Information processing system and information processing method

Abstract

To evaluate characteristics related to people while taking into account relationships between people.SOLUTION: An information processing system generates a learning model that outputs multiple types of adjusted label values related to relationships between objects and specified characteristics in response to input of explanatory variables, assigns each value of the multiple types of adjusted labels to an object to be analyzed based on explanatory variable values of the object to be analyzed and the generated learning model, and determines an adjusted label to be adopted as a final determination result for the object to be analyzed from the multiple types of adjusted labels based on each of the values assigned to the object to be analyzed and relationships between the adjusted labels.SELECTED DRAWING: Figure 1A

Description

[Technical field]

[0001] The present invention relates to an information processing system and an information processing method. [Background technology]

[0002] As a background technique in this technical field, there is JP 2007-317238 A (Patent Document 1), which states that "the system includes a user name input means for inputting a plurality of user names, a public data search means for searching public data including the user names inputted by the user name input means and acquiring the number of hits, a default value determination means for determining whether the number of hits acquired by the public data search means exceeds a default value, and a human relationship setting means for setting human relationship data between user names that are determined by the default value determination means to exceed the default value" (see abstract). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2007-317238 A Summary of the Invention [Problem to be solved by the invention]

[0004] The technology described in Patent Document 1 estimates relationships between people from public data. However, Patent Document 1 does not describe evaluating characteristics related to people while taking into account relationships between people. Therefore, one aspect of the present invention evaluates characteristics related to people while taking into account relationships between people. [Means for solving the problem]

[0005] In order to solve the above problem, one aspect of the present invention employs the following configuration: An information processing system includes a processor and a memory, and the memory holds label information indicating labels related to predetermined features of learning objects, relationship information indicating the presence or absence of relationships between the learning objects, explanatory variable information indicating values ​​of explanatory variables of the learning object and the analysis object, label adjustment method information indicating a plurality of adjustment rules for assigning a plurality of types of adjusted labels related to the relationships between objects and the predetermined features to the learning objects, and judgment criteria information indicating the relationships between the plurality of types of adjusted labels, and the processor assigns a plurality of types of adjusted labels to the learning objects based on the relationships indicated by the relationship information, the labels indicated by the label information, and the plurality of adjustment rules. The multiple types of adjusted labels are assigned to the object to be analyzed, and for each of the multiple types of adjusted labels assigned, a learning model is generated based on the values ​​of the explanatory variables of the object to be analyzed and the adjusted label, which outputs the value of the adjusted label when the explanatory variables are input. Based on the values ​​of the explanatory variables of the object to be analyzed and the generated learning model, the values ​​of each of the multiple types of adjusted labels are assigned to the object to be analyzed, and an adjusted label to be adopted as the final judgment result for the object to be analyzed is determined from the multiple types of adjusted labels based on the relationship between each of the values ​​assigned to the object to be analyzed and the judgment criteria information. Effect of the Invention

[0006] According to one aspect of the present invention, characteristics related to people can be evaluated taking into account relationships between people.

[0007] Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0008] [Figure 1A] 1 is a block diagram illustrating an example of a functional configuration of a relationship analysis system according to a first embodiment. [Figure 1B] 2 is a block diagram illustrating an example of a hardware configuration of a computer that configures the relationship analysis system in the first embodiment. [Diagram 2] FIG. 4 is a diagram illustrating an example of a data configuration of label information in the first embodiment. [Diagram 3] FIG. 11 is a diagram illustrating an example of a data configuration of application information during study in the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of a data configuration of learning-time relationship information in the first embodiment. [Diagram 5] FIG. 11 is a diagram illustrating an example of a data configuration of label adjustment method information in the first embodiment. [Figure 6] FIG. 4 is a diagram illustrating an example of a data configuration of criterion information according to the first embodiment. [Figure 7] FIG. 11 is a diagram illustrating an example of a data configuration of determination result information in the first embodiment. [Figure 8] FIG. 11 is a diagram illustrating an example of a data configuration of trend analysis information according to the first embodiment. [Figure 9] 1 is a flowchart illustrating an example of an overall process performed by the relationship analysis system in the first embodiment. [Figure 10] 1 is a flowchart illustrating an example of a label generation process in the first embodiment. [Figure 11] 1 is a flowchart illustrating an example of a learning model generation process in the first embodiment. [Figure 12] 11 is a flowchart illustrating an example of a determination process in the first embodiment. [Figure 13] FIG. 13 is a diagram showing an example of a screen configuration of a determination result display screen in the first embodiment. [Figure 14] FIG. 11 is a block diagram illustrating an example of a functional configuration of a relationship analysis system according to a second embodiment. [Figure 15] FIG. 11 is a diagram illustrating an example of a data configuration of determination result information in the second embodiment. [Figure 16] 13 is a flowchart illustrating an example of a determination process in the second embodiment. [Figure 17] FIG. 11 is a diagram showing an example of a screen configuration of a determination result display screen in the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment, the same components are generally designated by the same reference numerals, and repeated explanations are omitted. Note that this embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention.

[0010] There are many situations in which it is necessary to structure the relationships between objects such as people and organizations, and to identify objects that have certain characteristics by utilizing the information about these relationships.

[0011] For example, when an object is a person, it is necessary to identify a person with certain characteristics and their related parties, taking into account the relationships between people.To present one use case, when an applicant applies to register for a service provided by an institution (for example, opening an account at a financial institution or signing a real estate contract), the applicant is examined to see if he or she is appropriate to use the service.

[0012] In the preliminary stage of this screening, a risk assessment may be conducted on the applicant, and if the risk assessment determines that the applicant is at high risk of committing fraud, it is determined that further scrutiny is required in the screening of the applicant. For example, there is a risk assessment method that identifies individuals who are recognized to have characteristics in common or similar to individuals who have committed fraud in the past as high-risk individuals.

[0013] Based on the results of such pre-screening risk assessments, applicants are classified into levels of screening, such as whether to conduct a simplified or detailed screening, thereby ensuring efficient and rigorous screening.

[0014] When assessing the risk of an individual, it is necessary to pay attention to whether there are any people related to the person being assessed, the characteristics of the person being assessed and those related to them, and the structure of the relationships between the people.

[0015] For example, if a person with no related parties commits a misconduct, it is presumed that the misconduct was committed by that person's sole decision. In contrast, if a person with multiple related parties commits a misconduct, it is presumed that the misconduct was committed by a collective decision.

[0016] When it is presumed that misconduct was committed by collective decision-making, it is presumed that those involved may commit or have already committed misconduct, even if they have not been observed committing misconduct. Therefore, in such cases, it is important to check the people around the person who committed the misconduct during the investigation. In this way, investigations are conducted according to the investigation contents and investigation standards that correspond to the relationship between the person and the official. In addition, the investigation work requires transparency in judgments, and it is desirable for the grounds for judgments to be shown. EXAMPLES

[0017] In the first embodiment, a procedure for performing a risk assessment on a person, which is an example of an object, in a preliminary stage of a screening process for a person who has applied to use an arbitrary service or the like will be described.

[0018] In risk assessment for a person, it is determined whether the person has a high risk of performing a specified act or of corresponding to a specified attribute. A person who has a high risk of performing the specified act or of corresponding to the specified attribute is called a high-risk person, a person who performs the specified act alone or of corresponding to the specified attribute is called an individual HR (High Risk) person, and a person who belongs to a group that has a high risk of performing the specified act or of corresponding to the specified attribute is called a group HR person. Performing a specified act or corresponding to a specified attribute is an example of a specified characteristic.

[0019] In the examples described below, an HR person is an individual who is at high risk of committing fraud, an individual who is at high risk of committing fraud based on individual judgment, and a collective HR person is an individual who commits fraud based on an organized group of behavioral guidelines.

[0020] In this embodiment, a risk assessment is performed to determine whether a person is a single HR person, a group HR person, or a non-HR person. This risk assessment is performed, for example, prior to the user's screening of the person, and the user determines, for example, the level of the subsequent screening (whether simple or detailed) based on the risk assessment result.

[0021] 1A is a block diagram showing an example of a functional configuration of a relationship analysis system. The relationship analysis system 100, which is an example of an information processing system, includes, for example, functional units, a label adjustment unit 101, a learning unit 102, and a judgment unit 103. The relationship analysis system 100 also includes label information 104, learning-time application information 105, learning-time relationship information 106, label adjustment method information 107, learning model information 108, judgment criterion information 109, judgment result information 110, and inference-time application information 111, all of which are information generated, updated, and / or used by the functional units.

[0022] The relationship analysis system 100 has components necessary for two phases: a pre-learning process and a judgment process performed at the time of inference. The label adjustment unit 101, the learning unit 102, the label information 104, the learning-time application information 105, the learning-time relationship information 106, and the label adjustment method information 107 are used in the learning process. The learning model information 108 is generated in the learning process and used in the judgment process. The judgment unit 103, the judgment criterion information 109, and the inference-time application information 111 are used in the judgment process. The judgment result information 110 is generated in the judgment process.

[0023] The label adjustment unit 101 uses the label information 104, the learning-time application information 105, the learning-time relationship information 106, and the label adjustment method information 107 to generate multiple types of adjusted labels for the person to be learned that is included in the learning-time application information 105.

[0024] The learning unit 102 generates a learning model for each adjustment rule that generated the adjusted label, using the adjusted label and the learning-time application information 105. The judgment unit 103 uses the inference-time application information 111 and the learning model 1081 to perform judgment on the person to be analyzed that is included in the inference-time application information 111 (judgment as to whether the person corresponds to a single HR person or a group HR person).

[0025] Label information 104 indicates a label indicating whether the person to be studied corresponds to an HR person. Application information at the time of learning 105 indicates the application contents of the person to be studied (values ​​that become explanatory variables of the learning model). Relationship information at the time of learning 106 indicates relationships between the people to be studied.

[0026] The label adjustment method information 107 indicates an adjustment rule for generating an adjusted label. The learning model information 108 includes a plurality of learning models 1081 generated by the learning unit 102 and trend analysis information 1082 indicating the trend of the learning models 1081.

[0027] The judgment criteria information 109 indicates the criteria for executing judgment by the judgment unit 103. The judgment result information 110 indicates the judgment result by the judgment unit 103. The inference time application information 111 indicates the application content of the person to be analyzed.

[0028] The label adjustment process by the label adjustment unit 101 and the learning model generation process by the learning unit 102 are both performed based on label information 104, learning-time application information 105, and learning-time relationship information 106, which are generated from past performance, so they can be performed in advance before the judgment process by the judgment unit 103.

[0029] The determination process by the determination unit 103 uses the newly acquired inference time application information 111 and the previously created learning model 1081. When the label adjustment process and the learning model generation process are performed based on a large amount of past performance data, it is common to execute these processes in advance from the viewpoint of the calculation time required for these processes.

[0030] However, since the learning schedule depends on various conditions such as the frequency of data acquisition, the label adjustment process and the learning model generation process may be executed each time a request for a judgment process is received. These requirements may be set according to the application destination and are not limited to the above.

[0031] 1B is a block diagram showing an example of a hardware configuration of a computer constituting the relationship analysis system 100. The computer 1000 constituting the relationship analysis system 100 is configured by, for example, a computer 1000 having a CPU (Central Processing Unit) 1001, a memory 1002, an auxiliary storage device 1003, an input device 1004, a display device 1005, and a communication device 1006.

[0032] The CPU 1001 includes a processor and executes programs stored in the memory 1002. The memory 1002 includes a ROM (Read Only Memory), which is a non-volatile storage element, and a RAM (Random Access Memory), which is a volatile storage element. The ROM stores immutable programs (e.g., a Basic Input / Output System (BIOS)). The RAM is a high-speed, volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by the CPU 1001 and data used when the programs are executed.

[0033] The auxiliary storage device 1003 is a large-capacity non-volatile storage device such as a magnetic storage device (HDD (Hard Disk Drive)) or a flash memory (SSD (Solid State Drive)), and stores programs executed by the CPU 1001 and data used when the programs are executed. That is, the programs are read from the auxiliary storage device 1003, loaded into the memory 1002, and executed by the CPU 1001.

[0034] The input device 1004 is a device such as a keyboard or a mouse that receives input from an operator. The display device 1005 is a device such as a display device or a printer that outputs the results of program execution in a format that can be viewed by the operator.

[0035] The communication device 1006 is a network interface device that controls communication with other devices according to a predetermined protocol, and may include a serial interface such as a Universal Serial Bus (USB).

[0036] A part or all of the programs executed by the CPU 1001 may be provided to the computer 1000 from a removable medium (such as a CD-ROM or a flash memory) which is a non-transitory storage medium, or from an external computer equipped with a non-transitory storage device via a network, and may be stored in a non-volatile auxiliary storage device 1003 which is a non-transitory storage medium. For this reason, the computer 1000 may have an interface for reading data from the removable medium.

[0037] The relationship analysis system 100 is a computer system configured on a single physical computer 1000, or on multiple computers 1000 configured logically or physically, and may operate in separate threads on the same computer 1000, or may operate on a virtual computer constructed on multiple physical computer resources.

[0038] 1A. For example, the CPU 1001 functions as the label adjustment unit 101 by operating according to a label adjustment program loaded into the memory 1002, and functions as the learning unit 102 by operating according to a learning program loaded into the memory 1002. The relationships between programs and the functional units are similar for the other functional units included in the CPU 1001.

[0039] Some or all of the functions of the functional units included in the CPU 1001 may be realized by hardware such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA).

[0040] The auxiliary storage device 1003 holds various pieces of information described in Fig. 1A. Note that some or all of the various pieces of information described in Fig. 1A may be stored in the memory 1002, or may be stored in an external database connected to the computer 1000.

[0041] In this embodiment, the information used by the relationship analysis system 100 does not depend on the data structure and may be expressed in any data structure. In this embodiment, the information is expressed in a table format, but the information can be stored in a data structure appropriately selected from, for example, a list, a database, or a queue.

[0042] 2 is a diagram showing an example of the data configuration of the label information 104. The label information 104 holds, for each person ID (ID for identifying a person) of a person to be learned, an HR flag indicating whether or not the person is a high-risk person (if the HR flag is 1, the person is a high-risk person, and if it is 0, the person is not a high-risk person).

[0043] The HR flag assigned to the label information 104 is assigned in advance, for example, according to past performance. That is, in the label information 104, a person to whom the HR flag is assigned is, for example, a person who is recognized to have committed fraud in the past. The HR flag is an example of a label (before adjustment).

[0044] 3 is a diagram showing an example of the data configuration of the study-time application information 105. The study-time application information 105 holds application contents for each person ID to be studied. The application contents in the study-time application information 105 are registered in advance, for example, according to past performance. The age, qualifications, and career history of a person are all examples of application contents.

[0045] 3, processed values ​​of raw data of application contents are stored in the learning-time application information 105. Specifically, each piece of raw data is processed as a feature to be input to a machine learning model that estimates whether or not a person is a high-risk person, which is generated in a learning model generation process described later, and is stored in the learning-time application information 105.

[0046] For example, held qualifications, which is an example of application content, are turned into features by one-hot encoding processing depending on the type of qualification, and age, which is also an example of application content, is converted into a categorical variable in 10-year intervals (the value of age itself may be stored in the learning-time application information 105 without being converted). Note that the application content stored in the learning-time application information 105 can be created by any method (including unprocessed if no processing is required) as long as it has been turned into features that can be applied to a machine learning model.

[0047] In addition, the inference time application information 111 holds application details for each person ID to be analyzed, and has a similar data structure to the learning time application information 105, so an example of the data configuration of the inference time application information 111 is omitted from the illustration.

[0048] Fig. 4 is a diagram showing an example of the data configuration of the study-time relationship information 106. It holds the person IDs of study subjects having a relationship and the type of relationship. The relationships and the types of relationships in the study-time relationship information 106 are registered in advance, for example, according to past performance. In the study-time relationship information 106 in Fig. 4, for example, a person with a person ID "H0001" and a person with a person ID "H0002" have a relationship in which they belong to the same "workplace."

[0049] The data format of the learning-time relationship information 106 is not limited to the table format, and may be, for example, JSON (JavaScript (registered trademark) Object Notation) format.

[0050] 5 is a diagram showing an example of the data configuration of the label adjustment method information 107. The label adjustment method information 107 holds a label adjustment method ID for identifying a label adjustment method, information for identifying a label Gr. (Group) to which the label adjustment method ID belongs, and an adjustment rule corresponding to the label adjustment method ID.

[0051] The adjustment rules include adjustment rules for assigning an individual HR flag to the person being studied (a flag indicating whether the person is an individual HR person (if the individual HR flag is 1, the person is an individual HR person, and if the individual HR flag is 0, the person is not an individual HR person)), and adjustment rules for assigning a group HR flag to the person being studied (a flag indicating whether the person is an individual HR person (if the group HR flag is 1, the person is an individual HR person, and if the group HR flag is 0, the person is not an individual HR person)).

[0052] The individual HR flag and the group HR flag are both examples of adjusted labels. The adjusted labels are used as the objective variables in the training data.

[0053] The adjustment rule is a rule for generating an adjusted label to be assigned to a person ID of a learning subject, using at least one of the label information 104, the learning-time application information 105, and the learning-time relationship information 106.

[0054] For example, the adjustment rule for assigning an individual HR flag and the adjustment rule for assigning a collective HR flag may take into consideration at least the HR flag indicated by the label information 104 and the relationship indicated by the learning-time relationship information 106, and may further take into consideration the application content indicated by the learning-time relationship information 106. The adjustment rule can also be said to be a rule for adjusting (converting) the original HR label indicated by the label information 104 into an adjusted label, which is the objective variable on the learning data.

[0055] For example, on the relationship network generated from the learning-time relationship information 106 (a graph in which the person IDs included in the learning-time relationship information 106 are nodes and related person IDs in the learning-time relationship information 106 are connected by edges), if a high-risk person (a person ID whose HR flag indicated by the label information 104 is 1) is present within N steps from the person ID of the person to whom the adjusted label is to be assigned (i.e., a node included in the range that can be reached by tracing N edges starting from the person ID of the person to whom the adjusted label is to be assigned), and the HR flag indicated by the label information 104 of the person to whom the adjusted label is to be assigned is 0, then the population HR flag of the person to whom the adjusted label is assigned is 1. This is an example of an adjustment rule for assigning a population HR flag.

[0056] Each label group includes one or more adjustment rules. However, it is preferable that each label group includes an adjustment rule for assigning a single HR flag and an adjustment rule for assigning a collective HR flag.

[0057] The data format of the label adjustment method information 107 is not limited to the table format, and may be, for example, a JSON format or a binary format.

[0058] 6 is a diagram showing an example of the data configuration of the criterion information 109. The criterion information 109 holds, for each label group, information indicating the structure type between adjusted labels given by each adjustment rule belonging to the label group, and the positioning of the adjustment rule belonging to the label group.

[0059] The structure type indicates, for example, the relationship between adjusted labels corresponding to two adjustment rules, and includes three structure types: "A", "B", and "C".

[0060] Here, two adjustment rules x and y belong to the label Gr., and X is the set of people for whom an adjusted label indicating that they are HR persons (individually or collectively) is output as the judgment result by the learning model 1081 corresponding to adjustment rule x, and Y is the set of people for whom an adjusted label indicating that they are HR persons (individually or collectively) is output as the judgment result by the learning model 1081 corresponding to adjustment rule y.

[0061] In two adjustment rules with structure type "A", the adjusted labels corresponding to the two adjustment rules are treated as being in conflict with each other. In the two conflicting adjusted labels, a person belonging to XY or YX may exist, but a person belonging to X∩Y may not exist.

[0062] Therefore, the final judgment result of the learning model 1081 corresponding to two adjustment rules x and y with a structural type of "A" is adjusted so that X∩Y becomes an empty set. Specifically, for example, a person judged by the learning model 1081 corresponding to two adjustment rules to belong to X∩Y (i.e., to belong to both X and Y) is adjusted so that the person belongs to only either X or Y based on the probability value corresponding to the judgment result of each learning model 1081 for that person.

[0063] In addition, for individuals who are determined to belong to XY (i.e., belong only to X) from each learning model 1081 corresponding to the two adjustment rules with a structural type of "A", individuals who are determined to belong to YX (i.e., belong only to Y), and individuals who are determined to belong to neither X nor Y, the judgment results are adopted as the final judgment results as they are.

[0064] For example, it is desirable that the structure type of a label group consisting of a combination of an adjustment rule for determining an individual HR person and an adjustment rule for determining a group HR person is "A".

[0065] In two adjustment rules with a structure type of "B", the adjusted labels corresponding to the two adjustment rules are treated as being in a non-conflicting relationship and in a non-inclusive relationship. In two non-conflicting adjusted labels, a person belonging to X∩Y may exist. In addition, when the adjusted labels corresponding to the two adjustment rules are in a non-conflicting relationship, the values ​​of the two adjusted labels of a person belonging to X∩Y (i.e., the judgment results by the two learning models 1081) match.

[0066] Therefore, in the final judgment result of the learning model 1081 corresponding to the two adjustment rules x and y whose structure type is "B", X∩Y does not need to be an empty set. Specifically, for example, for a person judged to belong to X∩Y by the two learning models 1081, the judgment results for that person from each learning model 1081 match, so that judgment result is adopted as the final judgment result as it is.

[0067] In addition, in two adjustment labels that are in a non-inclusive relationship, there may be people who belong to XY and YX. Therefore, for people who are determined to belong to XY, people who are determined to belong to YX, and people who are determined to belong neither to X nor Y from the learning model 1081 corresponding to the two adjustment rules with the structure type "B", the judgment results are adopted as the final judgment results as they are.

[0068] In two adjustment rules with structure type "C", the adjusted labels corresponding to the two adjustment rules are treated as being in a non-conflict relationship and in an inclusion relationship. In two adjusted labels that are in an inclusion relationship, there must be no person who belongs to YX (Y is the "parent" in the inclusion relationship and X is the "child" in the inclusion relationship).

[0069] Therefore, the final judgment result of the learning model 1081 corresponding to two adjustment rules x and y whose structure type is "C" is adjusted to satisfy the inclusion relationship of X⊃(X∩Y)=Y. Specifically, for example, the final judgment result of a person judged to belong to YX by the learning model 1081 corresponding to two adjustment rules is adjusted so that the person belongs to either X∩Y(=Y) or belongs only to X, based on the probability value corresponding to the judgment result of each learning model 1081 for that person.

[0070] In addition, for individuals who are determined to belong to XY, individuals who are determined to belong to X∩Y (individuals whose determination results from the two learning models 1081 match), and individuals who are determined to belong to neither X nor Y from each learning model 1081 corresponding to the two adjustment rules whose structural type is "C", the determination results are adopted as the final determination results as they are.

[0071] For label Gr. with a structure type of "C", information indicating an inclusion relationship is defined as information indicating the positioning of the adjustment rule in the judgment criteria information 109. In the example of Fig. 6, the adjusted label corresponding to the adjustment rule with a label adjustment method ID of "L0008" corresponds to the "parent" in the inclusion relationship (the above-mentioned x), and the adjusted label corresponding to the adjustment rule with a label adjustment method ID of "L0009" corresponds to the "child" in the inclusion relationship (the above-mentioned y).

[0072] Furthermore, a label group may include three or more label adjustment rules. For a label group including three or more label adjustment rules, the criteria information 109 defines a structure type for each combination of two labels included in the three or more labels, and further defines the positioning of the adjustment rule for the combination of structure type "C."

[0073] The data format of the criteria information 109 is not limited to the table format, and may be, for example, a JSON format or a binary format.

[0074] Fig. 7 is a diagram showing an example of the data configuration of the determination result information 110. The determination result information 110 indicates the result of a determination process described later, and holds the determination content for the person ID of the analysis target. In the example of Fig. 7, the determination result information 110 stores a classification result indicating whether each person ID of the analysis target is an individual HR person, a group HR person, or a non-HR person (a person who is neither an individual HR person nor a group HR person), but in addition to or instead of the classification result, a probability value of being an individual HR person and a probability value of being a group HR person may be stored.

[0075] 8 is a diagram showing an example of the data configuration of the trend analysis information 1082. The trend analysis information 1082 holds an analysis result indicating the judgment tendency of the learning model 1081 corresponding to the adjustment rule. The trend analysis information 1082 holds a trend pattern ID for identifying a trend pattern, a learning model ID for identifying the learning model 1081 having the trend pattern, and information indicating the characteristics of the trend.

[0076] As trend characteristics in the trend analysis information 1082, for example, a specific judgment content and the conditions under which the specific judgment content occurs (for example, a certain characteristic amount (application content) takes a certain value or falls within a certain range) are stored.

[0077] The data format of the trend analysis information 1082 is not limited to the table format, and the data format of the criteria information 109 may be, for example, JSON format or binary format.

[0078] 9 is a flowchart showing an example of the overall processing by the relationship analysis system 100. First, a data input process is performed (S901). Specifically, for example, values ​​of the label information 104, the learning-time application information 105, the learning-time relationship information 106, and the label adjustment method information 107 are acquired by user input via the input device 1004, and these pieces of information are set. Note that some or all of these pieces of information may be transmitted from an external device connected to the relationship analysis system 100.

[0079] The label adjustment unit 101 executes a label adjustment process (S902). Details of the label adjustment process in step S902 will be described later with reference to FIG. 10. The learning unit 102 executes a learning model generation process (S903). Details of the learning model generation process in step S903 will be described later with reference to FIG. 11. The judgment unit 103 executes a judgment process (S904), and the entire process ends. Details of the judgment process in step S904 will be described later with reference to FIG. 12.

[0080] 10 is a flowchart showing an example of the label generation process in step S902. The label adjustment unit 101 constructs a relationship network by using the learning-time application information 105 and the learning-time relationship information 106 (S1001). Specifically, for example, the label adjustment unit 101 extracts person ID information included in the learning-time application information 105, determines each of the extracted person IDs as a node in a graph, and generates edges connecting nodes corresponding to related person IDs by referring to the learning-time relationship information 106.

[0081] The label adjustment unit 101 selects a label group from the label adjustment method information 107 (S1002). Specifically, for example, the label group is selected in accordance with a user input via the input device 1004.

[0082] For example, when extracting high-risk individuals, if a user wants to obtain judgment results that distinguish between group HR individuals who may commit fraudulent acts based on organized decision-making and individual HR individuals who may commit fraudulent acts based on individual judgment, the user can select a label Gr. that includes adjustment rules for assigning a group HR flag and adjustment rules for assigning an individual HR flag.

[0083] The label adjustment unit 101 selects one unselected adjustment rule from the label adjustment method information 107 among the adjustment rules included in the label group selected in step S1002 (S1003).

[0084] The label adjustment unit 101 applies the adjustment rule selected in the most recent step S1003 to each of the extracted person IDs, and generates an adjusted label corresponding to the adjustment rule for each of the extracted person IDs (S1004). In generating an adjusted label for each of the extracted person IDs, the label adjustment unit 101 uses, for example, at least one of the HR flag corresponding to the person ID indicated by the label information 104, the application content corresponding to the person ID indicated by the learning-time application information 105, and the relationship network constructed in step S1001.

[0085] A specific example of the process in step S1004 will be described below. It is assumed that the adjustment rule selected in the most recent step S1003 is an adjustment rule for assigning a collective HR flag, and is defined as follows.

[0086] Flag Judgment: If an individual satisfies both conditions 1 and 2 below, the group HR flag will be assigned a value of "1". All other individuals will be assigned a value of "0". Condition 1: A person has m or more related parties. Condition 2: There are n or more high-risk people among yourself and the people connected to you within N steps in your relationship network.

[0087] Here, m, n, and N are all predetermined integer values ​​equal to or greater than 1 (for example, m=1, n=1, n=2). If m≦n-1, then if condition 2 is satisfied, condition 1 is automatically satisfied, and therefore the determination of condition 1 may be omitted.

[0088] Another specific example of the process of step S1004 will be described below. It is assumed that the adjustment rule selected in the immediately preceding step S1003 is an adjustment rule for assigning a single HR flag, and is defined as follows.

[0089] Flag determination: A single HR flag of "1" is assigned to a person who satisfies condition 1 or condition 2, and a single HR flag of "0" is assigned to a person who satisfies neither condition 1 nor condition 2. Condition 1: Conditions 1-1 and 1-2 must be satisfied. Condition 1-1: The person has no related parties. Condition 1-2: You are a high-risk person. Condition 2: Conditions 2-1 and 2-2 must be satisfied. Condition 2-1: The person must have related parties. Condition 2-2: There are no high-risk individuals among those involved. Condition 2-3: You are a high-risk person.

[0090] In addition, the term "related person" in the conditions of the specific examples of the above two adjustment rules refers to other people who can be reached by tracing edges from the node corresponding to the person in the relationship network. Also, the term "high-risk person" in the conditions of the specific examples of the above two adjustment rules refers to a person whose HR flag in the label information 104 is "1".

[0091] Next, the label adjustment unit 101 determines whether all adjustment rules belonging to the label group selected in step S1002 have been selected in step S1003 (that is, whether adjusted labels have been generated using all adjustment rules) (S1005).

[0092] When the label adjustment unit 101 determines that there is an unselected adjustment rule among the adjustment rules belonging to the selected label Gr (S1005: NO), the process returns to step S1003. When the label adjustment unit 101 determines that all adjustment rules belonging to the selected label Gr have been selected (S1005: YES), the label adjustment process ends.

[0093] By using the above-described label adjustment process, it is possible to assign to the person ID of the learning subject an adjusted label indicating whether it is a solo HR who commits fraudulent acts based on individual judgment, and an adjusted label indicating whether it is a collective HR who commits fraudulent acts based on organized group behavioral guidelines.

[0094] 11 is a flowchart showing an example of the learning model generation process in step S903. The learning unit 102 selects learning data (S1101). The learning data is data obtained by merging the learning application information 105 and an adjusted label that is assigned by any one of the adjustment rules selected in the label adjustment process.

[0095] That is, the learning unit 102 can select the learning data by selecting one of the adjustment rules in step S1101. Note that each application content included in the learning data is an explanatory variable, and the adjusted label included in the learning data is a target variable.

[0096] The learning unit 102 uses the learning data to generate a learning model 1081 corresponding to the adjustment rule (S1102). Specifically, for example, the learning unit 102 generates the learning model 1081 by logistic regression. The learning unit 102 may also generate the learning model 1081 by using other machine learning algorithms such as a decision tree.

[0097] The learning unit 102 determines whether all the learning data have been selected in step S1101 (i.e., whether learning has been performed using the adjusted labels according to all the adjustment rules used in the label adjustment process as the objective variables) (S1103). If the learning unit 102 determines that there is unselected learning data (S1103: NO), it returns to step S1101. If the learning unit 102 determines that all the learning data have been selected (S1103: YES), it transitions to step S1104.

[0098] For example, in the label adjustment process, when adjusted labels are generated using an adjustment rule for assigning a collective HR flag and an adjustment rule for assigning an individual HR flag, the learning unit 102 generates two learning models 1081: a learning model 1081 corresponding to the adjustment rule for assigning a collective HR flag (i.e., a learning model 1081 with the collective HR flag as the objective variable), and a learning model 1081 corresponding to the adjustment rule for assigning an individual HR flag (i.e., a learning model 1081 with the individual HR flag as the objective variable).

[0099] The learning unit 102 analyzes differences in trends of the judgment results between the generated learning models (S1004), and stores the analysis results in trend analysis information 1082. A specific example of the trend analysis process will be described below.

[0100] The learning unit 102, for example, inputs application information for verification prepared in advance (information having a data structure similar to that of the application information at the time of learning 105, and the application information at the time of learning 105 itself may be the application information for verification) into each learning model 1081 and obtains the judgment results from each learning model 1081.

[0101] The learning unit 102 identifies sample data in which a predetermined difference in the judgment results of each learning model 1081 appears (for example, data of a person in the application information for verification in which the difference between the probability value based on one learning model and the probability value based on the other learning model 1081 is equal to or greater than a predetermined value), generates data that is close to the identified sample data in the feature space (for example, data in which the value of only one type of feature out of multiple types of features (application content) has been changed within a predetermined amount), and inputs the generated data into each learning model 1081 to obtain additional judgment results.

[0102] The learning unit 102 extracts conditions under which differences in judgment results occur from the judgment results in the reinforced verification information (the judgment result obtained from the application information for verification and the additional judgment result) (for example, when the value of feature X is within the range of Y to Z, the judgment result based on the first learning model indicates high risk and the judgment result based on the second learning model does not indicate high risk, etc.), and stores the extracted conditions as features in each learning model of the trend analysis information 1082.

[0103] In order to simplify the conditions to be more readable, the learning unit 102 may extract the above conditions using the learning results of a simple approximation model, such as a linear regression formula or a decision tree regression in which the tree depth is limited to suppress complexity. The means for analyzing the difference in trends are not limited to those described above.

[0104] After executing the trend analysis in step S1104, the learning unit 102 ends the learning process.

[0105] The above-mentioned learning model generation process makes it possible to generate a learning model for determining whether a person is a solo HR who commits fraudulent acts based on individual judgment, and a learning model for determining whether a person is a collective HR who commits fraudulent acts based on organized behavioral guidelines within a group.

[0106] 12 is a flowchart showing an example of the judgment process in step S904. Before the process of step S904 starts, judgment criteria information 109 is set. The judgment unit 103 acquires inference time application information 111 (S1201). The judgment unit 103 may acquire the inference time application information 111 by user input via the input device 1004, or the inference time application information 111 may be transmitted from an external device connected to the relationship analysis system 100. The inference time application information 111 is an example of evaluation target data.

[0107] The judgment unit 103 selects one of the unselected learning models 1081 (S1202). The judgment unit 103 inputs the application details of each person included in the application information at the time of inference 111 to the learning model 1081 selected in step S1202, and obtains a judgment result for each person (S1203). If the selected learning model 1081 is a logistic regression model, a probability value (a value in the range from 0 to 1) of being a (single or group) HR person is set as the judgment result in step S1203.

[0108] The determination unit 103 determines whether all learning models 1081 have been selected (S1204). If the determination unit 103 determines that there is an unselected learning model 1081 (S1204: NO), the process returns to step S1202. If the determination unit 103 determines that all learning models 1081 have been selected (S1204: YES), the process proceeds to step S1205.

[0109] For example, if there are two learning models, namely, a learning model 1081 corresponding to an adjustment rule that assigns a group HR flag, and a learning model 1081 corresponding to an adjustment rule that assigns an individual HR flag, the judgment unit 103 executes two judgment processes and obtains judgment results for each person included in the application information 111 at the time of inference using each of the two learning models 1081.

[0110] The determination unit 103 executes a final determination using each of the determination results of the learning model 1081 and the determination criterion information 109, and stores the final determination result in the determination result information 110 (S1205). After executing the final determination process of step S1205, the determination unit 103 ends the determination process.

[0111] A specific example of the final determination process in step S1205 will be described below.

[0112] For example, suppose that judgment results are obtained from a first learning model and a second learning model that each output a probability value of being an HR person (probability value of being a group HR person or probability value of being an individual HR person), and that the adjustment rules corresponding to the first learning model and the second learning model each belong to the same label Gr.

[0113] If the probability value for a certain person according to the first learning model exceeds a predetermined judgment threshold, the first learning model judges the person to be an HR person (a group HR person or an individual HR person), and if the probability value for the certain person according to the second learning model exceeds the predetermined judgment threshold, the second learning model judges the person to be an HR person (a group HR person or an individual HR person) (the judgment threshold for the first learning model and the judgment threshold for the second learning model may be different).

[0114] At this time, the determination unit 103 performs a final determination on the person using the structure type between the adjustment rules corresponding to the first learning model and the second learning model.

[0115] If the structural type is "A" and the person is determined to be an HR person by both the first learning model and the second learning model, the judgment unit 103 determines the judgment result corresponding to the higher value between the probability value that is the judgment result by the first learning model and the probability value that is the judgment result by the second learning model as the final judgment result.

[0116] If the structural type is "C" and the learning model corresponding to the child (among the first learning model and the second learning model) determines that the person is an HR person, and the learning model corresponding to the parent determines that the person is not an HR person, the judgment unit 103 determines the judgment result corresponding to the higher value of the probability value that is the judgment result by the learning model corresponding to the child and the probability value that is the judgment result by the learning model corresponding to the parent as the final judgment result.

[0117] In combinations other than those of the structural types and judgment results of each learning model described above, the judgment unit 103 simply adopts the judgment results of whether the person is an HR person obtained by each of the first learning model and the second learning model as the final judgment result.

[0118] An example in which more specific values ​​and learning models are applied to the specific example of the final judgment process described above will be described. For example, it is assumed that the first learning model outputs a probability value of being a group HR person, the second learning model outputs a probability value of being an individual HR person, the adjustment rules corresponding to the first learning model and the second learning model belong to the same label Gr., and the structure type between these adjustment rules is "A". Furthermore, it is assumed that the first learning model obtains a judgment result in which the probability value of a certain person being a group HR person is 0.7, and the second learning model obtains a judgment result in which the probability value of the person being an individual HR person is 0.4.

[0119] In this case, if the specified judgment threshold is 0.5, the final judgment result that the person corresponds to a group HR person is obtained from the judgment result by the first learning model that shows a probability value exceeding the specified judgment threshold.

[0120] On the other hand, in the above assumption, if the second learning model gives a result that the probability value of the person being an individual HR is 0.6, the person will be classified as both a group HR person and an individual HR person. However, since the structural type between the adjustment rules is "A," the adjusted labels output by the learning models corresponding to these adjustment rules are in a conflicting relationship, so the higher of the probability value of the person being a group HR person (0.7) and the probability value of the person being an individual HR (0.6) will be adopted. In other words, the final judgment result is that the person is a group HR person.

[0121] In addition, in a learning model that outputs adjusted labels that are non-conflicting, such as structural types "B" and "C," if the judgment result in one learning model exceeds the judgment threshold, the judgment threshold for the judgment result in the other learning model may be lowered by a predetermined value (for example, if a probability value of 0.6 is obtained as the judgment result by the first learning model against a judgment threshold of 0.5, the judgment threshold for the second learning model may be lowered to 0.4).

[0122] In addition, the judgment unit 103 stores, for example, the value "single HR person" for a person ID whose final judgment result corresponds to a single HR person but not a group HR person, "group HR person" for a person ID whose final judgment result does not correspond to a single HR person but corresponds to a group HR person, and "non-HR person" for a person ID that corresponds to neither a single HR person nor a group HR person in the judgment content column of the judgment result information 110.

[0123] The above-mentioned judgment process allows an optimal and consistent final judgment result to be obtained from the judgment results obtained by each learning model for each person ID to be analyzed. That is, it is possible to judge with high accuracy whether the person ID to be analyzed corresponds to a single HR person who commits fraudulent acts based on individual judgment, a group HR person who commits fraudulent acts based on organized behavioral guidelines in a group, or neither.

[0124] 13 is a diagram showing an example of the screen configuration of the determination result display screen. The determination result display screen 1300 is displayed on the display device 1005 by the determination unit 103 after the determination process in step S904 is completed, for example.

[0125] The judgment result display screen 1300 includes, for example, a judgment result display area 1301 and a trend analysis result display area 1302. A part or all of the judgment result information 110 is displayed in the judgment result display area 1301. Specifically, for example, the result of the final judgment in step S1205 is displayed in the judgment result display area 1301.

[0126] The trend analysis result display area 1302 displays a part or all of the trend analysis information 1082. Specifically, for example, the trend analysis result display area 1302 displays the conditions indicated by the trend characteristics obtained from the generated learning model, and the types and value ranges of application contents used for the conditions.

[0127] The judgment result display screen 1300 allows the user to easily check the final judgment result and the characteristics of the learning model 1081 used for the judgment. In the trend analysis result display area 1302, in particular, it is possible to improve the visibility for the user by displaying only the characteristics of the trend pattern that meets the conditions among the analyzed person, displaying the characteristics at the top of the list, or highlighting the characteristics. EXAMPLES

[0128] In this embodiment, differences from the first embodiment will be mainly described, and descriptions of similarities to the first embodiment will be omitted as appropriate. The relationship analysis system 100 of the second embodiment differs from the second embodiment in that it holds relationship information about a person to be analyzed, and the relationship information is further used in the determination process.

[0129] 14 is a block diagram showing an example of a functional configuration of the relationship analysis system 100. The relationship analysis system 100 of this embodiment differs from the first embodiment in that it further has inference time relationship information 1412, further uses the inference time relationship information 1412 in the determination by the determination unit 103, and generates determination result information 1410 instead of the determination result information 110.

[0130] The inference time relationship information 1412 holds relationship information related to the person ID of the analysis target. Since the data structure of the inference time relationship information 1412 is similar to the data structure of the learning time relationship information 106, an example of the data configuration of the inference time relationship information 1412 is omitted from the illustration.

[0131] 15 is a diagram showing an example of the data configuration of the determination result information 1410. The determination result information 1410 differs from the determination result information 110 of the first embodiment in that it further includes related party information of the person to be analyzed.

[0132] In the example of determination result information 1410 in Figure 15, the related person information "{Step 1: [H0201, H0203]}" of the person with person ID "H0102" indicates that in the relationship network, there are two HR persons (single HR persons or group HR persons) one step ahead of the person with person ID "H0102", and the person IDs of the two HR persons are "H0201" and "H0203", respectively.

[0133] The process executed in this embodiment will be described below. The overall process in this embodiment is similar to that in the first embodiment shown in Fig. 9, but the determination process in step S904 in this embodiment is different from that in the first embodiment.

[0134] 16 is a flowchart showing an example of the determination process in step S904. Instead of the process in step S1201, the determination unit 103 executes a process of acquiring the inference time application information 111 and the inference time relationship information 1412 (S1601). The determination unit 103 may acquire the inference time application information 111 and the inference time relationship information 1412 by a user input via the input device 1004, or the inference time application information 111 and the inference time relationship information 1412 may be transmitted from an external device connected to the relationship analysis system 100.

[0135] When the determination unit 103 determines that all learning models 1081 have been selected (S1204: YES), it constructs a relationship network (S1602) using the inference time application information 111 and the inference time relationship information 1412. Specifically, for example, the determination unit 103 extracts person ID information included in the inference time application information 111, determines each of the extracted person IDs as a node in a graph, and generates edges connecting nodes corresponding to related person IDs by referring to the inference time relationship information 1412.

[0136] Following the process of step S1602, the determination unit 103 executes a final determination process of step S1205. However, in the final determination process of step S1205 in the second embodiment, the determination unit 103 further uses the relationship network constructed in step S1602.

[0137] Specifically, for example, after performing the final judgment process in Example 1, the judgment unit 103 stores, for each person ID to be analyzed, the person IDs of other HR persons (single HR persons or group HR persons) existing on the relationship network including the person, and the number of steps to the other HR persons on the relationship network in the judgment result information 1410.

[0138] Furthermore, for example, the judgment unit 103 may store in the judgment result information 1410, for each person ID being analyzed, information indicating whether other HR persons existing on the relationship network including the person are individual HR persons or group HR persons, and the relationship type corresponding to the edge to the other HR persons.

[0139] In this way, the determination unit 103 can supplement information that is more useful in the screening by adding information about HR persons in the relationship network to each person to be analyzed in the final judgment. In particular, for example, when a person belongs to a relationship network that includes other people who are determined to be group HR persons, the user can consider that these people have a common background, and thus it can be considered desirable to perform a coordinated screening of these people.

[0140] In addition, if there are other HR persons in the relationship network that includes a person who has been determined to be a sole HR person in the final determination, the user can infer that the sole HR person may be a collective HR person who has a common background with the other HR persons.

[0141] 17 is a diagram showing an example of the screen configuration of the determination result display screen 1300. The determination result display screen 1300 of this embodiment further includes a relational network display area 1701. Furthermore, a part or all of the determination result information 1410 is displayed in the determination result display area 1301 of this embodiment.

[0142] A graph showing the relationship network constructed in step S1602 is displayed in relationship network display area 1701. Black nodes in the graph showing the relationship network displayed in relationship network display area 1701 indicate people who have been determined to be HR persons (single HR persons or group HR persons) in the final assessment, and white nodes indicate people who have been determined to be neither single HR persons nor group HR persons in the final assessment.

[0143] In addition, for each black node in the graph showing the relationship network displayed in the relationship network display area 1701, information indicating whether the person corresponding to the node is an individual HR person or a group HR person, as well as the probability value of being an individual HR person or a group HR person may also be displayed.

[0144] In addition, since the overall picture of the relationship network displayed in the relationship network display area 1701 may become complex and large-scale, for example, the person ID displayed in the judgment result display area 1301 may be selectable by user input, and a relationship network within a range of a predetermined number of steps (the predetermined number of steps may be specifiable by the user) centered on the selected person ID may be displayed.

[0145] The judgment result display screen 1300 of Example 2 enables the user to easily check the final judgment result, the characteristics of the learning model 1081 used in the judgment, as well as the relationship network to which the person being analyzed belongs and the HR persons included in the relationship network.

[0146] 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 to clearly explain the present invention, and is not necessarily limited to those having all the configurations described. It is also 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. It is also possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

[0147] 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 the program, table, file, etc. that realizes each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

[0148] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0149] 100 Relationship analysis system 100, 101 Label adjustment unit, 102 Learning unit, 103 Judgment unit 103, 104 Label information, 105 Application information at learning, 106 Relationship information at learning, 107 Label adjustment method information, 108 Learning model information, 109 Judgment criteria information, 110 Judgment result information, 111 Application information at inference, 1000 Computer, 1001 CPU, 1002 Memory, 10003 Auxiliary storage device, 1004 Input device, 1005 Display device, 1006 Communication device, 1081 Learning model, 1082 Trend analysis information, 1300 Judgment result display screen, 1410 Judgment result information, 1412 Application information at inference

Claims

1. An information processing system, A processor and a memory, The memory includes: Label information indicating a label for a predetermined feature of the learning object; Relationship information indicating whether or not there is a relationship between the learning objects; Explanatory variable information indicating values ​​of explanatory variables of the learning object and the analysis object; Label adjustment method information indicating a plurality of adjustment rules for assigning a plurality of types of adjusted labels relating to the relationship between objects and the predetermined features to the learning object; and retaining judgment criteria information indicating a relationship between the plurality of types of adjusted labels; The processor, assigning the plurality of types of adjusted labels to the learning object based on the relationship indicated by the relationship information, the label indicated by the label information, and the plurality of adjustment rules; generating a learning model for each of the multiple types of adjusted labels that have been assigned, based on the values ​​of the explanatory variables of the learning object and the adjusted labels, which outputs the value of the adjusted label when the explanatory variables are input; assigning values ​​of each of the plurality of types of adjusted labels to the object to be analyzed based on values ​​of explanatory variables of the object to be analyzed and the generated learning model; An information processing system that determines an adjusted label to be adopted as a final judgment result for the analysis target object from the multiple types of adjusted labels based on each of the values ​​assigned to the analysis target object and the relationship indicated by the judgment criteria information.

2. 2. The information processing system according to claim 1, The plurality of types of adjusted labels include a first adjusted label indicating whether the object alone has the predetermined characteristic; and a second adjusted label indicating whether the object belongs to a population having the predetermined characteristic.

3. 3. The information processing system according to claim 2, The processor, generating a relational network in which the learning objects are nodes and related learning objects are connected by edges with reference to the relational information; Identifying a learning object whose label in the label information indicates that it has the predetermined characteristic; An information processing system that assigns the first adjusted label, indicating that the identified learning object alone has the specified characteristic, to those of the identified learning object whose relational network does not include other learning objects or does not include other identified learning objects.

4. 3. The information processing system according to claim 2, The processor, generating a relational network in which the learning objects are nodes and related learning objects are connected by edges with reference to the relational information; Identifying a learning object whose label in the label information indicates that it has the predetermined characteristic; An information processing system that assigns a second adjusted label indicating that the identified learning object has the specified characteristic as a group to those learning objects for which a first predetermined number of other learning objects exist in the relational network including the learning object itself, and for which other learning objects reachable in the relational network by tracing a second predetermined number of edges include a third predetermined number or more of the identified learning objects.

5. 3. The information processing system according to claim 2, the study object and the analysis object are people, The predetermined characteristic indicates whether the person is a high-risk person who is at high risk of performing a predetermined act or falling under the predetermined attribute, The first adjusted label indicates whether the person is a high-risk individual who is at high risk of performing the predetermined act alone or who is at high risk of meeting the predetermined attribute alone, An information processing system, wherein the second adjusted label indicates whether the person is a high-risk group member belonging to a group that has a high risk of committing the specified behavior or a high risk of meeting the specified attributes.

6. 6. The information processing system according to claim 5, A display device is provided, The processor, An information processing system that displays on the display device information indicating whether an analysis target object for which the first adjusted label is adopted as the final judgment result corresponds to the individual high-risk individual, and information indicating whether an analysis target object for which the second adjusted label is adopted as the final judgment result corresponds to the group high-risk individual.

7. 2. The information processing system according to claim 1, the relationship between the adjusted labels indicated by the criterion information includes an adversarial relationship; a value corresponding to each of the plurality of types of adjusted labels is a probability value having the predetermined feature based on a relationship between objects; The processor, assigning the probability value corresponding to each of the plurality of types of adjusted labels to the object to be analyzed based on the values ​​of explanatory variables of the object to be analyzed and the generated learning model; an information processing system which determines that all of the assigned probability values ​​indicate that the object to be analyzed has the specified characteristic based on a result of comparing the assigned probability values ​​with a predetermined threshold value, and if the multiple types of adjusted labels are in a conflicting relationship, adopts the adjusted label corresponding to the maximum probability value as the final judgment result for the object to be analyzed.

8. 2. The information processing system according to claim 1, the relationship between the adjusted labels indicated by the criterion information includes an inclusion relationship; the criterion information indicates an adjusted label corresponding to a parent and an adjusted label corresponding to a child in the inclusion relationship, a value of each of the plurality of types of adjusted labels is a probability value of having the predetermined feature based on a relationship between objects; The processor, assigning the probability value corresponding to each of the plurality of types of adjusted labels to the object to be analyzed based on the values ​​of explanatory variables of the object to be analyzed and the generated learning model; an information processing system in which, when a comparison between a predetermined threshold value and the assigned probability values ​​determines that the probability value corresponding to the adjusted label corresponding to the child in the inclusion relationship indicates that the object to be analyzed has the predetermined characteristic, and that the probability value corresponding to the adjusted label corresponding to the parent in the inclusion relationship indicates that the object to be analyzed does not have the predetermined characteristic, the adjusted label corresponding to the maximum probability value is adopted as the final determination result for the object to be analyzed.

9. 2. The information processing system according to claim 1, The processor, Identifying samples of explanatory variable values ​​that exhibit a predetermined difference in output values ​​between the generated learning models; An information processing system that generates trend analysis information indicating conditions regarding the values ​​of explanatory variables and values ​​output by the learning model when the conditions are satisfied, based on the identified sample and nearby values ​​of the sample based on specified conditions.

10. 10. The information processing system according to claim 9, A display device is provided, The processor displays the generated trend analysis information on the display device.

11. 2. The information processing system according to claim 1, A display device is provided, The relationship information further indicates whether or not there is a relationship between the analysis target objects; The processor, generating a relational network in which the analysis target objects are nodes and the analysis target objects that are related to each other are connected by edges, with reference to the relational information; An information processing system that displays, on the display device, information indicating the adjusted label adopted as the final judgment result for the analysis target object and the generated relational network.

12. An information processing method by an information processing system, comprising: The information processing system includes a processor and a memory. The memory includes: Label information indicating a label for a predetermined feature of the learning object; Relationship information indicating whether or not there is a relationship between the learning objects; Explanatory variable information indicating values ​​of explanatory variables of the learning object and the analysis object; Label adjustment method information indicating a plurality of adjustment rules for assigning a plurality of types of adjusted labels relating to the relationship between objects and the predetermined features to the learning object; and retaining judgment criteria information indicating a relationship between the plurality of types of adjusted labels; The information processing method includes: the processor assigns the plurality of types of adjusted labels to the learning object based on the relationship indicated by the relationship information, the label indicated by the label information, and the plurality of adjustment rules; the processor generates a learning model for each of the multiple types of adjusted labels that have been assigned, based on the values ​​of explanatory variables of the learning object and the adjusted labels, which outputs the value of the adjusted label when the explanatory variables are input; the processor assigns values ​​of each of the plurality of types of adjusted labels to the object to be analyzed based on values ​​of explanatory variables of the object to be analyzed and the generated learning model; An information processing method in which the processor determines an adjusted label to be adopted as a final judgment result for the analysis target object from the multiple types of adjusted labels based on each of the values ​​assigned to the analysis target object and the relationship indicated by the judgment criteria information.

Images