JP7713428B2 - Examination device, method and program - Google Patents

Description

This disclosure relates to technology for assisting in the screening of users.

A loan information providing device has been proposed that receives notifications of store transaction information, etc. from a terminal device, inputs the transaction information, etc. into a trained model to obtain classification results, calculates a loan score based on this classification result, generates loan information based on the loan score and loan condition information of each financial institution, and notifies the terminal device of the loan information of each financial institution and recommended information to be notified to businesses (see Patent Document 1).

JP 2020-160949 A

Conventionally, in screening for loans, credit transactions, etc., determining a user's credit score using a machine learning model has been considered important from the perspective of optimizing screening. In addition, a technology has been proposed that selects a combination of input data to a model that results in a high loan score as recommendation information (see Patent Document 1). However, while the technology proposed in the past has been effective to a certain extent in terms of optimizing screening, there is room for improvement in terms of appropriately screening users who have insufficient accumulated attribute data such as credit information.

In view of the above problems, the present disclosure aims to provide proper screening even for users who have insufficient accumulated attribute data.

An example of the present disclosure is an assessment device including: a first score acquisition means for acquiring a first score for a user based on an output obtained by inputting input data related to the user into a first machine learning model; a second score acquisition means for acquiring a second score for the user based on an output obtained by inputting input data related to the user into a second machine learning model; a user segment identification means for identifying a user segment to which the target user belongs by performing segmentation of a user group including the target user; and an assessment result determination means for determining an assessment result for the target user based on the first score and/or second score of the target user according to the user segment to which the target user belongs.

The present disclosure can be understood as an examination device, a system, a method executed by a computer, or a program executed by a computer. The present disclosure can also be understood as such a program recorded on a recording medium readable by a computer or other device, machine, etc. Here, a recording medium readable by a computer, etc. refers to a recording medium that stores information such as data and programs through electrical, magnetic, optical, mechanical, or chemical action and can be read by a computer, etc.

This disclosure makes it possible to provide fair screening even to users who have insufficient accumulated attribute data.

1 is a schematic diagram showing a configuration of an information processing system according to an embodiment. 1 is a diagram showing an outline of the functional configuration of a screening device according to an embodiment; FIG. 2 is a diagram showing an overview of a screening process in an embodiment. FIG. 1 is a simplified diagram illustrating the concept of a decision tree of a machine learning model employed in an embodiment. 1 is a flowchart illustrating a flow of a machine learning process according to an embodiment. 11 is a flowchart showing the flow of a screening process according to the embodiment. FIG. 2 is a diagram showing an outline of the functional configuration of a review device for variations.

Below, embodiments of the screening device, method, and program according to the present disclosure will be described with reference to the drawings. However, the embodiments described below are merely examples, and the screening device, method, and program according to the present disclosure are not limited to the specific configurations described below. In implementing the present disclosure, a specific configuration according to the mode of implementation may be appropriately adopted, and various improvements and modifications may be made. In the technology according to the present disclosure, at least a portion of the configurations in the embodiments and variations described below may be appropriately adopted with each other.

In this embodiment, an embodiment of the technology disclosed herein will be described in a case where the technology disclosed herein is implemented in a system that provides loan screening for users. However, the technology disclosed herein can be widely used as a technology for supporting any screening for users, and the application of the present disclosure is not limited to the examples shown in the embodiment.

<System Configuration>
1 is a schematic diagram showing the configuration of an information processing system according to this embodiment. In the information processing system according to this embodiment, a screening device 1 and one or more service providing systems 5 are communicably connected to each other. A user is a user of a service provided by the service providing system 5, and receives the service by accessing the service providing system 5 from a user terminal.

The review device 1 is a computer equipped with a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage device 14 such as an EEPROM (Electrically Erasable and Programmable Read Only Memory) or a HDD (Hard Disk Drive), a communication unit 15 such as a NIC (Network Interface Card), etc. However, the specific hardware configuration of the review device 1 can be omitted, replaced, or added as appropriate depending on the embodiment. Furthermore, the review device 1 is not limited to a device consisting of a single housing. The examination device 1 may be realized by multiple devices using so-called cloud or distributed computing technology, etc.

The screening device 1 screens transactions related to each user and provides the screening results to the service provision system 5. Depending on the results of the screening of a given user provided by the screening device 1, the service provision system 5 can decide whether or not to provide a service such as a loan to that user (hereinafter, the "target user") (whether or not to conduct a transaction).

The service providing system 5 is a computer equipped with a CPU, ROM, RAM, a storage device, a communication unit, an input device, an output device, etc. (not shown). Furthermore, these systems and terminals are not limited to devices consisting of a single housing. These systems and terminals may be realized by multiple devices using so-called cloud or distributed computing technology, etc.

The services provided by the service providing system 5 include, for example, loan screening services, credit card/deferred payment services, online shopping services, online reservation services, electronic money payment services, operation center services, map information services, etc. Note that "deferred payment" is not limited to services known as Buy Now Pay Later (BNPL), and may include any purchase of goods/services with deferred payment.

The services provided by the service providing system 5 are not limited to the examples in this embodiment. The service providing system 5 notifies the review device 1 of the user's attribute data when providing the service. Here, the user's attribute data includes the service usage history data of the user. The content of the service usage history data varies depending on the content of the service, and may include, for example, history data of the user's location information, payment history data of credit card usage amount/postpaid payment usage amount, electronic money usage history data, transaction history data (including purchase history data of products, etc.), reservation history data, operation history data for the user from the operation center, etc.

Here, the attribute data includes data expressed in a data format such as a score (e.g., a continuous value between 0 and 1) or a label (e.g., a binary value corresponding to presence/absence or yes/no). However, the format of the attribute data is not limited to the examples in this disclosure. In addition, the attribute data may include, for example, the usage status of online services, the usage status of electronic value including points, the cancellation rate in online services, and data indicating default (non-performance of debt) in deferred payment.

The attribute data may include factual attribute data and inferred attribute data. Here, factual attribute data is attribute data that can be confirmed as factual for a user based on user-provided data obtained by providing it from the user himself or on history data collected about the user. Inferred attribute data is attribute data that is obtained by inferring it by inputting user-provided data, history data, factual attribute data, etc. into a VAE (Variational Autoencoder), for example.

The VAE used here expresses the values input to the encoder (here, training data including fact attribute data) in a different format by inputting the latent vector output by the encoder in the first half of the VAE to the decoder in the second half of the VAE. Examples of user-provided data include registration data including the name, email address, telephone number, address, place of employment, place of study, etc. registered by the user himself, and data obtained as a result of the user himself answering a questionnaire, etc. Examples of history data include the above-mentioned usage history data of the electronic commerce service provided by the service providing system 5. It is preferable that the fact attribute data is data obtained by converting the above-mentioned user-provided data and history data into a data format suitable for marketing and/or analysis purposes. For example, examples of fact attribute data that can be obtained based on usage history data include genres/categories and brands of products/services frequently used by the user, as well as commercial locations, recreational areas, and tourist spots frequently visited by the user. That is, in this embodiment, the attribute data includes the user's personality, behavioral tendencies, user persona, etc. estimated or predicted using machine learning technology.

In this embodiment, a weight is set for each attribute data. The weight indicates the degree of correlation between the attribute data and the score when the attribute data is used to calculate the score, and each time the appropriateness of the score is evaluated by the machine learning unit 27 described later, the parameters of the model are adjusted so that the score becomes a more appropriate value. The weight corresponding to each attribute data corresponds to the weight corresponding to each node (each regression tree) in a model for score calculation such as a decision tree model described later, for example, and is determined appropriately in the process of calculating the score. Note that the score is determined based on the weight of each node, for example.

Here, the attribute data group may include demographic attributes, behavioral attributes, or psychographic attributes. Demographic attributes are, for example, the user's gender, family structure, age, etc., and behavioral attributes may be based on service usage history data, such as whether or not cash advances are used, whether or not revolving payments are used, deposit and withdrawal history for a specific account, commercial transaction history for some product/service including gambling or lottery (may include online transaction history in an online marketplace, etc.), user movement history using location information or place information, etc., and psychographic attributes are, for example, preferences for gambling or lottery. However, the available user attributes are not limited to the examples in this embodiment. For example, "time required for operation (calls, etc.)" and "credit card usage amount/deferred payment usage amount" from an operation center service, etc. may also be used as attributes. Note that attributes similar to demographic attributes may be attributes estimated based on attributes based on user-provided data or history data. Similarly, attributes similar to behavioral attributes may be inferred attributes based on user-provided data or attributes based on historical data. Psychographic attributes may be attributes based on user-provided data, including, for example, the result of user input.

Figure 2 is a diagram showing an outline of the functional configuration of the review device 1 according to this embodiment. The review device 1 functions as a review device equipped with an attribute determination unit 21, a first score acquisition unit 22, a second score acquisition unit 23, a user segment identification unit 24, a correlation determination unit 25, a review result determination unit 26, and a machine learning unit 27, by a program recorded in the storage device 14 being read into the RAM 13 and executed by the CPU 11, which controls each piece of hardware equipped in the review device 1. Note that in this embodiment and the variations described below, each function equipped in the review device is executed by the CPU 11, which is a general-purpose processor, but some or all of these functions may be executed by one or more dedicated processors.

The attribute determination unit 21 determines factual attribute data that can be confirmed as fact for the user based on user-provided data provided by the user and/or the user's history data. In this embodiment, the attribute determination unit 21 determines factual attribute data for the user using a method such as aggregating user-provided data and/or history data, determining the relevant attribute by referring to other data such as a map, or using the user-provided data and/or history data as is. Note that, although this embodiment employs a method of determining factual attribute data for the user based on user-provided data and/or the user's history data, factual attribute data for the user may be obtained by other methods.

The attribute determination unit 21 determines estimated attribute data estimated for a user based on input data including at least one or more fact attribute data determined for the user. In this embodiment, the attribute determination unit 21 may determine estimated attribute data based on an output value obtained by inputting input data including one or more fact attribute data related to the user into an attribute estimation model, which is a machine learning model. In this embodiment, the output value from the attribute estimation model is a value indicating the probability that the user has a predetermined attribute, and the attribute determination unit 21 determines that the user has the attribute when the output value obtained from the attribute estimation model is within a predetermined range. When it is determined that the user has the predetermined attribute, the attribute determination unit 21 sets the label of the attribute data estimated for the user to a value indicating the presence or absence of the attribute or the type of the attribute. The attribute data may be represented by a score instead of a label. In this case, the attribute determination unit 21 sets a value indicating the degree (probability) that the estimated attribute can be applied to the score of the estimated attribute data estimated for the user. The degree may be the output value of the attribute estimation model.

The first score acquisition unit 22 acquires a first score for a user based on an output (e.g., a score normalized/standardized with 0 as the minimum value and 1 as the maximum value) obtained by inputting input data related to the user into a first machine learning model. More specifically, the first score acquisition unit 22 acquires a first score for a user based on an output obtained by inputting input data related to the user into a first machine learning model generated by learning the relationship between the attributes of the user in a first category transaction (first type of transaction) and the creditworthiness of the user. In the present disclosure, a first category transaction refers to a transaction that is for a period of a predetermined length or more and/or for an amount of a predetermined length or more. That is, examples of transactions in the first category include relatively long-term and/or high-value transactions, such as long-term loans or high-value loans.

In this embodiment, the first machine learning model is generated and/or updated using training data in which attribute data including the status of other transactions of the same type as the first category transaction (e.g., other long-term loans or high-value loans, etc.) is used as input values, and a value indicating the creditworthiness of the user (first score) is used as output values. The first machine learning model is a machine learning model for screening, which outputs screening results for transactions in the first category, such as whether or not a loan may be granted, or a result that corresponds to the probability of default. For this reason, the first machine learning model may appropriately adopt the types of input data used in the previous high-value loan screening model.

The input data to the first machine learning model includes a group of attribute data including the above-mentioned fact attribute data and/or estimated attribute data. Examples of the input data include annual income, type of employment, date of birth, living situation (length of residence (e.g., in months), length of employment (e.g., in months), borrowing status/repayment status (of other loans already contracted, etc.), etc.).

The second score acquisition unit 23 acquires a second score for the user based on an output (e.g., a score normalized/standardized with 0 as the minimum value and 1 as the maximum value) obtained by inputting input data related to the user into a second machine learning model. More specifically, the second score acquisition unit 23 acquires a second score for the user based on an output obtained by inputting input data related to the user into a second machine learning model generated by learning the relationship between the attributes of the user and the risk of the user in a second category transaction (second type of transaction) different from the first category transaction. In this disclosure, a transaction in the second category refers to a transaction that is shorter in duration and/or smaller in amount than transactions in the first category. That is, examples of transactions in the second category include transactions that are relatively short-term and/or small in amount, such as postpaid settlement, credit card use, short-term loans, or small loans.

In this embodiment, the second machine learning model is generated and/or updated using training data in which attribute data including the status of other transactions of the same type as the second category transaction (e.g., other deferred payment, credit card use, short-term loans, small loans, etc.) is used as an input value, and a value (second score) indicating a risk based on the probability that the user will default without making a payment (without the debt being collected from the user) is used as an output value. There is no limitation on the method of expressing risk, and various indices may be adopted. For example, the risk can be expressed using the probability of default. Indices other than the probability of default may be adopted as an indicator of risk. For example, the risk may be divided into classes (ranks), and the class (rank) may be used as an indicator. That is, the second machine learning model is a machine learning model for screening that outputs, for example, a score that changes depending on the risk that the target user will not properly settle the deferred payment when the target user purchases goods and/or services with a deferred payment, which is a transaction of the second category, or a score related to the probability of default when the target user purchases goods and/or services with a credit card payment, which is also a transaction of the second category.

For this reason, the second machine learning model may appropriately adopt the type of input data in a screening model for small loans such as credit card screening. In addition, the second machine learning model may adopt a BNPL (Buy-Now-Pay-Later) screening model. When a BNPL screening model is adopted as the second machine learning model, some kind of credit score or user attributes may be used to determine the final screening result in this disclosure.

In the second machine learning model, as in the first machine learning model, a group of attribute data including fact attribute data and/or estimated attribute data can be appropriately adopted as input data. Examples of input data include annual income, living situation (industry of employment, working style (whether remote work is possible, etc.)), credit card usage, etc.

The user segment identification unit 24 performs segmentation of a user group including the target user to identify a user segment to which the target user belongs. Here, segmentation is performed by referring to the attribute data of users included in the user group and placing users having common attribute data in the same user segment. For example, users having the same annual income, business type, gender, and age group are placed in the same user segment. Here, the user attributes referred to during segmentation are not limited, but for example, by using attributes common between the input data of the first machine learning model and the input data of the second machine learning model (in other words, attributes used as input data in both the first machine learning model and the second machine learning model) for segmentation, it becomes easier to determine the correlation described later. In addition, rule-based segmentation may be used for the segmentation, or inference-based segmentation using a machine learning model may be used. In addition, segmentation may be performed by performing clustering using a general clustering method on the user group and referring to the clustering results.

Based on the strength of the correlation between the first score and the second score in the user segment, the correlation determination unit 25 determines whether the user segment is a first user segment having a correlation between the first score and the second score that is equal to or greater than a predetermined standard, or a second user segment having a correlation less than a predetermined standard (including those with no correlation in the present disclosure). That is, the correlation determination unit 25 determines whether the user segment to which the user belongs is a user segment having a strong correlation between the first score and the second score, or a user segment having a weak correlation between the first score and the second score, according to a predetermined standard. More specifically, for example, the correlation determination unit 25 calculates the difference between a statistical value of the first score calculated in the past (e.g., the average value or the median value) and a statistical value of the second score calculated in the past (e.g., the average value or the median value) for any user segment, and determines that the user segment is a first user segment if the difference is equal to or greater than a predetermined standard, and determines that the user segment is a second user segment if the difference is less than the predetermined standard. However, the specific means for determining the correlation between the first score and the second score in the user segment is not limited to the examples in the present disclosure.

For example, if the user group constituting the segment is men in their 50s and 60s, it is assumed that there is a strong correlation (correlated) between the height of the first score, which is a score based on credit information determined by a credit information agency, and the height of the second score, which is a score based on the output from a screening model for small payments. On the other hand, if the user group constituting the segment is men in their 20s and 30s, it is assumed that there is a weak correlation between the first score and the second score, or that there is no correlation (NOT correlated). This is because younger people in their 20s and 30s tend not to have sufficient accumulated credit information, so the accuracy of the first score is insufficient, and there is a strong tendency for the accuracy of the second score based on a screening model for uses that are frequently used by young people, such as small payments, to be sufficient.

The screening result determination unit 26 determines the screening result of the target user based on the first score and/or second score of the target user depending on the user segment to which the target user belongs. In this embodiment, when the user segment to which the target user belongs is determined to be the first user segment, the screening result determination unit 26 determines the screening result of the target user based on the first score for transactions in the first category and based on the second score for transactions in the second category. That is, in this embodiment, in the case of a user who belongs to the first user segment in which the first score and the second score are highly correlated, if the first score is a value that works against the screening, the final screening result is determined to be negative regardless of the second score (for example, the loan is not granted).

On the other hand, when the user segment to which the target user belongs is determined to be the second user segment, the screening result determination unit 26 determines the screening result of the target user based on the score indicating the higher creditworthiness or the lower risk among the first score and the second score. In other words, in this embodiment, when a specific user segment is a second user segment in which the correlation between the first score and the second score is weak or there is no correlation, the screening result determination unit 26 determines the final screening result in a positive manner (the screening result is determined so that the screening is favorable) when the first score is a value that works against the screening when the user belongs to the user segment, but is a value that can be determined to be sufficiently trustworthy based on the second score. This means that even if the first score is high and it is determined on the surface that the user lacks credit, the screening result is determined in a positive manner if the more reliable score (second score) indicates that the user is trustworthy.

Figure 3 is a diagram showing an overview of the screening process in this embodiment. That is, in this embodiment, when screening a target user, the loan screening result of the user is determined based on the first score output by the machine learning model for screening and the second score output by the machine learning model for another purpose, and at this time, the screening result determination logic by the screening result determination unit 26 is determined according to the strength of correlation between the first score and the second score for each user segment, and the screening result is output.

However, the specific method of determining the screening result of the target user based on the first score and/or the second score is not limited to the disclosure in this embodiment. For example, the screening result determination unit 26 may determine the screening result of the target user based on the third score calculated based on the first score and the second score. More specifically, for example, when the user segment to which the target user belongs is determined to be the first user segment, the screening result determination unit 26 may determine the screening result of the target user based on the third score calculated by setting the weight of the first score large and the weight of the second score small for transactions in the first category, and based on the third score calculated by setting the weight of the first score small and the weight of the second score large for transactions in the second category. Also, for example, when the user segment to which the target user belongs is determined to be the second user segment, the screening result determination unit 26 may determine the screening result of the target user based on the third score calculated by setting the weight of the score indicating a higher creditworthiness or lower risk among the first score and the second score large and setting the weight of the other score small.

The machine learning unit 27 generates and/or updates a machine learning model used by the first score acquisition unit 22 to acquire the first score, and a machine learning model used by the second score acquisition unit 23 to acquire the second score. The machine learning model for acquiring the first score is a machine learning model that outputs a first score indicating the creditworthiness of a target user when data on one or more user attributes related to the target user is input. The machine learning model for acquiring the second score is a machine learning model that outputs a second score indicating the degree of risk based on the likelihood that the user will default without making a payment when data on one or more user attributes related to the target user is input.

When generating and/or updating the machine learning model for obtaining the first score, the machine learning unit 27 creates teacher data for each user, in which the attribute data group of the user is defined as an input value and the score related to the user is defined as an output value. Then, the machine learning unit 27 generates and/or updates the first machine learning model based on the teacher data. As described above, the attribute data group input to the first machine learning model includes fact attribute data and estimated attribute data, and is combined with the score of the corresponding user and input to the machine learning unit 27 as teacher data. The score set in the teacher data may be a rule-based score or may be a manually set (annotated) score. It may also be a score that was previously output by the first machine learning model and then corrected by an administrator or the like.

When generating and/or updating a machine learning model for obtaining the second score, the machine learning unit 27 creates a machine learning model based on training data in which a statistical quantity (in this embodiment, the average value; however, a statistical indicator such as the mode or median may also be used) relating to the default occurrence rate of multiple users having a specific attribute is defined as a second score indicating the degree of risk of the user having the attribute. The calculated second score is combined with a group of attribute data including the factual attribute data and estimated attribute data of the corresponding user, and is input to the machine learning unit 27 as training data.

A machine learning model generation/update framework that can be adopted as the first machine learning model or the second machine learning model when implementing the technology according to the present disclosure is based on an ensemble learning algorithm, for example. For example, a machine learning framework (for example, LightGBM) based on a gradient boosting decision tree (GBDT) may be adopted as the framework. In other words, the framework may be a machine learning framework based on a decision tree model that transfers the error between the correct answer and the predicted value between the previous and next weak learners (weak classifiers). The predicted value here refers to the predicted value of the score, for example. In addition to LightGBM, the framework may adopt boosting methods such as XGBoost and CatBoost. According to a framework using a decision tree, a machine learning model with relatively high performance can be generated/updated with less effort in parameter adjustment compared to a framework using a neural network. However, the machine learning model generation/update framework that can be adopted when implementing the technology according to the present disclosure is not limited to the example in this embodiment. For example, instead of a gradient boosting decision tree, another learning device such as a random forest may be adopted as a learning device, or a learning device that is not called a weak learning device such as a neural network may be adopted. In particular, when a learning device that is not called a weak learning device such as a neural network is adopted, ensemble learning may not be adopted.

Figure 4 is a simplified diagram of the concept of a decision tree in a machine learning model employed in this embodiment. When a gradient boosting machine learning framework based on a decision tree algorithm is employed, the branching conditions of each node in the decision tree are optimized. Specifically, in the gradient boosting machine learning framework based on a decision tree algorithm, scores are calculated for each user group having attributes indicated by two child nodes branched from one parent node, and the branching conditions of the parent node are optimized so that the difference between these scores is large (for example, so that the difference is maximized or is equal to or greater than a predetermined threshold), that is, so that the two child nodes branch neatly. For example, if the attribute indicated as the branching condition of the node is age, the age set as the branching threshold may be changed, or the branching condition may be changed to an attribute other than age. In this way, the branching conditions of all nodes in the decision tree are recursively optimized, thereby improving the accuracy of estimating the score based on the attribute data group.

<Processing flow>
Next, a flow of processing executed by the screening device according to the present embodiment will be described. Note that the specific contents and processing order of the processing described below are an example for implementing the present disclosure. The specific contents and processing order may be appropriately selected according to the embodiment of the present disclosure.

Figure 5 is a flowchart showing the flow of the machine learning process according to this embodiment. The process shown in this flowchart is executed periodically or at a timing specified by the administrator.

In this embodiment, in the machine learning process, a first machine learning model and a second machine learning model are generated and/or updated. The machine learning unit 27 creates teacher data including a combination of a group of attribute data for each user accumulated in the past and a first score predetermined for the corresponding user (step S101). Then, the machine learning unit 27 uses the created teacher data to generate and/or update the first machine learning model (step S102). In addition, the machine learning unit 27 creates teacher data including a combination of a group of attribute data for each user accumulated in the past and a second score predetermined for the corresponding user (step S103). Then, the machine learning unit 27 creates and/or updates the second machine learning model using the created teacher data (step S104). Thereafter, the process shown in this flowchart ends.

Figure 6 is a flowchart showing the flow of the screening process according to this embodiment. The process shown in this flowchart is executed periodically or at a specified time for each target user.

In steps S201 to S203, a first score and a second score are obtained. The attribute determination unit 21 obtains a group of attribute data of the target user (step S201). Then, the first score acquisition unit 22 inputs input data corresponding to the first machine learning model from the group of attribute data obtained in step S201 into the first machine learning model, and obtains a first score to be set for the user based on the output value (step S202). Also, the second score acquisition unit 23 inputs input data corresponding to the second machine learning model from the group of attribute data obtained in step S201 into the second machine learning model, and obtains a second score to be set for the user based on the output value (step S203). After that, the process proceeds to step S204.

In steps S204 and S205, a user segment to which the target user belongs is identified, and a correlation between the first score and the second score in the user segment is determined. The user segment identification unit 24 identifies the user segment to which the target user belongs by performing segmentation of a user group including the target user (step S204). When the user segment to which the target user belongs is identified, the correlation determination unit 25 determines whether the user segment is a first user segment having a correlation between the first score and the second score that is equal to or greater than a predetermined standard, or a second user segment having a correlation that is less than a predetermined standard, based on the strength of the correlation between the first score and the second score in the identified user segment (step S205). If the user segment is determined to be the first user segment (YES in step S205), the process proceeds to step S206. On the other hand, if the user segment is determined to be the second user segment (NO in step S205), the process proceeds to step S207.

In step S206, the screening result of the target user is determined based on the score for the target category. If it is determined that the user segment to which the target user belongs is the first user segment, the screening result determination unit 26 determines the screening result of the target user based on the first score if this screening process was performed for a transaction in the first category, and based on the second score if this screening process was performed for a transaction in the second category. Then, the process proceeds to step S208.

In step S207, the screening result of the target user is determined based on the score indicating higher credibility or lower risk. If it is determined in step S205 that the user segment to which the target user belongs is the second user segment, the screening result determination unit 26 determines the screening result of the target user based on the score indicating higher credibility or lower risk out of the first score and the second score. Then, the process proceeds to step S208.

In step S208, the screening results are output. The screening result determination unit 26 outputs the screening results in step S206 or step S207. After that, the processing shown in this flowchart ends.

＜Effects＞
According to this embodiment, it is possible to more accurately support loan screening and the like for any user who belongs to a user segment with insufficient accumulation of credit information, such as the younger generation. Also, according to this embodiment, it is possible to more appropriately support user screening by considering the correlation between different credit scores by combining different input data and models.

<Variations>
Hereinafter, as a variation of the embodiment described above, a variation in which the assessment device 1b determines the second score based on the attribute data group of the target user will be described. Here, the system configuration of the assessment device 1b according to this variation and an example of the user attribute data used in this variation are roughly similar to those in the embodiment described above, and therefore the description will be omitted (see FIG. 1).

Figure 7 is a diagram showing an outline of the functional configuration of the review device 1b relating to the variation. In this variation, the review device 1b includes a product-related data acquisition unit 28 in addition to the attribute determination unit 21, first score acquisition unit 22, second score acquisition unit 23, user segment identification unit 24, correlation determination unit 25, review result determination unit 26, and machine learning unit 27 described in the embodiment described above.

The product-related data acquisition unit 28 acquires product-related data of the product/service that the target user intends to purchase by receiving the loan related to the transaction. Here, the product-related data acquisition unit 28 acquires at least one of attribute data of the product/service and attribute data of the store (including online store) selling the product/service as the product-related data. Here, the product-related data may be any information related to the product/service, and the data acquired as the product-related data is not limited. However, the product/service attribute data may include, for example, price and product/service category/genre, and the store attribute data may include, for example, store price range and store category/genre. Furthermore, the product-related data may include business rules and the type of transaction associated with the transaction of the target product/service. Here, the business rules and the type of transaction may include commercial practices related to the product/service, the contents of the contract associated with the transaction, and the delivery mode of the product/service (delivery method, delivery at the store, etc.).

In this variation, the second score acquisition unit 23 acquires a second score for determining whether to approve or reject a loan for the purchase of goods and/or services by the target user based on the output of the second machine learning model. Here, the second machine learning model may be trained appropriately for each item-related data.

More specifically, when the product-related data of the target product/service is acquired, the second score acquisition unit 23 inputs the attribute data of the target user into the latest second machine learning model generated or updated by the machine learning unit 27, and obtains, as an output value, a second score that changes according to the risk that the target user will not properly settle the loan if the target user purchases the target product/service with a loan. The processing after the second score is obtained is the same as the processing described in the above embodiment, and therefore will not be described.

<Other variations>
Hereinafter, as a variation of the embodiment described above, the screening device 1 may determine a screening result for a user based on the outputs (first to n-th scores) of a number n of machine learning models (first to n-th machine learning models) including a first machine learning model and a second machine learning model (n>1). Furthermore, the input data to the first to n-th machine learning models may be the input data to the above-mentioned first machine learning model, or may be the input data to the second machine learning model.

The first to nth machine learning models may be generated by learning the relationship between a user's attributes and the creditworthiness of the user in first to nth categories of transactions classified according to the period and/or amount of the transaction. In addition, the first to nth machine learning models may be generated by learning the relationship between a user's attributes and the creditworthiness of the user in first to nth categories of transactions classified according to other characteristics such as interest rates, the presence or absence of collateral, and repayment method, without being limited to the period or amount.

The review device 1 determines the review result of the target user based on the first to nth scores of the target user depending on the user segment to which the target user belongs. The review device 1 may determine the review result based on any of the first to nth scores depending on the user segment to which the target user belongs, or may determine the review result based on the weighted first to nth scores.

1. Examination equipment

Claims (11)

a first score acquisition means for acquiring a first score of the user based on an output indicating the user's creditworthiness, the output being obtained by inputting input data including the user's attribute data into a first machine learning model generated by learning the relationship between the user's attributes in a first type of transaction and the user's creditworthiness;
a second score acquisition means for acquiring a second score for the user based on an output indicating the risk of the user, the output being obtained by inputting input data including the user's attribute data into a second machine learning model generated by learning the relationship between the user's attributes and the risk of the user in a second type of transaction different from the first type of transaction;
A user segment identification means for identifying a user segment to which a target user belongs by performing segmentation of a user group including the target user;
a correlation determination means for determining whether the user segment is a first user segment having a correlation between the first score and the second score that is equal to or greater than a predetermined standard, or a second user segment having a correlation between the first score and the second score that is less than a predetermined standard, based on the strength of the correlation between the first score and the second score in the user segment;
an examination result determination means for determining an examination result of the target user based on the score indicating a higher creditworthiness or a lower risk among the first score and the second score when the user segment to which the target user belongs is determined to be the second user segment;
An examination device comprising: The first score acquisition means acquires a first score for the user based on an output obtained by inputting the input data related to the user into a first machine learning model, which is a machine learning model for screening the first type of transaction related to a predetermined period or more and/or a predetermined amount or more;
The second score acquisition means acquires a second score for the user based on an output obtained by inputting the input data related to the user into a second machine learning model, which is a machine learning model for screening the second type of transaction related to a shorter period and/or a smaller amount than the first type of transaction.
The examination device according to claim 1. The first score acquisition means acquires a first score for the user by using the first machine learning model generated and/or updated using user attribute data including the status of other transactions of the same type as the first type of transaction and teacher data including a value indicating the creditworthiness of the user.
The examination device according to claim 2. The second score acquisition means acquires a second score for the user by using the second machine learning model generated and/or updated using user attribute data including statuses of other transactions of the same type as the second type of transaction and teacher data including a value indicating a risk related to the user.
The examination device according to claim 2. The first score obtaining means estimates the first score using the first machine learning model generated and/or updated using a machine learning framework based on a gradient boosting decision tree;
The second score obtaining means estimates the second score using the second machine learning model generated and/or updated using a machine learning framework based on a gradient boosting decision tree.
The examination device according to claim 1. the correlation determination means determines, for the user segment, that the user segment is a first user segment when a difference between a statistical value of a first score calculated in the past and a statistical value of a second score calculated in the past is equal to or greater than a predetermined criterion, and determines, for the user segment, that the user segment is a second user segment when the difference is less than the predetermined criterion;
The examination device according to claim 1 . When it is determined that the user segment to which the target user belongs is the first user segment, the screening result determination means determines a screening result of the target user for the first type of transaction based on the first score.
The examination device according to claim 1 . When it is determined that the user segment to which the target user belongs is the first user segment, the screening result determination means determines a screening result of the target user for the second type of transaction based on the second score.
The examination device according to claim 1 . The first type of transaction is a type of transaction other than a deferred payment transaction,
the first score acquisition means acquires the first score of the user based on an output obtained by inputting the input data related to the user into the first machine learning model, which is a machine learning model for screening the first type of transaction;
The second score acquisition means acquires the second score of the user based on an output obtained by inputting the input data related to the user into the second machine learning model, which is a machine learning model for screening the transaction by deferred payment, and which is generated and/or updated using teacher data including user attribute data including the status of the transaction by deferred payment and a value indicating a risk related to the user;
The output value of the second machine learning model is a judgment score that changes depending on the risk that the target user will not correctly settle the deferred payment when the target user purchases goods and/or services with deferred payment,
The examination device according to claim 8 . The computer
a first score acquisition step of acquiring a first score of the user based on an output indicating the user's creditworthiness obtained by inputting input data including the user's attribute data into a first machine learning model generated by learning the relationship between the user's attributes in a first type of transaction and the user's creditworthiness;
a second score acquisition step of acquiring a second score for the user based on an output indicating the risk of the user, the output being obtained by inputting input data including user attribute data into a second machine learning model generated by learning the relationship between the user attributes and the risk of the user in a second type of transaction different from the first type of transaction;
a user segment identification step of identifying a user segment to which the target user belongs by performing segmentation of a user group including the target user;
a correlation determination step of determining whether the user segment is a first user segment having a correlation between the first score and the second score that is equal to or greater than a predetermined standard, or a second user segment having a correlation between the first score and the second score that is less than a predetermined standard, based on the strength of the correlation between the first score and the second score in the user segment;
an examination result determination step of determining an examination result of the target user based on the score indicating a higher creditworthiness or a lower risk among the first score and the second score when the user segment to which the target user belongs is determined to be the second user segment;
How to do it. Computer,
a first score acquisition means for acquiring a first score of the user based on an output indicating the user's creditworthiness, the output being obtained by inputting input data including the user's attribute data into a first machine learning model generated by learning the relationship between the user's attributes in a first type of transaction and the user's creditworthiness;
a second score acquisition means for acquiring a second score for the user based on an output indicating the risk of the user, the output being obtained by inputting input data including the user's attribute data into a second machine learning model generated by learning the relationship between the user's attributes and the risk of the user in a second type of transaction different from the first type of transaction;
A user segment identification means for identifying a user segment to which a target user belongs by performing segmentation of a user group including the target user;
a correlation determination means for determining whether the user segment is a first user segment having a correlation between the first score and the second score that is equal to or greater than a predetermined standard, or a second user segment having a correlation between the first score and the second score that is less than a predetermined standard, based on the strength of the correlation between the first score and the second score in the user segment;
an examination result determination means for determining an examination result of the target user based on the score indicating a higher creditworthiness or a lower risk among the first score and the second score when the user segment to which the target user belongs is determined to be the second user segment;
A program that functions as a