Credit learning device, credit learning method, credit estimation device, credit estimation method, and program

Abstract

In order to address the problem of improving the accuracy of estimating an individual's creditworthiness, a learning device 1 is provided with: a first training unit 23a that trains a first model on first training data including combinations of first attribute data groups relating to users who use first service and labels corresponding to a first score that changes with the risk borne by a service provider when the users use the first service; and a second training unit 23b that trains a second model, which has some parameters of the configuration of the trained first model as fixed parameters, on second training data including combinations of second attribute data groups relating to users who use second service and labels corresponding to a second score that changes with the risk borne by the service provider when the users use the second service.

Description

Technical Field

[0001] This disclosure relates to a technique for estimating a user's creditworthiness.

Background Art

[0002] Conventionally, an acquisition unit that acquires asset information, which is information obtained by arbitrarily combining individual assets owned by a first user, and the behavior history of the first user on the network, and a determination unit that estimates information related to the assets of a second user based on the asset information and the behavior history of the first user acquired by the acquisition unit have been proposed (see Patent Document 1).

[0003] Also, conventionally, in estimating credit risk, it has been proposed to perform transfer learning from the domain of credit card financing and debt restructuring to the domain of small business financing (see Non-Patent Document 1).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] Conventionally, various technologies have been proposed to determine the trustworthiness of users regarding a given service. However, while these conventional technologies are somewhat effective in accurately determining an individual's trustworthiness, there is room for improvement in terms of flexibly improving the accuracy of trustworthiness estimation within specific domains.

[0007] In light of the issues described above, this disclosure aims to improve the accuracy of flexibly estimating an individual's creditworthiness within a specific domain. [Means for solving the problem]

[0008] An example of the present disclosure is a credit learning device comprising: a first learning means for training a first model with first training data, which includes a combination of a first set of attribute data relating to a user of a first service and a label corresponding to a first score that changes according to the degree of risk that the provider of the first service would incur if the user used the first service; and a second learning means for training a second model, which has some parameters of the configuration of the trained first model as fixed parameters, with second training data, which includes a combination of a second set of attribute data relating to a user of a second service different from the first service and a label corresponding to a second score that changes according to the degree of risk that the provider of the second service would incur if the user used the second service.

[0009] Another example of the present disclosure is a credit learning device comprising: a first learning means for training a first model with first training data, which includes a combination of a first attribute data set relating to users of the first service who belong to a first segment, and a label corresponding to a score that changes according to the degree of risk that the provider of the second service would incur if the user were to use a second service different from the first service; and a second learning means for training a second model, which has some parameters of the configuration of the trained first model as fixed parameters, with second training data, which includes a combination of a second attribute data set relating to users of the first service who belong to a second segment different from the first segment, and a label corresponding to a score that changes according to the degree of risk that the provider of the second service would incur if the user were to use the second service.

[0010] Another example of this disclosure is a credit estimation device that includes estimation means for estimating the second score to be set for a target user by inputting a set of attribute data relating to the target user into the second model which has been trained by the learning method described above.

[0011] This disclosure can be understood as an information processing device, a system, a method executed by a computer, or a program to be executed by a computer. Furthermore, this disclosure can also be understood as such a program recorded on a recording medium readable by a computer or other device or machine. Here, a recording medium readable by a computer refers to a recording medium that stores information such as data and programs through electrical, magnetic, optical, mechanical, or chemical means and can be read by a computer. [Effects of the Invention]

[0012] According to this disclosure, it will be possible to flexibly improve the accuracy of estimating an individual's creditworthiness within a specific domain. [Brief explanation of the drawing]

[0013] [Figure 1] This is a schematic diagram showing the configuration of the information processing system according to the first embodiment. [Figure 2] This figure shows a schematic representation of the functional configuration of the information processing device according to the first embodiment. [Figure 3] This is a simplified diagram illustrating the concept of a decision tree in the machine learning model employed in the first embodiment. [Figure 4] This figure shows an overview of the first model generated when a neural network-based framework is employed in the first embodiment. [Figure 5] This figure shows an overview of the second model generated when a neural network-based framework is adopted in the first embodiment. [Figure 6] This is a simplified diagram of the score estimation process according to the first embodiment. [Figure 7] This is a flowchart showing the machine learning processing flow according to the first embodiment. [Figure 8] This is a flowchart showing the flow of the score estimation process according to the first embodiment. [Figure 9] This is a schematic diagram showing the configuration of the information processing system according to the second embodiment. [Figure 10] This figure shows a schematic representation of the functional configuration of the information processing device according to the second embodiment. [Figure 11] This is a flowchart showing the machine learning processing flow according to the second embodiment. [Modes for carrying out the invention]

[0014] Hereinafter, embodiments of an information processing apparatus, method, and program according to the present disclosure will be described based on the drawings. However, the embodiments described below are merely illustrative of the embodiments, and do not limit the information processing apparatus, method, and program according to the present disclosure to the specific configurations described below. In implementation, a specific configuration corresponding to the implementation mode may be appropriately adopted, and various improvements and modifications may be made. The present invention can appropriately adopt at least a part of the configurations in each of the embodiments and variations described later with respect to each other. In the embodiments and variations described later, a credit learning apparatus, a credit learning method, a credit estimation apparatus, and a credit estimation method are simply referred to as a learning apparatus, a learning method, an estimation apparatus, and an estimation method.

[0015] <First Embodiment> In this embodiment, the technology according to the present disclosure is retrained using a dataset related to a financial-related service different from the post-payment settlement service based on a machine learning model generated to calculate the default risk (hereinafter referred to as "post-payment risk") of a user related to the post-payment settlement service. An aspect in the case of implementing the label indicating the post-payment risk as a label capable of estimating risks related to different financial-related services will be described. However, the technology according to the present disclosure can be widely used for a technology for calculating the score of a user related to a different service by using a model generated using a dataset related to the service to calculate the score of a user related to a certain service, and the application target of the present disclosure is not limited to the example shown in the embodiment. For example, the technology according to the present disclosure may be implemented to enable risk estimation for a plurality of different services belonging to fields other than finance.

[0016] <<System Configuration>> Figure 1 is a schematic diagram showing the configuration of the information processing system according to this embodiment. In the information processing system according to this embodiment, the information processing device 1 and one or more service provision systems 5 are connected to each other in a manner that allows them to communicate with one another. The user is a user of the services provided by the service provision system 5, and receives the services by accessing the service provision system 5 from the user terminal. Here, the information processing device 1 represents a learning device, an estimation device, or a combination thereof.

[0017] The information processing device 1 is a computer equipped with a CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, storage devices 14 such as EEPROM (Electrically Erasable and Programmable Read Only Memory) and HDD (Hard Disk Drive), a communication unit 15 such as a NIC (Network Interface Card), and the like. However, the specific hardware configuration of the information processing device 1 can be omitted, replaced, or added as appropriate depending on the implementation. Furthermore, the information processing device 1 is not limited to a device consisting of a single enclosure. The information processing device 1 may be implemented by multiple devices using so-called cloud or distributed computing technologies.

[0018] The service provision system 5 is a computer equipped with a CPU, ROM, RAM, storage device, communication unit, input device, output device, etc. (not shown in the diagram). Furthermore, these systems and terminals are not limited to devices consisting of a single enclosure. These systems and terminals may be implemented by multiple devices using technologies such as so-called cloud computing or distributed computing.

[0019] The services provided by the service provision system 5 include, for example, online shopping services, online reservation services, loan services, banking services, credit card services, deferred payment services, electronic money payment services, advertising services, operation center services, subscription services, or map information services. Note that "deferred payment" is not limited to services known as Buy Now Pay Later (BNPL), but may include any purchase of goods / services on a deferred payment basis.

[0020] The services provided by the service provision system 5 are not limited to those exemplified in this embodiment. The service provision system 5 notifies the information processing device 1 of user-related data when providing services. Here, user-related data includes the user's service usage history data. The content of the service usage history data varies depending on the content of the service, and may include, for example, user location information history data, credit card usage amount / post-payment usage amount payment history data, electronic money usage history data, transaction history data (including purchase history data of goods, etc.), reservation history data, operation history data from the operation center to the user, etc.

[0021] In this embodiment, among the various services provided by the service provision system 5, financial services including deferred payment services, loan services, banking services, and credit card services (i.e., services belonging to the "finance" category in business classifications and domains), are described in which a machine learning model generated to calculate the user's deferred payment risk for the deferred payment service (corresponding to the "first service" described later) is retrained using a dataset related to a different financial services (for example, a loan service, corresponding to the "second service" described later), and the label indicating the deferred payment risk is changed to a label corresponding to the risk for a different financial services. Here, retraining refers to transfer learning, additional learning, etc., depending on the learning method. In this embodiment, the service group may be a group of services that share a common membership program (user identifier), a group of services that share a common points program (electronic value such as points that can be awarded or used), or a group of services that grant benefits usable in one service in conjunction with the use of the other service.

[0022] The information processing device described in this disclosure uses a machine learning model that takes user attribute data as input and outputs a credit score (in this embodiment, a score indicating deferred payment risk). The attribute data used as input may also include data indicating default (debt default). By using various attribute data as input, the information processing device 1 described in this disclosure calculates a universal and generalized credit score that uniformly reflects all of the user's attributes. The calculated credit score can be used for credit screening (determination involving approval or rejection of deferred payment) in deferred payment services, etc., but as described above, in this embodiment, the calculated credit score is used to determine whether or not to include the user in the settlement method option proposal.

[0023] Here, user attribute data includes factual attribute data and estimated attribute data. Attribute data includes data expressed in data formats such as scores (e.g., continuous values ​​between 0 and 1) or labels (e.g., binary values ​​corresponding to presence or absence or right / wrong). However, the format of attribute data is not limited to the examples provided in this disclosure. Attribute data may also include, for example, online service usage and the usage of electronic value, including points. Online service usage may include at least one of the number of cancellations, cancellation rates, and number of orders in online shopping services or online reservation services.

[0024] Factual attribute data is data that indicates factual attributes of a user that can be confirmed as factual, based on user-provided data obtained from the user themselves, historical data collected about the user, etc. Examples of user-provided data include registration data such as the user's name, email address, telephone number, address, workplace, school, etc., and data obtained as a result of the user's responses to questionnaires, etc. Examples of historical data include the usage history data of e-commerce services provided by the service provision system 5 mentioned above. Preferably, factual attribute data is data obtained by converting the aforementioned user-provided data and historical data into a data format suitable for marketing and / or analysis purposes. For example, factual attribute data that can be obtained based on usage history data includes the genre / category and brand of products / services that the user frequently uses, as well as commercial areas, leisure spots, and tourist destinations that the user frequently visits.

[0025] Furthermore, estimated attribute data is data that represents estimated attributes (inferred attributes) obtained by estimating based on user-provided data, historical data, factual attribute data, etc. In this embodiment, estimated attribute data includes the user's personality, etc., estimated or predicted using machine learning techniques. Preferably, estimated attribute data is data about attributes that influence the user's behavior when used for targeting.

[0026] In this embodiment, for example, when a decision tree model is used, an importance value is assigned to each attribute data. This importance value indicates the degree of correlation between the attribute data and the credit score when the attribute data is used in calculating the credit score. Each time the appropriateness of the credit score is evaluated by the learning unit 23, described later, the model parameters are adjusted so that the credit score becomes a more appropriate value. The importance value corresponding to each attribute data corresponds, for example, to the weights corresponding to each node (each regression tree) in a model for calculating the credit score, such as the decision model described later, and is determined as appropriate during the process of calculating the credit score.

[0027] Here, the attribute data set may include demographic attributes, behavioral attributes, or psychographic attributes. Demographic attributes may include, for example, the user's gender, family structure, age, etc. Behavioral attributes may be based on service usage history data, for example, whether or not cash advances are used, whether or not revolving credit is used, deposit and withdrawal history for a specified account, commercial transaction history for any goods / services including gambling or lotteries (may include online transaction history on online marketplaces, etc.), user movement history using location information or place information, etc. Psychographic attributes may include, for example, preferences regarding gambling or lotteries. However, the available user attributes are not limited to the examples in this embodiment. For example, "time required for operations (calls, etc.)" from operation center services, etc., and "credit card usage amount / post-payment usage amount" may also be used as attributes. Demographic attributes and behavioral attributes may be treated as factual attributes. Psychographic attributes may be treated as inferred attributes. Furthermore, attributes similar to demographic attributes may be estimated attributes based on factual attributes derived from user-provided data or historical data. Similarly, attributes similar to behavioral attributes may be estimated attributes derived from factual attributes derived from user-provided data or historical data. Psychographic attributes may be factual attributes based on user-provided data, including, as an example, the results of user input.

[0028] Figure 2 is a diagram illustrating the schematic configuration of the information processing device 1 according to this embodiment. The information processing device 1 functions as an information processing device comprising a fact attribute determination unit 21, an estimated attribute determination unit 22, a learning unit 23, and a score estimation unit 24, by having a program recorded in the storage device 14 read into the RAM 13, executed by the CPU 11, and controlling each piece of hardware provided in the information processing device 1. In this embodiment and other embodiments described later, each function of the information processing device 1 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.

[0029] The fact attribute determination unit 21 determines fact attribute data that can be confirmed as facts about the user based on user-provided data and / or the user's history data provided by the user. In this embodiment, the fact attribute determination unit 21 determines the fact attribute data pertaining to the user by methods such as aggregating user-provided data and / or history data, determining the relevant attributes by referring to other data such as maps, or using the user-provided data and / or history data as is. In this embodiment, the method of determining fact attribute data pertaining to a user based on user-provided data and / or the user's history data is employed, but fact attribute data pertaining to a user may be obtained by other methods.

[0030] The Estimated Attribute Determination Unit 22 determines estimated attribute data estimated for a user based on user-related data that includes at least one or more factual attribute data determined for the target user by the Factual Attribute Determination Unit 21. In this embodiment, the Estimated Attribute Determination Unit 22 determines estimated attributes based on the output value obtained by inputting user-related data, which includes one or more factual attribute data relating to the target 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 target user has a predetermined estimated attribute, and the Estimated Attribute Determination Unit 22 determines that the target user has the estimated attribute if the output value obtained from the attribute estimation model is within a predetermined range. If it is determined that the target user has a predetermined estimated attribute, the Estimated Attribute Determination Unit 22 sets the label of the attribute data estimated for the target user to a value indicating the presence or absence of the attribute or the type of attribute. Alternatively, the estimated attribute data may be represented by a score instead of a label. In this case, the Estimated Attribute Determination Unit 22 sets a value indicating the degree (probability) to which the estimated attribute can be applied to the score of the attribute data estimated for the target user. This degree may be the output value of the attribute estimation model.

[0031] Furthermore, the attribute estimation unit 22 can also estimate missing (insufficient) attribute data based on other attribute data. Therefore, if attribute data is missing or insufficient for a particular user, the attribute data estimated by the attribute estimation unit 22 may be used instead of the actual attribute data. The attribute data supplementation by the attribute estimation unit 22 can also be applied to supplement the attribute data of users whose attribute data is insufficient in quantity or quality (for example, light users of the first service) in the second embodiment described later.

[0032] The learning unit 23 generates and / or updates the score estimation model generated by the score estimation unit 24. The score estimation model is a machine learning model that, when one or more attribute data (group of attribute data) relating to a target user is input, outputs a credit score indicating the degree of risk that the target user will not properly settle the deferred payment when using deferred payment. By using such a machine learning model, in this embodiment, it is possible to obtain a credit score that is not just a simple credit score, but an output that takes into account factors specific to deferred payment. Furthermore, in this embodiment, an example is shown and explained in which a credit score is adopted in which a larger credit score value indicates higher creditworthiness (lower risk), and a smaller credit score value indicates lower creditworthiness (higher risk), but the relationship between the magnitude of the score value and the level of risk may be reversed.

[0033] In generating and / or updating the score estimation model, the learning unit 23 creates training data for each user, which includes a combination of the user's attribute data set and the credit score of users whose attribute data set is common to that user. Then, the learning unit 23 generates and / or updates the score estimation model based on this training data. As described above, the attribute data set input to the score estimation model includes factual attribute data determined by the factual attribute determination unit 21 and estimated attribute data estimated by the estimated attribute determination unit 22 based on user-related data including the factual attribute data. These are combined with labels indicating the risk and corresponding score for the corresponding user and input to the learning unit 23 as training data. In this embodiment, the credit score set in the training data is a credit score determined based on the deferred payment history data of the user corresponding to the combination of user attributes. Here, the deferred payment history data includes data indicating whether or not there has been a default (failure to pay) in the deferred payment and the amount of the default. In this case, the credit score may be a rule-based credit score or a manually set (annotated) credit score. Furthermore, the credit score may be one that was previously generated by a score estimation model and subsequently modified by an administrator or other relevant party.

[0034] A machine learning model generation / update framework that can be used as a score estimation model, etc., when implementing the technology related to this disclosure is, for example, based on an ensemble learning algorithm. This framework may employ, for example, a machine learning framework based on a Gradient Boosting Decision Tree (GBDT) (e.g., LightGBM). In other words, the framework may employ a machine learning framework based on a decision tree model that carries over the error between the correct answer and the predicted value between preceding and succeeding weak learners (weak classifiers). Here, the predicted value refers, for example, to the predicted value of a credit score. In addition to LightGBM, the framework may employ boosting methods such as XGBoost and CatBoost. A decision tree framework allows for the generation / update of machine learning models with relatively high performance with less parameter adjustment effort compared to a neural network framework. However, the machine learning model generation / update frameworks that can be used when implementing the technology related to this disclosure are not limited to the examples given in this embodiment. For example, a random forest or other learning mechanism may be used instead of a gradient boosting decision tree as the learning mechanism, and a learning mechanism that is not considered a so-called weak learning mechanism, such as a neural network, may also be used. Furthermore, if a learning mechanism that is not considered a so-called weak learning mechanism, such as a neural network, is used, ensemble learning does not necessarily have to be employed.

[0035] Figure 3 is a simplified diagram of the concept of a decision tree in a machine learning model used as a score estimation model in this embodiment. When a gradient boosting machine learning framework based on a decision tree algorithm is adopted, the branching conditions of each node in the decision tree are optimized. Specifically, in a gradient boosting machine learning framework based on a decision tree algorithm, a credit score is calculated for each user group having the attributes indicated by the two child nodes that branch off from a single parent node, and the branching conditions of the parent node are optimized so that the difference between these credit scores is large (for example, so that the difference is maximized or exceeds a predetermined threshold), that is, so that the two child nodes branch cleanly. For example, if the attribute indicated as the branching condition of a 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, by recursively optimizing the branching conditions of all nodes in the decision tree, the estimation accuracy of credit scores based on attribute data can be improved.

[0036] In this embodiment, the learning unit 23 comprises at least a first learning unit 23a and a second learning unit 23b.

[0037] The first learning unit 23a trains the first model (first score estimation model) with first training data, which includes a combination of first attribute data related to a user of the first service (in this embodiment, a deferred payment service) and a label corresponding to a score (user credit score in the first service) that changes according to the degree of risk (deferred payment risk) that the provider of the first service incurs when the user uses the first service. As described above, in this embodiment, the first service is a deferred payment service. The risk borne by the provider of the first service is deferred payment risk, which is determined based on the payment history in the deferred payment service and represents the risk that the user will not properly settle the deferred payment when the user uses the deferred payment service using some indicator (in this embodiment, a label that changes according to the magnitude of the deferred payment risk).

[0038] The second learning unit 23b trains a second model (second score estimation model) which has some of the configuration of the first model (intermediate layers, nodes, and other parameters such as weights) trained by the first learning unit 23a as fixed parameters, with second training data including a combination of a second attribute data set relating to a user of the second service (in this embodiment, a loan service) and a label corresponding to a second score that changes according to the degree of risk that the provider of the second service will incur if the user uses the second service. The second service is a different service from the first service, and in this embodiment, an example in which the second service is a loan service will be mainly described. However, the second service is not limited to the examples in this embodiment, as long as a correlation can be confirmed or inferred between the risk that the provider of the first service will incur if the user uses the first service and the risk that the provider of the second service will incur if the user uses the second service. For example, a combination of services may be adopted in which, for the same group of users, there is a correlation between the score corresponding to the risk in the first service and the score corresponding to the second service.

[0039] If a machine learning framework based on gradient boosting decision trees is adopted for both the first and second models, the second learning unit 23b trains the second model, which has some of the nodes of the first model's nodes as fixed parameters, with the second training data, and updates the parameters (weights) of the nodes that were not fixed. Alternatively, the second learning unit 23b may train the second model, which has some of the tree structure of the first model fixed, with the second training data.

[0040] Here, when determining the parameters of the nodes to be fixed, for example, a method may be adopted in which nodes of high importance, i.e., nodes corresponding to the branching conditions (attributes) that most greatly affect the model's estimation (output), or nodes that have an influence on the output above a predetermined standard, are given priority in fixing (see Figure 3). Furthermore, when updating parameters, for example, parameters (weights) excluding the fixed (freezed) layers and nodes can be updated so as to minimize the loss between the target domain's (repurposed destination; in this case, the second service) dataset and the network output.

[0041] If a machine learning framework based on a neural network is employed for both the first and second models, the second learning unit 23b trains the second model, which has some parameters of the intermediate layer (hidden layer) of the first model fixed, with the second training data. In this case, the second learning unit 23b may train the second training data on a second model in which multiple copies of the trained first model are arranged in parallel and share a common output layer, and in which some parameters of the intermediate layer (hidden layer) are fixed.

[0042] Figure 4 is a diagram showing an overview of the first model generated by the first learning unit 23a when a neural network-based machine learning framework is employed in this embodiment. The first learning unit 23a uses a neural network-based machine learning framework to train the first model on first training data, which includes a combination of first attribute data related to a user of the first service and labels that change according to the degree of risk borne by the provider of the first service when the user uses the first service. In the example shown in Figure 4, the first model is an example of a neural network including hidden layers a to c.

[0043] Figure 5 shows an overview of the second model generated by the second learning unit 23b when a neural network-based machine learning framework is adopted in this embodiment. The second learning unit 23b generates the second model by connecting multiple copies of the neural network trained on the dataset of the source domain (the source of the repurposed data; in this case, the first service) in parallel and retraining some of its intermediate layers. Referring to the example shown in the figure, for example, the second learning unit 23b copies multiple neural networks (four in the example shown in the figure, NN1 to NN4) that include intermediate layers a to c, which are included in the first model (see Figure 4) trained on the dataset related to the first service, which is the source domain (first training data), and connects these multiple neural networks NN1 to NN4 in parallel, sharing the input and output layers. Then, it creates the second model by freezing some of the intermediate layers (in the example shown in Figure 5, intermediate layers a to c of NN1, intermediate layers a and b of NN2, and intermediate layer a of NN3, shown by solid lines). Then, the second learning unit 23b updates the unfixed intermediate layers (in the example shown in Figure 5, intermediate layers c of NN2, intermediate layers b and c of NN3, and intermediate layers a to c of NN4, shown by dashed lines).

[0044] Note that the method for fixing the hidden layers, as explained using Figure 5, is just one example in this embodiment, and various variations can be used for fixing the hidden layers (see, for example, Non-Patent Document 1). Also, in the example explained using Figure 5, a second model was created by connecting multiple copies of the neural network trained on the source domain dataset in parallel, but the second learning unit 23b may also fix some of the hidden layers of a single neural network trained on the source domain dataset (i.e., without creating multiple copies of the hidden layers) and retrain it. Furthermore, the output layer may also be tuned by the second learning unit 23b during the retraining process.

[0045] The score estimation unit 24 estimates a second score (credit score, creditworthiness) for the target user based on an attribute data set that includes factual attributes and estimated attributes related to the target user. In this case, the score estimation unit 24 may perform some processing (normalization, ranking, labeling, etc.) on the determined factual attributes and / or estimated attributes and make them part of the attribute data set, or it may make all or part of the attribute data set other types of scores (e.g., so-called credit scores, etc.) or labels calculated using the determined factual attributes and / or estimated attributes. Here, other machine learning models may be involved in the calculation of other types of scores or labels. The score estimation unit 24 may also estimate a score for a virtual user by inputting an attribute data set of a virtual user for whom there is no usage history or transaction history in the service into the second model.

[0046] Figure 6 is a simplified diagram of the score estimation process according to this embodiment. In this embodiment, the score estimation unit 24 estimates (calculates) the user's score by inputting the user's attribute data set into a score estimation model. Here, the output value of the score estimation model is, for example, a score with a minimum value of 0 and a maximum value of 1. In this embodiment, the score estimation unit 24 inputs the attribute data set relating to the target user into a second model generated by the learning unit 23 to estimate a second score that changes according to the degree of risk borne by the provider of the second service if the target user uses the second service.

[0047] <<Processing Flow>> Next, the processing flow performed by the information processing device according to this embodiment will be described. Note that the specific processing content and processing order described below are examples for implementing this disclosure. The specific processing content and processing order may be appropriately selected depending on the embodiment of this disclosure.

[0048] Figure 7 is a flowchart showing the flow of machine learning processing according to this embodiment. The processing shown in this flowchart is executed periodically or at times specified by the administrator.

[0049] In steps S101 and S102, a first score estimation model is generated. The first learning unit 23a creates first training data for the user group of the first service, which includes a combination of user attribute data accumulated in the past and a label predetermined for the corresponding user (step S101). Then, the first learning unit 23a performs training processing on the first score estimation model based on the created first training data (step S102). After that, the process proceeds to step S103.

[0050] In steps S103 and S104, a second score estimation model is generated. The second learning unit 23b creates second training data for the second service user group, which includes combinations of previously accumulated user attribute data and pre-determined labels for the corresponding users (step S103). Then, the second learning unit 23b retrains the second score estimation model, which has some of the parameters of the first score estimation model generated in step S102 as fixed parameters, using the second training data, thereby constructing a second score estimation model to be used for score estimation by the score estimation unit 24 (step S104). After that, the process shown in this flowchart is completed.

[0051] Figure 8 is a flowchart showing the flow of the score estimation process according to this embodiment. The process shown in this flowchart is executed periodically or at specified intervals for each target user.

[0052] In steps S201 and S202, factual attribute data and estimated attribute data are determined. The factual attribute determination unit 21 determines factual attribute data relating to the target user based on the user-provided data and / or historical data of the target user (step S201). Then, the estimated attribute determination unit 22 determines estimated attribute data relating to the target user based on at least the factual attribute data determined in step S201 (step S202). After that, the process proceeds to step S203.

[0053] In steps S203 and S204, the score of the target user is estimated. The score estimation unit 24 determines an attribute data set including the factual attribute data determined in step S201 and the estimated attribute data determined in step S202 (step S203). The score estimation unit 24 then inputs the attribute data set determined in step S203 into a second estimation model and obtains the output value as a score that changes according to the degree of risk that the target user will not properly settle the deferred payment when using the deferred payment service (step S204). However, the method of estimating the score is not limited to the examples in this embodiment. For example, the score may include a value calculated by inputting the attribute data set into a predetermined function or statistical model other than a machine learning model. After that, the process shown in this flowchart is completed.

[0054] <<Variations>> In this embodiment, the learning unit 23 may further include a third learning unit, a fourth learning unit, ... nth (n>2) learning units. Here, the nth learning unit trains the nth model (nth score estimation model), which has some of the parameters of the nth model trained by the nth learning unit as fixed parameters, etc., with the nth training data, which includes a combination of the nth attribute data set relating to a user who uses the nth service different from the first, second, ... nth-1 services, and a label corresponding to the nth score which changes according to the degree of risk that the provider of the nth service would incur if the user uses the nth service. At this time, the score estimation unit 24 may estimate the score of the user by inputting an attribute data set including factual attributes and estimated attributes relating to the target user to the nth model trained by the nth learning unit. In this embodiment, for the same group of users, a combination of services may be adopted such that there is a correlation between the score corresponding to the risk in the nth service and the score corresponding to the nth service.

[0055] In this embodiment, the nth learning unit may determine two or more models from a group of models including the (n-1)th model and the (n-2)th model, and train the nth model, which has some of the parameters of each of the determined models as fixed parameters, with the nth training data. In this case, the nth learning unit may determine two or more models from the group of models, including at least the (n-1)th model.

[0056] In this embodiment, a second service different from the first service, or an nth service different from the first, second, ..., n-1th services, may be determined based on the respective attribute data sets of user groups that have usage history or transaction history in each service. For example, a second service different from the first service may be another service that includes a predetermined proportion of users similar to or matching the attribute data sets of users of the first service. Here, the similarity between users may be based on the proportion of attribute data common to users, whether predetermined attribute data is common or not, whether there are more than a predetermined number of combinations in which the embedded representations for users of each service are similar, etc. Also, for example, an nth service different from the n-1th service may be another service that includes a predetermined proportion of users similar to or matching the attribute data sets of users of at least the n-1st service.

[0057] <<Effect>> According to this embodiment, by retraining a machine learning model that learns based on the payment status of a deferred payment service and can output a credit score using data from a service different from the deferred payment service, it becomes possible to improve the accuracy of estimating an individual's creditworthiness even for services with a relatively small amount of training data.

[0058] <Second Embodiment> This embodiment relates to a technique for training a model using the attributes of users (including their transaction history) belonging to one of two different user segments that use a common service as a dataset, and then retraining the model using the dataset of the other segment.

[0059] More specifically, this embodiment describes a method of applying the technology of this disclosure to train a machine learning model for calculating the default risk of users related to financial services (for example, the deferred payment risk described in the first embodiment) using the attributes of users in one of several user segments that use commercial transaction-related services as a dataset, and then retraining the machine learning model using datasets of users in other segments that use the commercial transaction-related services. However, the technology of this disclosure can be broadly applied to techniques for calculating user scores for different services using a model generated with multiple datasets belonging to multiple different segments in a given service, and the scope of application of this disclosure is not limited to the examples shown in the embodiments. For example, the technology of this disclosure may be implemented to enable risk estimation for multiple different services belonging to fields other than finance.

[0060] <<System Configuration>> Figure 9 is a schematic diagram showing the configuration of the information processing system according to this embodiment. In the information processing system according to this embodiment, the information processing device 1b and one or more service provision systems 5 are connected to each other in a communicative manner. The user is a user of the services provided by the service provision system 5, and receives the services by accessing the service provision system 5 from the user terminal. Here, the information processing device 1b represents a learning device, an estimation device, or a combination thereof. The configuration of the information processing system according to this embodiment is substantially the same as the configuration of the information processing system according to the first embodiment described with reference to Figure 1, except that the functional configuration of the information processing device 1b, which will be described later with reference to Figure 10, differs from that of the information processing device 1 according to the first embodiment. Therefore, the description of the common parts will be omitted.

[0061] In this embodiment, we describe a method of training a machine learning model to calculate the default risk of users related to financial services (corresponding to the "second service" described later), which are in a different business classification or domain, using the attributes of users in one of the multiple user segments that use commercial transaction-related services (corresponding to the "first service" described later) among the various services provided by the service provision system 5 as a dataset, and then retraining the model to calculate the default risk of users in other segments that use the aforementioned commercial transaction-related services, using the dataset of users in those other segments. Here, commercial transaction-related services are services that belong to the "commercial transaction" category in business classification or domain (segment, category, domain), such as online shopping services or online reservation services. Financial services are services that belong to the "finance" category in business classification or domain, such as deferred payment services. Furthermore, retraining refers to transfer learning, additional learning, etc., depending on the learning method. In this embodiment, the group of services may be a group of services that share a common membership program (user identifier), a group of services that share a common points program (electronic value such as points that can be awarded or used), or a group of services that provide benefits usable in one service in conjunction with the use of the other service.

[0062] The information processing device 1b relating to this disclosure uses a machine learning model that takes user attribute data obtained in the first service (in this embodiment, the online shopping service) as input and outputs a credit score (in this embodiment, a score indicating deferred payment risk) for the second service (in this embodiment, the deferred payment settlement service). Details of the attribute data used as input here, including the fact that it may include factual attribute data and estimated attribute data, have been explained in the first embodiment and will therefore be omitted. Similarly, the framework that can be used to generate / update the machine learning model has been explained in the first embodiment and will therefore be omitted.

[0063] Figure 10 is a diagram illustrating the schematic functional configuration of the information processing device 1b according to this embodiment. The information processing device 1b functions as an information processing device comprising a fact attribute determination unit 21, an estimated attribute determination unit 22, a learning unit 25, a score estimation unit 24, and a segmentation unit 26, by having a program recorded in the storage device 14 read into the RAM 13, executed by the CPU 11, and controlling each hardware component of the information processing device 1b. In this embodiment and other embodiments described later, each function of the information processing device 1b is executed by the general-purpose processor CPU 11, but some or all of these functions may be executed by one or more dedicated processors.

[0064] The fact attribute determination unit 21, the estimated attribute determination unit 22, and the score estimation unit 24 are as described in the first embodiment, so their explanation will be omitted. On the other hand, the processing content of the learning unit 25 differs from that of the learning unit 23 described in the first embodiment, and the segmentation unit 26 is a functional unit that did not appear in the first embodiment, so the learning unit 25 and the segmentation unit 26 will be explained below.

[0065] The segmentation unit 26 segments users of the first service (in this embodiment, the online shopping service) to determine which users belong to the first segment and which belong to the second segment. Here, the segmentation unit 26 segments users of the first service based on the user attribute data obtained in the first service.

[0066] The type of attribute data referenced for segmentation is not limited, but in this embodiment, an example is described in which the frequency of use of the first service by each user over a predetermined period is used as the attribute data referenced for segmentation. In this embodiment, the segmentation unit 26 obtains and references the frequency of use of the first service (e.g., the number of times) over a predetermined period from the first service, places users whose frequency of use is above a predetermined threshold into a first segment consisting of users who conduct transactions frequently (a segment consisting of heavy users who conduct transactions frequently), and places users whose frequency of use is below a predetermined threshold into a second segment consisting of users who conduct transactions infrequently (a segment consisting of light users who conduct transactions infrequently).

[0067] However, as mentioned above, the types of attribute data referenced for segmentation are not limited. For example, the attribute data referenced for segmentation may be attribute data relating to service usage history or transaction history. Alternatively, for example, the attribute data referenced for segmentation may include the login frequency of each user to the first service over a predetermined period. When the login frequency of each user to the first service is referenced as attribute data, the segmentation unit 26 can obtain and reference the login frequency (e.g., number of logins) to the first service over a predetermined period from the first service, and place users whose login frequency is above a predetermined threshold into the first segment (a segment consisting of active users), and users whose login frequency is below a predetermined threshold into the second segment (a segment consisting of dormant users). Furthermore, the segmentation unit 26 may perform heuristic segmentation, for example, for users with a service usage history, by determining the user's segment uniquely based on rules according to attribute data including factual attribute data and estimated attribute data.

[0068] Furthermore, the segmentation unit 26 may encode the attribute data of each user with a service usage history into an embedded representation (vector representation), and identify similar / proximity user groups as segments based on the Euclidean distance or cosine similarity between the embedded representations. In other words, the segmentation unit 26 treats users who are close and similar in some feature space (embedded space, latent space) as users belonging to a single segment. Here, the method used for encoding may be to obtain the embedded representation using a pre-trained model (neural network or Transformer-based pre-trained model such as BERT), or to obtain the embedded representation by embedding each user as a node in a graph network. The learning unit 25, described later, may learn using a dataset of user groups belonging to a segment segmented in this way, then preferentially select segments that are close (or have high similarity) to the segment in question, and retrain using a dataset of user groups belonging to the selected segments.

[0069] The learning unit 25 generates and / or updates the score estimation model generated by the score estimation unit 24. The score estimation model is a machine learning model that, when one or more attribute data (attribute data group) relating to a target user is input, outputs a credit score indicating the degree of risk that the target user will not properly settle the deferred payment when using the deferred payment service. This embodiment differs from the first embodiment in that the user attribute data included in the training data is attribute data obtained in a first service (in this embodiment, an online shopping service), and the score is the credit score of the same user in a second service (in this embodiment, a deferred payment service) that is different from the first service.

[0070] In this embodiment, the learning unit 25 comprises at least a first learning unit 25a and a second learning unit 25b.

[0071] The first learning unit 25a trains the first model (first score estimation model) with first training data, which includes a combination of first attribute data sets relating to users belonging to the first segment among users of the first service, and labels corresponding to a score (user credit score in the second service) that changes according to the degree of risk borne by the provider of the second service when the user uses a second service different from the first service. The second service is a different service from the first service, and as described above, in this embodiment, the first service is an online shopping service and the second service is a deferred payment service. The risk borne by the provider of the second service is deferred payment risk, which is determined based on the payment history in the deferred payment service and represents the risk that the user will not properly settle the deferred payment when the user uses the deferred payment service using some indicator (in this embodiment, a label that changes according to the magnitude of the deferred payment risk).

[0072] The second learning unit 25b trains a second model (second score estimation model) which has some of the configuration of the first model (intermediate layers, nodes, and other parameters such as weights) that has been trained by the first learning unit 25a as fixed parameters, with second training data which includes a combination of a second set of attribute data relating to users of the first service that belong to the second segment among the users of the first service, and labels corresponding to a score that changes according to the degree of risk that the provider of the second service would incur if the user were to use the second service.

[0073] The specific framework for training a second model (second score estimation model), which has some of the parameters of the configuration of the first model trained by the first learning unit 25a as fixed parameters, with respect to the second training data is the same as that described in the first embodiment with reference to Figures 3 to 5, so the explanation will be omitted.

[0074] <<Processing Flow>> Next, the processing flow performed by the information processing device according to this embodiment will be described. Note that the specific processing content and processing order described below are examples for implementing this disclosure. The specific processing content and processing order may be appropriately selected depending on the embodiment of this disclosure.

[0075] Figure 11 is a flowchart showing the machine learning processing flow according to this embodiment. The processing shown in this flowchart is executed periodically or at times specified by the administrator.

[0076] In step S300, user group segmentation is performed. The segmentation unit 26 segments users who use the first service based on the attribute data of the users who use the first service, thereby determining which users belong to the first segment and which belong to the second segment. After that, the process proceeds to step S301.

[0077] In steps S301 and S302, a first score estimation model is generated. The first learning unit 25a creates first training data that includes a combination of first attribute data for users of the first service who belong to the first segment, and labels corresponding to a score that changes according to the degree of risk borne by the provider of the second service if the user uses a second service different from the first service (step S301). Then, the first learning unit 25a performs training processing on the first score estimation model based on the created first training data (step S302). After that, the process proceeds to step S303.

[0078] In steps S303 and S304, a second score estimation model is generated. The second learning unit 25b creates second training data that includes a combination of a second set of attribute data relating to users of the first service who belong to the second segment, and labels corresponding to a score that changes according to the degree of risk that the provider of the second service would incur if such users used the second service (step S303). Then, the second learning unit 25b retrains the second score estimation model, which has some of the parameters of the first score estimation model generated in step S302 as fixed parameters, using the second training data, thereby constructing a second score estimation model to be used for score estimation by the score estimation unit 24 (step S304). After that, the process shown in this flowchart is completed.

[0079] The score estimation process flow according to this embodiment is substantially the same as the process flow described with reference to Figure 8 in the first embodiment, so its explanation will be omitted.

[0080] <<Variations>> In the second embodiment described above, an example was described in which users of the first service are divided into two segments and learning is performed in two stages. However, the number of segments to be divided and the number of learning stages are not limited to the example described above. Similar to the first embodiment, the learning unit 25 may further include a third learning unit, a fourth learning unit, ... an nth (n>2) learning unit. Here, the nth learning unit trains an nth model (the nth score estimation model) which has some of the parameters of the nth-1 model trained by the nth-1 learning unit as fixed parameters, etc., with the nth training data, which includes a combination of the nth attribute data set relating to users included in the nth segment among the users of the first service, and a label corresponding to a score that changes according to the degree of risk that the provider of the second service would incur if the user were to use a second service different from the first service. At this time, the score estimation unit 24 may estimate the score of a user by inputting an attribute data set including factual attributes and estimated attributes relating to the target user to the nth model trained by the nth learning unit. For example, as a variation of the embodiment described above, the segments may be divided into three segments: "heavy users," "medium users," and "light users," and learning may be performed in three stages.

[0081] Furthermore, in this embodiment as well, the nth learning unit may determine two or more models from a group of models including the (n-1)th model and the (n-2)th model, and train the nth model, which has some of the parameters of each of the determined models as fixed parameters, with the nth training data. In this case, the nth learning unit may determine two or more models from the group of models, including at least the (n-1)th model.

[0082] <<Effect>> For users belonging to a certain segment (for example, light users of the first service), the accuracy of score estimation may be insufficient due to insufficient quantity or quality of attribute data (for example, relatively small quantity, missing attribute data, etc.). Therefore, in this embodiment, a first score estimation model is created using attribute data of users in the first segment who have sufficient quantity or quality of attribute data (for example, heavy users of the first service), and a second score estimation model is created by fixing a part of the first score estimation model and then retraining it using attribute data of users in the second segment who have insufficient quantity or quality of data (for example, light users of the first service). This makes it possible to improve the accuracy of estimating individual creditworthiness even for users in segments with relatively small training datasets (in this case, light users). [Explanation of Symbols]

[0083] 1,1b Information Processing Device

Claims

1. A first learning means for training a first model with first training data, which includes a combination of first attribute data sets relating to users belonging to a first segment among users of the first service, and labels corresponding to scores that change according to the degree of risk borne by the provider of the second service if the user uses a second service different from the first service; A second learning means for training a second model having some parameters of the configuration of the trained first model as fixed parameters, with second training data including a combination of a second set of attribute data relating to users of the first service who belong to a second segment different from the first segment, and a label corresponding to a score that changes according to the degree of risk borne by the provider of the second service when the user uses the second service, A trust learning device equipped with the following features.

2. The system further includes segmentation means for determining users included in the first segment and users included in the second segment by segmenting users who use the first service. The trust learning device according to claim 1.

3. The segmentation means performs segmentation of users of the first service based on the attribute data of users of the first service. The trust learning device according to claim 2.

4. The segmentation means segments users of the first service based on the frequency of use of the first service during a predetermined period. The trust learning device according to claim 3.

5. The second service mentioned above is a deferred payment service. The risk borne by the provider of the second service is the risk of the user not properly settling the deferred payment when the user uses the deferred payment service, as determined based on the payment history in the deferred payment service. The trust learning device according to claim 1.

6. The second service is a service in which a correlation can be confirmed or inferred between the risk borne by the provider of the first service when the user uses the first service and the risk borne by the provider of the second service when the user uses the second service. The trust learning device according to claim 1.

7. The first model and the second model are gradient boosting decision trees, The second learning means causes the second model, which has some of the nodes of the first model as fixed parameters, to learn the second training data. The trust learning device according to claim 1.

8. The first model and the second model are neural networks. The second learning means causes the second model, which has some of the parameters of the intermediate layer of the trained first model as fixed parameters, to learn the second training data. The trust learning device according to claim 1.

9. The second learning means is a second model in which a plurality of copies of the trained first model are arranged in parallel, and the second model has some of the parameters of the intermediate layers included in the second model as fixed parameters, and the second means trains the second training data on the second model. The trust learning device according to claim 8.

10. A fact attribute determination means that determines factual attributes that can be confirmed as facts about the user based on user-provided data or the user's historical data provided by the user, The system further comprises: an estimated attribute determination means for determining estimated attributes estimated for a user based on at least the factual attributes relating to the user; The trust learning device according to claim 1.

11. On the computer, A first learning step involves training a first model with first training data, which includes a combination of first attribute data sets relating to users belonging to a first segment among users of the first service, and labels corresponding to scores that change according to the degree of risk borne by the provider of the second service if the user uses a second service different from the first service. A second learning step involves training a second model, which has some parameters of the configuration of the trained first model as fixed parameters, with second training data including a combination of a second set of attribute data relating to users of the first service who belong to a second segment different from the first segment, and a label corresponding to a score that changes according to the degree of risk incurred by the provider of the second service when such users use the second service. A trust-based learning method that enables execution.

12. Computers, A first learning means for training a first model with first training data, which includes a combination of first attribute data sets relating to users belonging to a first segment among users of the first service, and labels corresponding to scores that change according to the degree of risk borne by the provider of the second service if the user uses a second service different from the first service; A second learning means for training a second model having some parameters of the configuration of the trained first model as fixed parameters, with second training data including a combination of a second set of attribute data relating to users of the first service who belong to a second segment different from the first segment, and a label corresponding to a score that changes according to the degree of risk borne by the provider of the second service when the user uses the second service, A program that makes something function as such.

13. A credit estimation device comprising estimation means for estimating a second score to be set for a target user by inputting a group of attribute data relating to the target user to the second model which has been trained by the learning method described in claim 11.

14. On the computer, A credit estimation method comprising performing an estimation step of estimating a second score to be set for a target user by inputting a group of attribute data relating to the target user into the second model that has been trained by the learning method described in claim 11.

15. Computers, A program that functions as an estimation means for estimating a second score to be set for a target user by inputting a group of attribute data relating to the target user into the second model that has been trained by the learning method described in claim 11.

Images