasons via the relative degree of satisfaction input unit 202.

[0052] In step S403, the individual satisfaction degree calculation unit 107 calculates an individual satisfaction degree for each judgment reason by, for example, the analytic hierarchy process.

[0053] First, when the judgment reason creating unit 106 creates N explanations for judgment reasons, each explanation is represented as nx (1≦x≦N). First, the person being explained inputs a numerical value expressing which explanation, n1 or n2, the person being explained is relatively more convinced of via the relative degree of convinced input unit 202. Here, the numerical value itself is not directly input, but may be input indirectly via a selection of a sentence expression corresponding to the numerical value.

[0054] The individual satisfaction degree calculation unit 107 performs this comparison for all combinations of judgment reasons. N This is performed C2 times to create each element of the paired comparison matrix A. However, when the number n of created judgment reasons is large, the number of elements constituting the paired comparison matrix A will be large, and the comparison work will take time. Therefore, N may be divided into similar judgment reasons, and a paired comparison matrix A' may be created for each subset of N, and weights may be calculated for individual reason explanations as described below, and finally, these may be integrated when calculating the weights of each judgment reason. Also, it is not necessary to input relative degrees of satisfaction for all combinations of judgment reasons. In this case, a paired comparison matrix may be created that is inferred by complementing the relative degrees of satisfaction that have already been input.

[0055] Next, the individual satisfaction degree calculation unit 107 calculates the weight for each reason explanation n from the pairwise comparison matrix A. The pairwise comparison matrix A is calculated by dividing the maximum eigenvalue of the matrix by λ max , and the principal eigenvector associated with the maximum eigenvalue is w, then equation (1) is satisfied.

[0056] Aw=λ max w (1)

[0057] In this case, the elements w1, w2, ..., wN of the principal eigenvector w that satisfy this formula become the weights of the reason explanations of n1, n2, ..., nN, respectively. Other methods for calculating the weights include the geometric mean method, the harmonic mean method, and the power method, but there are no particular limitations on the method of calculating and approximating the weights, and any method can be applied.

[0058] In step S404, the individual satisfaction level calculation unit 107 calculates the weight of each explanatory element constituting the judgment reason.

[0059] If we assume that the degree of persuasion for each reason explanation is equal to the sum of the degrees of persuasion for the explanatory elements that make up each reason explanation, when N reason explanations are composed of a maximum of P explanatory elements, the principal eigenvector w representing the degree of persuasion for each reason explanation can be expressed as shown in Equation (2) using an N-row, P-column matrix B indicating the presence or absence of a component py (1≦y≦P) that makes up each judgment reason, and a vector g.

[0060] w = Bg (2)

[0061] Now, by solving these N simultaneous equations to obtain vector g, the weight (degree of convincing) gy for each explanatory element py can be calculated.

[0062] In step S405, the important explanatory element extraction unit 108 identifies the explanatory element with the largest weight as the important explanatory element from the results of the weight calculation of each element in the above steps. For example, the important explanatory element extraction unit 108 extracts the explanatory element pg_max with the largest weight from among the above-mentioned components.

[0063] In step S406, the extended explanation creation unit 109 creates an extended explanation with the important explanatory element identified in the above step as its root, and presents it via the display output unit 201. For example, the extended explanation creation unit 109 creates an extended explanation E1 with the important explanatory element pg_max as its explanatory element. Next, when the extended explanation creation unit 109 rearranges the explanatory elements py in descending order of weight to produce branch and leaf elements pg2, pg3, ..., it sequentially creates extended explanations E2, E3, ... by adding the branch and leaf elements pg2, pg3, ... to the root element pg_max. At this time, there is no limit to the number of explanatory elements to be added, and if there are a total of P explanatory elements, Σ P C k There are (k=1 to P) possible combinations. However, it is not necessary to create extended explanations for all combinations, and it may be limited to a few. For example, the explanatory elements that improve the prediction accuracy may be added in order, or the number of extended explanations to be presented may be limited by adding the explanatory elements in order of the weights of the explanatory elements calculated in advance.

[0064] In step S407, the person being explained inputs the degree of satisfaction with the multiple extended explanations Ex presented via the display output unit 201 through the extended explanation satisfaction input unit 203. This degree of satisfaction may be a binary value, such as whether the person being explained was satisfied or not satisfied, or may be a numerical representation of the degree of satisfaction of the person being explained. This degree of satisfaction is input for each extended explanation, but it is not essential to input it for all extended explanations.

[0065] In step S408, the extended explanation creation unit 109 detects whether the person being explained to is convinced, based on the degree of satisfaction for each extended explanation input in step S407. If the input value is a degree of satisfaction, it may be possible to detect whether the person is convinced by comparing it with a predetermined threshold value. If it is determined that the person being explained to is convinced of a certain extended explanation, the process proceeds to S409. On the other hand, if it is determined that the person is not convinced, the process returns to S406 and another extended explanation is created.

[0066] In step S409, the final explanation content creation unit 110 identifies the extended explanation that can achieve the highest prediction accuracy from the group of extended explanations that are determined to have convinced the recipient, and adopts that extended explanation as the final explanation content.

[0067] In step S410, even if there is one or more satisfactory extended explanations, if it is desired to improve the prediction accuracy by providing other extended explanations, the process returns to S406 for further trials. On the other hand, if confirmation using other extended explanations is not to be performed any more, the process ends and returns to the flow diagram of FIG.

[0068] According to the above process, it becomes possible to adjust the content of the reason explanation according to the degree of satisfaction of the person to whom the explanation is given.

[0069] FIG. 5 is a diagram showing an example of a screen display of the initial reason explanation content by the information processing system 4 in the embodiment of the present invention.

[0070] Here, an example of an application form item entered by an applicant regarding a spouse will be described. As shown in FIG. 5(a), there are multiple input items in the category of answers regarding spouse, and each is assigned an item ID of 01, 02, .... The attributes related to spouse include item 01, which indicates whether the applicant has a spouse, item 02, which indicates whether the applicant and spouse live together, and item 03, which indicates whether the spouse is supported by the applicant. When the initial input values ​​are blank for item 01, "1" (living together) for item 02, and "1" (support) for item 03, the information processing system 4 presents a corrected predicted value of "1" (having a spouse) for item 01. The recipient who receives the corrected predicted value can select whether to press a button to apply the corrected predicted value, press a button to display an explanation of the prediction that led to the presentation of this corrected predicted value, or do nothing with the corrected predicted value. Here, the explanation will continue assuming that the button for explaining the reason is pressed.

[0071] As shown in FIG. 5(b), for the corrected prediction of item 01, the initial judgment reason and buttons for acceptance / rejection are displayed. If the person being explained agrees with this initial judgment reason, he / she presses the apply button in the acceptance / rejection section. On the other hand, if he / she does not agree, he / she presses the not apply button. If he / she wishes to check other explanations of the reason, he / she may press the reserve button. Here, the explanation will continue assuming that the reserve button has been pressed to check other explanations of the reason.

[0072] FIG. 6 is a diagram showing an example of a screen display for inputting a relative degree of satisfaction regarding a plurality of reason explanations by the information processing system 4 according to the embodiment of the present invention.

[0073] FIG. 6(a) is an example of a screen when the information processing system 4 creates multiple reason explanations n in addition to the initial reason explanation and displays them to the person being explained about, and the person being explained about inputs the relative degree of satisfaction with the reason explanations.

[0074] For example, in order to input the relative degree of convincing between reason explanation n1 and reason explanation n2, a line segment with the left direction assigned to reason explanation n1 and the right direction assigned to reason explanation n2 may be displayed, and the person being explained to may operate a pointer to point to any position on the line segment. For example, the person being explained to may operate the pointer so that if the degree of convincing of both reason explanations is judged to be about the same, the person being explained to point to the midpoint of the line segment, if the person being explained to judges that the degree of convincing of reason explanation n2 is relatively high, the person being explained to point to a position to the right of the midpoint of the line segment, and the higher the relative degree of convincing of reason explanation n2, the closer to the right end of the line segment the pointer is.

[0075] 6(a) shows an example of input in which reason explanation n2 is overwhelmingly more convincing than reason explanation n1, reason explanation n1 is relatively more convincing than reason explanation n3, and reason explanation n2 is slightly more convincing than reason explanation n3. After the person receiving the explanation has completed inputting the degree of convincing, he or she notifies the information processing system 4 of the completion of input by pressing a button at the bottom of the screen, and the process proceeds to extraction of important explanatory elements.

[0076] Fig. 6(b) shows an example of a screen on which the information processing system 4 displays a plurality of created extended explanations E to the person being explained to, and the person being explained to inputs whether or not he or she agrees with the extended explanation or the degree of agreement. For example, a line segment may be displayed with the left direction assigned to "not agreeable" and the right direction assigned to "agreeable," and the person being explained to may input the degree of agreement by operating a pointer to indicate any position on the line segment in the same manner as in Fig. 6(a).

[0077] In this case, if the person being explained about the extended explanations E1, E2, E3, ... is satisfied, the input will be shifted to the right, and if the person is not satisfied, the input will be shifted to the left. At this time, a predetermined threshold may be determined, and it may be determined that the person being explained about is satisfied when an input of a degree of satisfaction equal to or greater than the threshold is made. After completing the input of the degree of satisfaction, a button at the bottom of the screen is pressed to notify the information processing system 4 that the input has been completed, and the process proceeds to the creation of the final reason explanation.

[0078] Note that Figure 6 is an example of a screen display to assist the recipient in intuitively inputting the relative degree of assent and the level of assent, and in reality the input may be made in a manner other than that described above, such as the recipient directly inputting a numerical value indicating the degree of assent, etc.

[0079] FIG. 7 shows an example of a screen display of the final reason explanation content by the information processing system 4 according to the embodiment of the present invention.

[0080] As a result of the process of creating a final explanation of reasons following the input of the degree of satisfaction shown in Figure 6(b) above, the item for reasons for judgment has the highest prediction accuracy, and the content of the item for reasons for judgment on the screen is updated based on the content of the final explanation of reasons that was most satisfactory to the person being explained to. Here, an example is shown in which the final explanation of reasons is extended explanation E2. For this reason, "E2", indicating extended explanation E2, is displayed in the item for reasons for judgment in Figure 7. However, the actual reason for judgment is not displayed as "E2", but rather the group of explanation elements that make up E2.

[0081] In the acceptance / rejection section corresponding to the reason for judgment in Fig. 7, an apply button, a not apply button, and a hold button are displayed. Since the person being explained to has input that he / she is satisfied in Fig. 6(b) described above, by pressing the apply button, he / she obtains final agreement regarding the application of the corrected predicted value. If he / she is still not satisfied, he / she can press the not apply button to not apply the corrected predicted value, or press the hold button to reserve the judgment and check another extended explanation.

[0082] FIG. 8 is a diagram showing an example of a table of a data structure relating to the presence or absence of correction and the presence or absence of update application of a target case held by the information processing system 4 in the embodiment of the present invention.

[0083] A table 800 of a data structure relating to whether or not a correction is made and whether or not an update is applied for a target case shown in FIG. 8 includes a case ID 801 , a distribution classification 802 , a classification determination process execution date and time 803 , whether or not an update is applied 804 , and an update process execution date and time 805 .

[0084] The case ID 801 is a value that allows all cases to be identified by a unique value. Here, a case is a collection of data that is the subject of processing by the information processing system 4. For example, if the user of the information processing system 4 is a government agency and the subject of processing by the information processing system 4 is a document submitted to the government agency, a single submitted document (for example, a document submitted by an applicant to make some kind of application) may be treated as one case. In this case, the recipient is an employee at the government agency who is responsible for processing the document.

[0085] The sorting classification 802 is a value that indicates the predicted classification result for each case, such as no correction required, correction required, etc. When this item is "correction required," it is understood that there is one or more items in this case that require correction.

[0086] The classification determination process execution date / time 803 is a value indicating the date / time when the above-mentioned classification determination process was performed.

[0087] The update application status 804 is a value that indicates whether or not an update has been made to the initial input value using a corrected predicted value or a manually corrected value, for cases that have been determined to require correction after the above-mentioned classification and judgment process. When this item is "Yes," the reason for the judgment regarding the corrected predicted value can be confirmed.

[0088] The update process execution date and time 805 is a value that indicates the date and time when the update using the corrected predicted value or the like from the initial input value was applied.

[0089] In the example of FIG. 8, for the case with case ID 801 of "1", it was determined that correction was not necessary, and therefore update using the corrected predicted value, etc. was not performed. For the case with case ID 801 of "2", it was determined that correction was required, and update using the corrected predicted value, etc. was applied. For the case with case ID 803 of "3", it was determined that correction was required (i.e., a corrected predicted value was obtained), but update using the corrected predicted value, etc. was not applied. Examples of cases in which update using a corrected predicted value is not applied even though it has been obtained include cases in which the recipient is not convinced by the explanation for the reason output together with the corrected predicted value, and therefore non-application, pending, or no input is made in the adoption / rejection item shown in FIG. 5 or FIG. 7, or cases in which the recipient judges that the initial input value is more correct than the corrected predicted value.

[0090] FIG. 9 is a diagram showing an example of a table of a data structure related to updating of item values ​​included in a target case held by the information processing system 4 in the embodiment of the present invention.

[0091] A table 900 of a data structure related to updating of item values ​​included in a target case shown in FIG. 9 includes item ID 901, attribute name 902, update application status 903, initial input value 904, corrected predicted value 905, and final result value 906.

[0092] The item ID 901 is a value for uniquely identifying an entry item that constitutes a case. For example, when a submitted document corresponds to one case, the item ID identifies the items to be filled in by the submitter in the document (such as name, address, date of birth, whether or not the submitter has a spouse, whether or not the spouse lives together, etc.).

[0093] The attribute name 902 is a value that indicates the name of an entry item that constitutes a case.

[0094] The update application status 903 is the same as the update application status 804 described above with reference to FIG.

[0095] Initial input value 904 is the initial value of the item entered by the applicant.

[0096] The corrected predicted value 905 is an estimated value calculated by the information processing system 4 for the initial input value.

[0097] The final result value 906 is the value finally applied after processing such as correction prediction has been performed, and is updated by applying the initial input value, the corrected prediction value, or a manually corrected value.

[0098] FIG. 9 shows, as examples of the table 900, tables 900A, 900B, and 900C.

[0099] Table 900A shown in FIG. 9(a) is an example of stored data related to a case that is determined to be unnecessary for correction. For example, for the item ID 901 "01" for spouse presence / absence and the item ID 901 "02" for spouse cohabitation, the initial input value 904 is both "1" (i.e., the applicant has a spouse and the spouse cohabitation is entered as the initial input value), and if it is determined that this initial input value is appropriate, the corrected predicted value is not generated and the value is left blank. In this case, the final result value 906 may be stored in advance as the initial input value, which is "1", or may be stored when the person to whom the explanation is given approves the value. The column for update application 903 may be left blank when there is no update of values ​​for all items by the corrected predicted value or the like.

[0100] Table 900B shown in FIG. 9(b) is an example of stored data related to a case that is determined to require correction. For example, the initial input value of the item "01" regarding spouse presence / absence is blank, and the items "02" regarding spouse cohabitation and other spouses have initial input values. In this case, the information processing system 4 does not specify whether the cause is an omission in the item "01", but creates "1" as the corrected predicted value of the item "01" and presents it to the person to whom the explanation is given. In this case, the person to whom the explanation is given can refer to the explanation of the reason for the corrected predicted value of this item as shown in FIG. 10. The person to whom the explanation is given checks the reason for the judgment of the corrected predicted value, and if the corrected predicted value is applied to the final result value, the item "Update applied / not applied" 903 is changed to "Yes". On the other hand, when it is determined that the initial input value of the item 02 does not require correction, the item "Update applied / not applied" 903 is left blank.

[0101] Table 900C shown in FIG. 9(c) is an example of stored data related to a case that is determined to require correction. For example, the initial input value of the spouse presence / absence in item "01" is "1", and when the information processing system 4 determines that this initial input value is appropriate, it does not generate a corrected predicted value but leaves it blank. At this time, the final result value may be stored in advance as the same value as the initial input value, "1", or may be stored as the final result value when the person receiving the explanation approves the value. Next, when the initial input value of the spouse cohabitation in item "02" is 1 (living together), but the corrected predicted value is predicted to be "0" (i.e., living separately) in relation to other items, the person receiving the explanation can refer to the explanation of the reason for the corrected predicted value as shown in FIG. 10, as in FIG. 9(b). Note that when this corrected predicted value is incorrect, the initial input value "1" is adopted as the final result value, not the corrected predicted value "0". At this time, the update application presence / absence 903 for both items "01" and "02" remains blank.

[0102] FIG. 10 is a diagram showing an example of a table of a data structure related to explanations of reasons for prediction determinations, explanatory elements, and extended explanations of items included in a target case held by the information processing system 4 in the embodiment of the present invention.

[0103] FIG. 10(a) shows a table 1000 for managing a plurality of reason explanations created by the information processing system 4.

[0104] The table 1000 shown in FIG. 10( a ) includes a reason explanation ID 1001 , a judgment condition 1002 , and a degree of satisfaction 1003 .

[0105] The reason explanation ID 1001 is a value for uniquely identifying a reason explanation created for a certain prediction determination.

[0106] The judgment condition 1002 is a logical configuration of the explanatory elements that constitute each explanation of the reason for a certain prediction judgment.

[0107] The satisfaction level 1003 stores the satisfaction level calculated in step S403 for each individual explanation of the reason based on the relative satisfaction level input by the person who received the explanation of the reason for a certain prediction judgment.

[0108] Fig. 10 shows, as an example, data stored regarding the explanation of the reason corresponding to the corrected predicted value for the initial input value of the spouse presence item. Table 1000 in Fig. 10(a) shows an example of the explanation of the reason when the initial input value of the spouse presence item is "absent" (i.e., "0") and the corresponding predicted corrected value is "present" (i.e., "1"). This corresponds to the example of item ID "01" in Fig. 9(b).

[0109] In this example, the reason explanation with reason explanation ID 1001 of "n1" corresponds to outputting "Yes" as the corrected predicted value of the spouse presence / absence field when the determination condition that the initial input value of the spouse presence / absence field is "No" and the initial input value of the spouse cohabitation field is "Yes" is satisfied. In this case, the fact that the determination condition is satisfied is output as the reason explanation.

[0110] On the other hand, the reason explanation with reason explanation ID 1001 of "n2" corresponds to outputting "Yes" as the corrected predicted value of the spouse presence item when the determination conditions that the initial input value of the spouse presence item is "No," the initial input value of the spouse cohabitation item is "Yes," and the initial input value of the spouse not required item is "Yes" are satisfied. In this case as well, the fact that the determination conditions are satisfied is output as the reason explanation.

[0111] In the example of Fig. 10(a), the persuasiveness 1003 of the reason explanation with the reason explanation ID 1001 of "n1" and the reason explanation with the reason explanation ID 1001 of "n2" is "0.35" and "0.45", respectively. This indicates that when a similar predicted correction value was presented in a case with a similar situation in the past, the latter reason explanation was more likely to convince the person who was explained.

[0112] FIG. 10B shows a table 1010 for managing explanation elements that constitute a reason explanation created by the information processing system 4.

[0113] The table 1010 shown in FIG. 10( b ) includes an explanation element ID 1011 , an explanation element 1012 , and a weight 1013 .

[0114] The explanation element ID 1011 is a value for uniquely identifying an explanation element that constitutes the reason explanation.

[0115] The explanation element 1012 is composed of an attribute name and a judgment condition regarding the attribute value.

[0116] Weight 1013 is an importance value calculated for each explanatory element.

[0117] In the example of Figure 10(b), "Cohabitation with spouse = Yes", "Whether spouse is present = No", "Spouse support = Yes", etc. are recorded as explanatory elements 1012 constituting the reason explanation illustrated in Figure 10(a), and further, "0.2", "0.3", "0.1", etc. are recorded as the respective weights 1013.

[0118] FIG. 10(c) shows a table 1020 for managing the degree of understanding of the explained person and the presence or absence of an explanation element that constitutes the extended explanation created by the information processing system 4.

[0119] The table 1020 shown in FIG. 10(c) includes an extended explanation ID 1021, a satisfaction degree 1022, a satisfaction level 1023, and a plurality of explanation elements (eg, an explanation element p1_1024, an explanation element p2_1025, and an explanation element pN_1026, etc.).

[0120] The extended explanation ID 1021 is a value for uniquely identifying an extended explanation created by the information processing system 4.

[0121] The satisfaction / dissatisfaction 1022 is a value expressing whether the explained person is satisfied or not satisfied with the extended explanation inputted by the explained person.

[0122] The degree of assent 1023 is a value entered by the person to whom the explanation was given, expressing the degree to which the person was assent to the extended explanation.

[0123] Explanation element p1_1024 through explanation element pN_1026, etc., are values ​​indicating whether each explanation element is a component of each extended explanation, and store 1 if it is a component.

[0124] In the example of Figure 10(c), the extended explanation whose extended explanation ID 1021 is "E0x" includes explanatory element p2 (according to Figure 10(b) "Spouse status = No") and explanatory element p1 (according to Figure 10(b) "Living with spouse = Yes") as components, but does not include explanatory element pN as a component. On the other hand, the extended explanation whose extended explanation ID 1021 is "E0x+1" includes explanatory element p2, explanatory element p1, and explanatory element pN as components.

[0125] Here, an example will be described in which "Yes" (i.e., the user is satisfied with the extended explanation) is input as the satisfaction / dissatisfaction 1022 for the extended explanation ID 1021 of "E0x" and "No" is input as the satisfaction / dissatisfaction 1022 for the extended explanation ID 1021 of "E0x+1". In this case, the extended explanation "E0x" that is acceptable to the recipient and achieves the highest prediction accuracy among the acceptable extended explanations can be determined. In addition, when the satisfaction level is used, for example, when a predetermined threshold is set to 0.3, the satisfaction level of the extended explanation "E0x" is greater than the threshold and the satisfaction level of the extended explanation "E0x+1" is less than the threshold, so that the final extended explanation E0x can be determined by making a judgment in the same way as the satisfaction / dissatisfaction.

[0126] Here, referring again to Fig. 4, an example of the flow of processing based on the specific examples of Fig. 8 to Fig. 10 will be described. For example, weights 1013 of explanatory elements p1 to pN are calculated in step S404, and one of them (for example, explanatory element p2 with the largest weight 1013) is determined as an important explanatory element in step S405.

[0127] In step S406, which is executed first after step S405, a reason explanation including only the important explanatory element p2 is created and displayed as an extended explanation. In step S407, the person being explained to inputs whether or not he / she is satisfied with the extended explanation. If the person being explained to is satisfied (step S408: Yes), the process returns to step S406. In step S406, a new extended explanation is created by adding an explanatory element not yet included (for example, explanatory element p1 with the second largest weight 1013) to the extended explanation that the person being explained to is satisfied with, and the new extended explanation is displayed.

[0128] Each time the loop of steps S406 to S408 is repeated, an explanatory element is added to the extended explanation in step S406, so the number of explanatory elements included in the extended explanation gradually increases. The explanatory elements may be added in any order, for example, in order of the magnitude of the weights 1013. In general, the more explanatory elements included in the reason explanation, the higher the prediction accuracy. However, on the other hand, the more explanatory elements there are, the more complex the explanation becomes, making it difficult to determine whether the explanation is valid, and so on, and the degree of assent of the person being explained to tends to decrease.

[0129] When it is determined for the first time during the loop of steps S406 to S408 that the recipient is not convinced (step S408: No), the extended explanation (e.g., extended explanation "E0x") created just before the current extended explanation (e.g., extended explanation "E0x+1") can be said to be the one that the recipient can be convinced of and has the highest predictive accuracy among the extended explanations created so far. Therefore, that extended explanation is selected as the final extended explanation that satisfies both the predictive accuracy and the degree of persuasion (step S409).

[0130] The above process may be performed for all the explanatory elements from the beginning, or may be performed for some combinations of the explanatory elements. In this case, once the final extended explanation is selected, if it is determined that other extended explanations are to be tried (step S409: Yes), the above process is performed again for another combination of the explanatory elements.

[0131] The calculated degree of persuasion of each reason explanation, the weight of each explanation element calculated based on the calculated degree of persuasion, and the result of correction of the input value made by the person being explained based on the prediction result and the extended explanation therefor are stored in the storage unit 102 of the information processing system 4 as shown in Figures 8 to 10. At this time, the information shown in Figures 8 to 10 is stored in association with the identification information of the person being explained. The example in Figures 8 to 10 is one example of information corresponding to a certain person being explained, and similar information is actually stored for each person being explained.

[0132] Then, when each person to whom an explanation is given handles a new case, the information processing system 4 may create an extended explanation, etc., by referring to information corresponding to the person to whom the explanation is given in the past. Each person to whom an explanation is given has a different amount of knowledge, experience, etc., and therefore the type of explanation that convinces the person may differ. However, by generating an extended explanation based on information corresponding to the person to whom an explanation is given as described above, it is possible to provide the person to whom the explanation is given with an easy understanding of the reason.

[0133] On the other hand, the information processing system 4 may create an extended explanation or the like by referring to information corresponding to another person to whom the explanation is being given. For example, the person to whom the explanation is being given may refer to an extended explanation created based on the degree of understanding of a person other than himself (e.g., a person with more work experience than himself), and may learn what kind of explanation would be convincing to the other person, and may utilize this in his own work.

[0134] Note that the machine learning model 5 may be of any type, and the explanation for the corrected prediction result based on the machine learning model 5 may be generated in any manner, but an example is as follows: That is, a decision tree may be used as the machine learning model 5, and in that case, the explanation for the prediction result may include, as an explanatory element, an attribute name and an attribute value corresponding to at least one of a plurality of nodes on a path from the root node of the decision tree to the terminal node corresponding to the prediction result.

[0135] According to the present embodiment described above, it is possible to search for and adjust an explanation for the reason for a certain prediction judgment that is convincing to the person being explained about, and it is possible to determine the content of the explanation for the reason that is convincing to the person being explained about and that will result in the highest prediction accuracy.

[0136] Furthermore, the system according to the embodiment of the present invention may be configured as follows.

[0137] (1) An information processing system (e.g., information processing system 4) having a control unit (e.g., control unit 101) and a memory unit (e.g., memory unit 102) connected to the control unit, the memory unit holds a prediction model (e.g., machine learning model 5) that outputs a prediction result based on an input value, and a plurality of reason explanations (e.g., reason explanations in table 1000) for explaining why the prediction result was predicted, each of the plurality of reason explanations including one or more explanatory elements (e.g., the explanatory elements in table 1010), the control unit calculates a weight indicating the magnitude of the contribution of each explanatory element included in each reason explanation to the degree of persuasion based on the persuasion degree of each reason explanation indicating the degree of persuasion of the person who received each reason explanation (e.g., step S404), creates an extended explanation that is a new reason explanation including one or more explanatory elements based on the weight (e.g., step S406), and determines whether to create a new extended explanation or select one of the one or more extended explanations already created as the final extended explanation based on whether the person who received the explanation was persuasive with the created extended explanation (e.g., step S408).

[0138] This makes it possible to search for and adjust explanations for a certain predicted result that will convince the person being explained about it.

[0139] (2) In the above (1), the control unit calculates the degree of persuasion of each reason explanation based on the relative degrees of persuasion of a plurality of pairs of two reason explanations input by the explained person (eg, steps S402 to S403).

[0140] This makes it easier for the person receiving the explanation to input his / her degree of satisfaction.

[0141] (3) In (2) above, the control unit stores at least one of the relative degree of persuasion, the degree of persuasion of each reason explanation, the weight of each explanation element, one or more extended explanations, and the final extended explanation in the memory unit as an explanation history (e.g., step S307).

[0142] This allows the previously input degree of satisfaction to be used to create an explanation for a new prediction result.

[0143] (4) In (3) above, the explanation history includes identification information of the person being explained corresponding to the degree of persuasion of each reason explanation, and the control unit creates a reason explanation corresponding to the prediction result to be output to the person being explained together with the prediction result by the prediction model, based on the explanation history corresponding to the identification information of the person being explained.

[0144] This allows an appropriate explanation of the reasons to be prepared according to the level of understanding of the person being explained to.

[0145] (5) In (3) above, the explanation history includes identification information of the person being explained to corresponding to the degree of persuasion of each reason explanation, and the control unit creates a reason explanation corresponding to the prediction result to be output to the person being explained to together with the prediction result by the prediction model, based on the explanation history corresponding to the identification information of other people being explained to.

[0146] This allows the person receiving the explanation to use the other person's level of understanding as a reference for his or her own work.

[0147] (6) In (1) above, the control unit creates an extended explanation including one or more explanatory elements, and if information is input indicating that the recipient is convinced of the created extended explanation (e.g., step S408: Yes), it creates a new extended explanation by adding explanatory elements, and if information is input indicating that the recipient is not convinced of the created extended explanation (e.g., step S408: No), it selects the extended explanation created immediately before the extended explanation that the recipient was not convinced of (e.g., if the loop of steps S406 to S408 has been executed multiple times, the extended explanation created in step S406 of the previous loop of the loop that was judged to be No in step S408) as the final extended explanation.

[0148] This allows the recipient of the explanation to be convinced and allows for the selection of a highly accurate explanation of the reason.

[0149] (7) In (6) above, the control unit creates an extended explanation including the explanatory element with the highest weight (e.g., step S406), and if information is input indicating that the person being explained to is convinced by the created extended explanation (e.g., step S408: Yes), it creates a new extended explanation by sequentially adding explanatory elements in order of magnitude of weight (e.g., step S406 in the next or subsequent loops).

[0150] This allows for the creation of highly accurate explanations of reasons that are easy for the recipient to understand.

[0151] (8) In (1) above, the prediction model is a decision tree, and each explanatory element included in the multiple reason explanations for explaining why the prediction result by the prediction model was predicted includes an attribute name and an attribute value corresponding to each of multiple nodes on the path from the root node of the decision tree to the terminal node corresponding to the prediction result.

[0152] This allows you to create a proper explanation for the reason.

[0153] The present invention is not limited to the above-mentioned embodiment, and various modifications are included. For example, the above-mentioned embodiment has been described in detail for better understanding of the present invention, and is not necessarily limited to those having all the configurations described. Also, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Also, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

[0154] In addition, the above-mentioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as a program, table, file, etc. that realizes each function can be stored in a storage device such as a non-volatile semiconductor memory, a hard disk drive, or an SSD (Solid State Drive), or in a computer-readable non-transitory data storage medium such as an IC card, an SD card, or a DVD.

[0155] In addition, the control lines and information lines are shown as those considered necessary for the explanation, and not all control lines and information lines in the product are shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0156] 1: Information processing device 2: Client terminal 3: Document screening screen display section 4: Information processing system 5: Machine learning models 101: Control unit 102: Storage section 103: Model Creation Department 104: Prediction and Judgment Creation Department 105: Judgment condition extraction part 106: Judgment reason creation section 107: Individual satisfaction calculation unit 108: Important explanatory element extraction unit 109: Extended explanation generation unit 201: Display output unit 202: Relative satisfaction input section 203: Consent level input section

Claims

1. An information processing system having a control unit and a storage unit connected to the control unit, The storage unit holds a prediction model that outputs a prediction result based on an input value, and a plurality of reason explanations for explaining why the prediction result was predicted; each of the plurality of reason explanations includes one or more explanation elements; The control unit is Calculating a weight indicating the degree of contribution of each of the explanation elements included in each of the reason explanations to the degree of persuasion of the person receiving each of the reason explanations, based on the persuasion degree of each of the reason explanations, Based on the weights, an extended explanation is created, which is a new reason explanation including one or more of the explanatory elements; An information processing system characterized by determining whether to create a new extended explanation or select one of one or more already created extended explanations as the final extended explanation based on whether the person being explained is convinced by the created extended explanation.

2. 2. The information processing system according to claim 1, The information processing system is characterized in that the control unit calculates the degree of persuasiveness of each of the reason explanations based on the relative persuasiveness of multiple sets of two reason explanations input by the person being explained.

3. 3. The information processing system according to claim 2, The information processing system is characterized in that the control unit stores in the memory unit at least one of the relative degree of persuasion, the degree of persuasion of each reason explanation, the weighting of each explanation element, the one or more extended explanations, and the final extended explanation as an explanation history.

4. 4. The information processing system according to claim 3, The explanation history includes identification information of the person to whom the explanation is given corresponding to the degree of persuasion of each of the reason explanations, The control unit creates the explanation of the reason corresponding to the prediction result to be output to the recipient together with the prediction result by the prediction model, based on the explanation history corresponding to the identification information of the recipient. An information processing system characterized by the above.

5. 4. The information processing system according to claim 3, The explanation history includes identification information of the person to whom the explanation is given corresponding to the degree of persuasion of each of the reason explanations, The control unit creates the explanation of the reason corresponding to the prediction result to be output to the recipient together with the prediction result by the prediction model, based on the explanation history corresponding to the identification information of recipients other than the recipient.

6. 2. The information processing system according to claim 1, The control unit is creating said extended description including one or more of said description elements; When information indicating that the explained person is convinced of the created extended explanation is input, the new extended explanation is created by adding the explanation element; An information processing system characterized in that, when information indicating that the recipient is not convinced of the created extended explanation is input, the extended explanation created immediately before the extended explanation that the recipient was not convinced of is selected as the final extended explanation.

7. 7. The information processing system according to claim 6, The control unit is creating the expanded explanation that includes the explanation element with the highest weight; An information processing system characterized in that, when information indicating that the recipient is convinced of the created extended explanation is input, the new extended explanation is created by sequentially adding the explanatory elements in order of the magnitude of the weighting.

8. 2. The information processing system according to claim 1, the predictive model is a decision tree; An information processing system characterized in that each explanatory element included in a plurality of reason explanations for explaining why a prediction result by the prediction model was predicted includes an attribute name and an attribute value corresponding to each of a plurality of nodes on a path from a root node of the decision tree to a terminal node corresponding to the prediction result.

9. An information processing method executed by an information processing system having a control unit and a storage unit connected to the control unit, The storage unit holds a prediction model that outputs a prediction result based on an input value, and a plurality of reason explanations for explaining why the prediction result was predicted; each of the plurality of reason explanations includes one or more explanation elements; The information processing method includes: The control unit calculates a weight indicating the magnitude of contribution of each of the explanation elements included in each of the reason explanations to the degree of persuasion based on the persuasion degree of each of the reason explanations indicating the degree of persuasion of the person who received each of the reason explanations; A step of creating an extended explanation, which is a new reason explanation including one or more of the explanatory elements, based on the weight; An information processing method comprising: a step of determining whether to create a new extended explanation or select one of one or more already created extended explanations as the final extended explanation based on whether the recipient is convinced by the created extended explanation.

Images