JP2023159664A - Information processing device, information processing method, and information processing program - Google Patents

Abstract

To allow for appropriately generating a learning model that is created using a user's history information.SOLUTION: An information processing device comprises an exclusion unit and a learning unit. The exclusion unit excludes history information satisfying a second condition from a user's history information narrowed down by a first condition. The learning unit generates a learning model using the user's history information obtained by having the exclusion unit exclude the history information satisfying the second condition.SELECTED DRAWING: Figure 1

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

BACKGROUND ART Conventionally, a technique for generating a learning model by machine learning using user information including user history information is known. For example, Patent Document 1 discloses a technique for generating a learning model that calculates an index value regarding whether or not a user will default on a debt based on the user's history information.

JP 2019-185595 Publication

However, the user's historical information includes various types of information, and if a learning model is generated using the user's historical information as is, discriminatory judgments may be made, and there is still room for improvement. be.

The present application has been made in view of the above, and provides an information processing device, an information processing method, and an information processing program that can appropriately generate a learning model generated using history information of a user. The purpose is to

The information processing device according to the present application includes an exclusion section and a learning section. The exclusion unit excludes history information that satisfies the second condition from the history information of the users narrowed down based on the first condition. The learning unit generates a learning model using the user's history information from which the history information satisfying the second condition has been excluded by the exclusion unit.

According to one aspect of the embodiment, it is possible to appropriately generate a learning model using history information of a user.

FIG. 1 is a diagram for explaining an information processing method performed by an information processing apparatus according to an embodiment. FIG. 2 is a diagram illustrating an example of the configuration of the information processing system according to the embodiment. FIG. 3 is a diagram illustrating an example of the configuration of the information processing device according to the embodiment. FIG. 4 is a diagram illustrating an example of a user information table stored in the user information storage unit according to the embodiment. FIG. 5 is a flowchart illustrating an example of information processing by the processing unit of the information processing device according to the embodiment. FIG. 6 is a hardware configuration diagram showing an example of a computer that implements the functions of the information processing device according to the embodiment.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Below, an information processing apparatus, an information processing method, and a form for implementing an information processing program (hereinafter referred to as an "embodiment") according to the present application will be described in detail with reference to the drawings. Note that the information processing apparatus, information processing method, and information processing program according to the present application are not limited to this embodiment. Moreover, each embodiment can be combined as appropriate within the range that does not conflict with the processing contents. Further, in each of the embodiments below, the same parts are given the same reference numerals, and redundant explanations will be omitted.

[1. An example of information processing]
First, with reference to FIG. 1, an information processing method performed by an information processing apparatus according to an embodiment will be described. FIG. 1 is a diagram for explaining an information processing method performed by an information processing apparatus according to an embodiment.

The information processing device 1 shown in FIG. 1 cooperates with the terminal devices of each user U ₁ , U ₂ , . . . , _Un , and provides various services online to each user U ₁ , U ₂ , . , _Un , and is realized by one or more servers or a cloud system. In the following, when each of the users U ₁ , U _{2 ,} .

For example, the services provided by the information processing device 1 include search service, payment service, Internet connection, SNS (Social Networking Service), electronic commerce (EC), online games, online banking, online trading, accommodation and ticket reservations. , video/music distribution, news, maps, route searches, route guidance, route information, operation information, weather forecasts, and other services may be provided. The information processing device 1 may cooperate with various servers that provide the above-mentioned services, and may mediate the services or may be in charge of processing the services.

The information processing device 1 has a storage unit that stores user information that is information about each user U, and generates a learning model based on the user information of each user U. The user information includes history information including information such as the usage history of the above-mentioned services by the user U, attribute information of the user U, and the like.

The attribute information of the user U is, for example, information on demographic attributes such as the user U's age, gender, occupation, and place of residence. User U's history information includes search history information that is history information of user U's search queries, payment history information that is usage history information of user U's payment services, and history of web pages accessed by user U. It includes information such as browsing history information, but may also include other information.

The learning model generated by the information processing device 1 is, for example, a prediction model that inputs user information of user U and outputs a predicted score regarding a specific behavior of user U. This is a loan loss prediction model that outputs a loan loss score, which is a score related to loan loss risk in payment services. Although the following description assumes that the learning model generated by the information processing device 1 is a bad debt prediction model, the learning model generated by the information processing device 1 may be a learning model generated based on search history information. However, the present invention is not limited to bad debt prediction models.

As shown in FIG. 1, the information processing device 1 performs a narrowing process of narrowing down history information to be extracted from history information included in user information stored in a storage unit for each attribute based on a first condition (step S1). . The attributes are, for example, psychographic attributes, and are objects that may be of interest to the user U, such as golf, baseball, soccer, fashion, travel, politics, economy, technology, and the like. Note that in the following, the history information narrowed down by the information processing device 1 based on the first condition may be referred to as first history information.

The first condition is, for example, the history information of a search query that is highly common to a plurality of users U having a specific attribute. For example, when the attribute is golf, the information processing device 1 treats the user U who has made a payment at a golf course or purchased golf equipment as a user U who has an interest in golf, that is, the attribute "golf". Identified as user U who has. Then, the information processing device 1 extracts history information of search queries that are highly common among a plurality of users U having the attribute "golf".

A search query that has a high degree of commonality among multiple users U having the attribute "golf" is, for example, a search word such as "golf course reservation," "golf ball," "golf wear," "iron," or "driver." is a search query containing . The information processing device 1 extracts a preset number of search words in descending order of commonality among the users U, and sets the extracted search words as first history information.

The information processing device 1 calculates commonality among a plurality of users U for each search word based on, for example, the number of users U who have searched for the same search word and the number of search queries that include the same search word. Judgment is made.

For example, the information processing device 1 calculates a value obtained by weighting and adding the number of users U who searched the same search word, the number of search queries including the same search word, etc. Calculated as a value indicating the commonality in The information processing device 1 extracts a preset number of search words in descending order of the value indicating commonality, and sets the extracted plurality of search words as first history information.

Next, the information processing device 1 performs an exclusion process of excluding history information that satisfies the second condition from the first history information for each attribute (step S2). The second condition is, for example, a condition that the information is included in a rejection list that lists information to be excluded, or a condition that it is not included in a permission list that lists information that is not to be excluded. Note that, hereinafter, history information obtained by the information processing device 1 excluding history information that satisfies the second condition may be referred to as second history information.

The history information that satisfies the second condition is, for example, consideration-required personal information that is information that requires consideration for the user U. The rejection list includes, for example, sensitive personal information as information to be excluded. Further, the permission list includes, for example, information that does not correspond to special consideration personal information as information that is not to be excluded. The sensitive personal information is, for example, information defined in the Personal Information Protection Law or sensitive information.

In the denial list and the permission list, information on exclusion targets and information on non-exclusion targets may be vectorized. In this case, the second condition is, for example, a condition that the vector is similar to exclusion target information included in the rejection list, or a condition that the vector is not similar to non-exclusion target information included in the permission list.

Next, the information processing device 1 collects second history information that is history information excluding history information that satisfies the second condition in step S2, user information of a plurality of users U who used the payment service in the past, etc. A learning model generation process for generating a bad debt prediction model is performed using the learning model (step S3).

In the process of step S3, the information processing device 1 performs an attribute estimation model generation process (step S3-1) and a predictive model generation process (S3-2). In the attribute estimation model generation process of step S3-1, the information processing device 1 generates an attribute estimation model that includes a plurality of search words included in the second history information in its features and outputs an attribute score for each attribute. The attribute score increases as the possibility that the user U has the attribute to be estimated by the attribute estimation model increases, but the attribute score is not limited to this example.

The attribute estimation model for each attribute is, for example, an attribute estimation model for the attribute "golf", an attribute estimation model for the attribute "baseball", an attribute estimation model for the attribute "soccer", an attribute estimation model for the attribute "fashion", and an attribute estimation model for the attribute "travel". attribute estimation models, etc. The features of the attribute estimation model include, for example, attribute information, but are not limited to such examples.

In the predictive model generation process of step S3-2, the information processing device 1 uses information for learning including the user information of the user U who used the payment service in the past and the attribute score for each attribute obtained by the attribute estimation model. As information, a bad debt prediction model is generated that includes in its features each attribute for which an attribute score can be obtained using the attribute estimation model.

The user information of the user U who used the service in the past includes, for example, attribute information of the user U, browsing history information of the user U, history information of the payment service of the user U, and the like. The history information of the payment service includes information indicating the presence or absence of bad debts. Information indicating the presence or absence of bad debt is used as label data (correct data) in the predictive model generation process of step S3-2.

The information processing device 1 satisfies the second condition with an attribute score obtained by an attribute estimation model whose feature is information included in second history information that is history information after excluding history information that satisfies the second condition. A score difference is calculated for each attribute, which is the difference from the attribute score obtained by the attribute estimation model whose background is information included in the history information before excluding the history information.

The information processing device 1 can also generate a bad debt prediction model without using the attribute scores of the attributes for which the score difference described above is greater than or equal to a predetermined value. As a result, since the attribute score output by the attribute estimation model, which deteriorates the accuracy of the attribute score by excluding personal information requiring special consideration, is not taken into account, it is possible to prevent the accuracy of the bad debt score from deteriorating.

In this way, the information processing device 1 according to the embodiment excludes the history information that satisfies the second condition from the history information of the user U narrowed down based on the first condition, and excludes the history information that satisfies the second condition. Using the history information of the excluded user U, learning models such as an attribute estimation model and a bad debt prediction model are generated. Thereby, the information processing device 1 can appropriately generate a learning model generated using the user U's history information.

Hereinafter, the configuration of an information processing system including the information processing apparatus 1 that performs such processing will be described in detail.

[2. Information processing system configuration]
FIG. 2 is a diagram illustrating an example of the configuration of the information processing system according to the embodiment. As shown in FIG. 2, the information processing system 100 according to the embodiment includes an information processing device 1 and terminal devices 2 ₁ , 2 ₂ , . . . , 2 _n .

Terminal device 2 ₁ is used by user U ₁ , terminal device 2 ₂ is used by user U 2 , and terminal device _{2 n} is used by user U _n . In the following, when each of _the terminal devices 2 ₁ , 2 ₂ , .

The information processing device 1 and the terminal device 2 are connected to each other via a network N so that they can communicate with each other by wire or wirelessly. Note that the information processing system 100 shown in FIG. 2 may include a plurality of information processing apparatuses 1.

The terminal device 2 is, for example, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a mobile phone, a PDA (Personal Digital Assistant), or the like. Note that the terminal device 2 is not limited to the above example, and may be, for example, a smart watch or a wearable device.

[3. Configuration of information processing device 1]
FIG. 3 is a diagram showing an example of the configuration of the information processing device 1 according to the embodiment. As shown in FIG. 3, the information processing device 1 includes a communication section 10, a storage section 11, and a processing section 12.

[3.1. Communication department 10]
The communication unit 10 is realized by, for example, a NIC (Network Interface Card). The communication unit 10 is connected to the network N by wire or wirelessly, and transmits and receives information to and from various other devices. For example, the communication unit 10 transmits and receives information to and from the terminal device 2 via the network N.

[3.2. Storage unit 11]
The storage unit 11 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 11 includes a user information storage unit 20.

[3.2.1. User information storage unit 20]
The user information storage unit 20 stores various types of user information regarding the user U. FIG. 4 is a diagram showing an example of a user information table stored in the user information storage unit 20 according to the embodiment. In the example shown in FIG. 4, the user information table stored in the user information storage unit 20 includes information on items such as "user ID", "user name", "attribute information", and "history information". include.

“User ID” is an identifier that identifies user U. “User name” is information indicating the name of user U. "Attribute information" is attribute information regarding the attributes of the user U associated with the "user ID." For example, the attribute information is information on demographic attributes such as the user U's gender, age, occupation, and place of residence. Note that the above-mentioned attribute information is only an example, and may include various information such as information on psychographic attributes such as interest, lifestyle, and the like.

"History information" is history information including information such as service usage history, and includes, for example, user U's payment history information, user U's search history information, user U's browsing history information, and the like.

The user U's payment history information is, for example, information about the user U's payment history using a payment service. The payment service is a deferred payment payment service, such as a payment service using a credit card, a payment service using deferred payment electronic money, a bill payment service, a payment service using a deferred payment application, a carrier payment service, etc. The examples are not limited to such examples.

The payment history information of the user U includes information indicating the date and time of lending to the user U, the loan amount, and whether or not there is a default on the loan. The loan amount information is information indicating the amount lent to user U. In a payment service, if a loan is not repaid even after the payment deadline has passed, it is determined that there is a default on the loan, information indicating that there is a default on the loan is set in the payment history information of user U, and if the loan is repaid by the payment deadline, It is determined that there is no bad debt, and information indicating that there is no bad debt is set in the payment history information of user U.

User U's payment history information includes purchase history information regarding products that User U purchased online, at a physical store, or at a physical facility using a payment service; This includes service usage history information regarding services used for a fee at stores or actual facilities.

The purchase history information includes information on products purchased by user U, purchase cost information, purchase date and time, purchase store information, and the like. The service usage history information includes information on services used by user U, information on usage fees, date and time of usage, information on stores used, and the like.

The search history information of the user U is, for example, information on the search history of web content on search sites, information on the search history on various websites, and the like. The user U's search history information is, for example, information on the search history of web content on search sites and portal sites.

User U's browsing history information is information on user U's browsing history of web content, and includes various information such as user U's browsing history of property information, browsing information on leaflets of stores such as convenience stores and supermarkets, etc. Contains information.

Note that the user information may also include movement history information of the user U. The movement history information of the user U is information on the movement history of the user U, and includes, for example, information on the position of the user U for each unit time. Note that the movement history information of the user U may be information on visited stores, facilities, and the like.

[3.3. Processing unit 12]
The processing unit 12 is a controller, and is stored in a storage device (for example, the storage unit 11) inside the information processing device 1 by a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). This is realized by executing various programs (an example of an information processing program) using the RAM as a work area. Furthermore, the processing unit 12 may be partially or entirely realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the processing unit 12 includes an acquisition unit 30, a display processing unit 31, a narrowing down unit 32, an exclusion unit 33, and a learning unit 34, and has information processing functions described below. accomplish or carry out an action. Note that the internal configuration of the processing unit 12 is not limited to the configuration shown in FIG. 3, and may be any other configuration as long as it performs information processing to be described later.

[3.3.1. Acquisition unit 30]
The acquisition unit 30 acquires various information. The acquisition unit 30 acquires various information from the storage unit 11. The acquisition unit 30 acquires various types of information from the user information storage unit 20.

Further, the acquisition unit 30 can also acquire user information, which is information about the user U, from the external device 3 and store the acquired user information in the user information storage unit 20, for example.

[3.3.2. Display processing unit 31]
The display processing section 31 displays various information on the display section 14. For example, the display processing unit 31 can cause the display unit 14 to display information according to an input operation by an operator to the input unit 13.

Further, the display processing unit 31 can display the information processed by the narrowing down unit 32, the information processed by the exclusion unit 33, and the information processed by the learning unit 34 on the display unit 14. In the example shown in FIG. 1, the display unit 14 is provided separately from the information processing device 1, but may be included in the information processing device 1.

[3.3.3. Narrowing down section 32]
The narrowing down section 32 performs a narrowing down process of narrowing down history information to be extracted from the history information included in the user information stored in the user information storage section 20 for each attribute based on a first condition. The attributes to be narrowed down are, for example, psychographic attributes, and are objects that may be of interest to the user U, such as golf, baseball, soccer, fashion, travel, politics, economy, technology, and the like.

The first condition is, for example, that history information has a high degree of commonality among a plurality of users U who have attributes to be narrowed down. Here, history information with high commonality is, for example, history information of search queries with high commonality.

The narrowing down unit 32 determines, for each attribute, a plurality of users U having attributes to be narrowed down, based on search history information included in the history information, for example. For example, if the attribute to be narrowed down is golf, the narrowing down unit 32 may classify the user U who has made a payment at a golf course or purchased golf equipment from a user U who has an interest in golf, that is, the attribute " Identified as user U who has "Golf".

In addition, for example, instead of or in addition to the payment history information included in the history information, the narrowing down unit 32 selects a plurality of users U having the attributes to be narrowed down based on the viewing history information included in the history information. It is also possible to judge each time. For example, when the attribute to be narrowed down is golf, the narrowing down unit 32 can identify a user U who frequently views web content related to golf as a user U having the attribute "golf."

Then, the narrowing down unit 32 extracts history information of search queries that are highly common among a plurality of users U having the attribute "golf". A search query that has a high degree of commonality among multiple users U having the attribute "golf" is, for example, a search word such as "golf course reservation," "golf ball," "golf wear," "iron," or "driver." is a search query containing . The narrowing down unit 32 extracts a preset number of search words in descending order of commonality among a plurality of users U, and sets the extracted plurality of search words as first history information.

The narrowing-down unit 32 selects a plurality of results based on, for example, a first number that is the number of users U who have searched for the same search word, and a second number that is the number of search queries that include the same search word. Commonality among users U is determined for each search word. The first number is the number of users U who have searched for the same search word among a plurality of users U who have attributes to be narrowed down. Moreover, the second number is the number of search queries that include the same search word among the search queries of a plurality of users U who have attributes to be narrowed down.

For example, the narrowing down unit 32 calculates a value obtained by weighting and adding the first number, the second number, etc. as a value indicating the commonality among the plurality of users U. The narrowing down unit 32 extracts a preset number of search words in descending order of the value indicating commonality, and sets the extracted plurality of search words as first history information.

Note that the narrowing down section 32 can, for example, narrow down browsing history information included in the history information instead of or in addition to the search history information included in the history information, and may narrow down other history information. You can also. When the narrowing down target is browsing history information, the narrowing down unit 32 can narrow down web content or words included in the web content, for example.

[3.3.4. Exclusion part 33]
The exclusion unit 33 obtains second history information by excluding history information that satisfies the second condition from the history information of the users narrowed down by the first condition by the narrowing down unit 32.

The second condition is, for example, a condition that the information is included in a rejection list that lists information to be excluded, or a condition that it is not included in a permission list that lists information that is not to be excluded. The second condition is set for each attribute. That is, a deny list and a permit list are set for each attribute.

The history information that satisfies the second condition is, for example, consideration-required personal information that is information that requires consideration for the user U. The rejection list includes, for example, sensitive personal information as information to be excluded. Further, the permission list includes, for example, information that does not correspond to special consideration personal information as information that is not to be excluded.

The sensitive personal information is, for example, historical information that includes information defined in the Personal Information Protection Act, or sensitive historical information. Personal information that requires special consideration as defined in the Personal Information Protection Act includes the person's race, creed, social status, medical history, criminal history, the fact that he or she has been harmed by a crime, and any other information that may cause unfair discrimination, prejudice, or other disadvantages to the person. This is personal information that includes descriptions that require special consideration when handling to prevent this from occurring. For example, the sensitive personal information is a search history regarding a specific hospital gown. In addition, sensitive information includes, for example, searches related to economic conditions such as poverty.

Note that even if the information is sensitive, the exclusion unit 33 treats it as information that has a specific relationship with bad debts, such as information related to economic conditions such as annual income and including descriptions required for credit screening. It may not be included in the history information that satisfies the second condition. Thereby, it is possible to suppress deterioration in the accuracy of the bad debt score output from the bad debt prediction model.

In the denial list and the permission list, information on exclusion targets and information on non-exclusion targets may be vectorized. In this case, the second condition is, for example, a condition that the vector is similar to exclusion target information included in the rejection list, or a condition that the vector is not similar to non-exclusion target information included in the permission list. In this case, the exclusion unit 33 vectorizes each search word included in the first history information, and compares the vectorized search word with the rejection list and the permission list.

Information on exclusion targets and information on non-exclusion targets are vectorized using a natural language processing method such as Word2Vec, for example. For example, information on excluded targets and information on non-excluded targets is vectorized using an interest model. The interest model is a model that has been trained by learning the features of each of a plurality of search words, assuming that two or more search words that meet predetermined conditions have similar features.

Two or more search words that satisfy predetermined conditions are multiple search words included in the same search query, or multiple search queries sent from the terminal device 2 by the same user U within a predetermined time. is a search word included in . The trained model is, for example, a DSSM (Deep Structured Semantic Model) that uses LSTM (Long Short-Term Memory), which is a type of RNN (Recurrent Neural Network), also called a recurrent neural network, for vector generation (for example, distributed representation generation). ), but is not limited to this example.

Note that the exclusion unit 33 can, for example, target browsing history information included in the history information instead of or in addition to the search history information included in the history information, and exclude other history information. It can also be used as a processing target. When the narrowing-down target is browsing history information, the exclusion unit 33 can target web content or words included in the web content, for example.

[3.3.5. Learning Department 34]
The learning unit 34 acquires history information from which history information that satisfies the second condition is excluded from the history information of the user U narrowed down based on the first condition and users of the plurality of users U who have used the payment service in the past. An attribute estimation model and a bad debt prediction model that output an attribute score for each attribute are generated using information such as information.

The learning section 34 includes an estimated model generation section 40 and a predictive model generation section 41. Each of the estimated model generation section 40 and the prediction model generation section 41 will be explained below.

[3.3.5.1. Estimated model generation unit 40]
The estimation model generation unit 40 generates attributes for each attribute using second history information that is history information obtained by excluding history information that satisfies the second condition from the history information of the user U narrowed down based on the first condition. Generate an attribute estimation model that outputs a score.

The attribute estimation model that outputs attribute scores for each attribute has a configuration that includes an attribute estimation model for each attribute, but it may also be a single attribute estimation model that outputs attribute scores for multiple attributes, and other configurations may also be used. It may be. The attribute estimation model for each attribute is generated using different second history information for each attribute.

The attribute estimation model for each attribute is, for example, an attribute estimation model for the attribute "golf", an attribute estimation model for the attribute "baseball", an attribute estimation model for the attribute "soccer", an attribute estimation model for the attribute "fashion", and an attribute estimation model for the attribute "travel". attribute estimation models, etc.

The estimated model generation unit 40 generates an attribute estimation model for each attribute, each of which includes, as features, a plurality of pieces of information including a plurality of search words included in the second history information of the attribute that is the target of the attribute estimation model. The attribute score increases as the possibility that the user U has the attribute to be estimated by the attribute estimation model increases, but the attribute score is not limited to this example. The attribute estimation model for each attribute is a model whose feature is a search word included in mutually different second history information extracted for each attribute.

Based on learning information including user information of a plurality of users U who have attributes to be estimated, the estimation model generation unit 40 generates, for example, a search history in the search history of users U who have attributes to be estimated. A process of generating an attribute estimation model by machine learning is performed for each attribute so that the attribute score of the user U who is similar to the user U is high.

For example, the estimation model generation unit 40 generates a model whose search history is similar to the search history of the user U who has the attribute "golf" based on learning information including user information of a plurality of users U who has the attribute "golf". An attribute estimation model for the attribute "golf" is generated such that the attribute score of the user U becomes high.

The user information included in the learning information includes, for example, user U's search query history information, user U's attribute information, and user U's browsing history information, and the background of the attribute estimation model includes: In addition to the second history information, user U's attribute information and browsing history information are included, but the information is not limited to such examples. For example, the user information included in the learning information may be information that does not include one or both of user U's attribute information and user U's browsing history information, or may include other information. good.

[3.3.5.2. Prediction model generation unit 41]
The prediction model generation unit 41 receives information including the attribute score for each attribute output from the attribute estimation model, and generates a prediction model that outputs a prediction score regarding a specific behavior of the user U.

The prediction model is, for example, a loan loss prediction model that inputs information including an attribute score for each attribute output from an attribute estimation model and outputs a score related to loan loss risk, but is not limited to such an example. For example, the prediction model may be a model that outputs a score indicating the possibility of clicking on an advertisement or a model that outputs a score indicating the possibility of purchasing a product.

The predictive model generation unit 41 uses learning information including, for example, user information of multiple users U who have used the payment service in the past and attribute scores for each attribute output from the attribute estimation model to predict bad debts. Generate the model. The background of the bad debt prediction model includes, for example, user U's attribute information, user U's browsing history information, and each attribute from which an attribute score is obtained by the attribute estimation model.

The user information of the user U who used the service in the past and is included in the learning information includes, for example, the attribute information of the user U, the browsing history information of the user U, and the information of the payment service of the user U. Contains history information, etc. The history information of the payment service includes information indicating the presence or absence of bad debts. Information indicating the presence or absence of bad debt is used as label data (correct data). Note that the background of the bad debt prediction model is not limited to the above example as long as it includes an attribute for which an attribute score can be obtained by the attribute estimation model.

The predictive model generation unit 41 inputs the user information of a plurality of users U who used the payment service in the past into the attribute estimation model for each attribute, and calculates the attribute score for each attribute output from the attribute estimation model for each attribute. get.

Then, the predictive model generation unit 41 generates information including attribute scores for each attribute of each user U who used the payment service in the past, and information indicating whether or not each user U who used the payment service in the past had bad debts. Based on this, a bad debt prediction model is generated which takes as input information including the attribute score for each attribute of user U and outputs the bad debt score.

The predictive model generation unit 41 calculates the second condition and the attribute score obtained by the attribute estimation model whose feature is information included in the second history information, which is the history information after excluding the history information that satisfies the second condition. A score difference is calculated for each attribute, which is the difference between the attribute score and the attribute estimation model whose feature is the information included in the history information before excluding the history information that satisfies the requirement.

The prediction model generation unit 41 can also generate a bad debt prediction model without using the attribute scores of the attributes for which the score difference is greater than or equal to a predetermined value. As a result, since the attribute score output by the attribute estimation model, which deteriorates the accuracy of the attribute score by excluding personal information requiring special consideration, is not taken into account, it is possible to prevent the accuracy of the bad debt score from deteriorating.

[4. Processing procedure]
Next, a procedure of information processing by the processing unit 12 of the information processing device 1 according to the embodiment will be described. FIG. 5 is a flowchart illustrating an example of information processing by the processing unit 12 of the information processing device 1 according to the embodiment.

As shown in FIG. 5, the processing unit 12 of the information processing device 1 selects one attribute from a plurality of target attributes (step S10). Next, for the attribute selected in step S10, the processing unit 12 performs a narrowing process to narrow down the history information to be extracted from the history information using a first condition (step S11). The first condition is that history information of search queries that are highly common to a plurality of users U having the attribute selected in step S10 is included.

Next, the processing unit 12 performs an exclusion process to exclude history information that satisfies the second condition from the history information of the users narrowed down based on the first condition in step S11 (step S12). The second condition is, for example, a condition that the information is included in a rejection list that lists information to be excluded, or a condition that it is not included in a permission list that lists information that is an exclusion target.

Next, the processing unit 12 determines whether there are any unselected attributes (step S13). When the processing unit 12 determines that there is an unselected attribute (step S13: Yes), the process proceeds to step S10.

When the processing unit 12 determines that there is no unselected attribute (step S13: No), the processing unit 12 uses the history information of the user U from which the history information that satisfies the second condition for each attribute has been excluded, to select the An attribute estimation model that outputs an attribute score is generated (step S14).

Then, the processing unit 12 receives information including the attribute score for each attribute outputted from the attribute estimation model, and generates a prediction model that outputs a prediction score regarding a specific behavior of the user U (step S15). When the processing of step S14 is completed, the processing unit 12 ends the processing shown in FIG. 5.

[5. Modified example]
The processing unit 12 can also generate an attribute estimation model that outputs an attribute score for each attribute by vectorizing a plurality of search words included in the second history information. In this case, a plurality of search words included in the search history of user U are vectorized and input into the attribute estimation model.

In the above example, the processing unit 12 generates an attribute estimation model that outputs an attribute score for each attribute, and generates a prediction model using the attribute score for each attribute outputted by the attribute estimation model. It is also possible to generate a predictive model using the second history information as a feature without generating a model.

[6. Hardware configuration]
The information processing apparatus 1 according to the embodiments described above is realized, for example, by a computer 80 having a configuration as shown in FIG. FIG. 6 is a hardware configuration diagram showing an example of a computer 80 that implements the functions of the information processing device 1 according to the embodiment. The computer 80 includes a CPU 81, a RAM 82, a ROM (Read Only Memory) 83, an HDD (Hard Disk Drive) 84, a communication interface (I/F) 85, an input/output interface (I/F) 86, and a media interface (I/F). )87.

The CPU 81 operates based on a program stored in the ROM 83 or HDD 84 and controls each part. The ROM 83 stores a boot program executed by the CPU 81 when the computer 80 is started, programs depending on the hardware of the computer 80, and the like.

The HDD 84 stores programs executed by the CPU 81 and data used by the programs. The communication interface 85 receives data from other devices via the network N (see FIG. 2) and sends it to the CPU 81, and sends data generated by the CPU 81 to the other devices via the network N.

The CPU 81 controls output devices such as a display and a printer, and input devices such as a keyboard and a mouse via an input/output interface 86. The CPU 81 obtains data from an input device via the input/output interface 86. Further, the CPU 81 outputs the generated data to an output device via the input/output interface 86.

Media interface 87 reads programs or data stored in recording medium 88 and provides them to CPU 81 via RAM 82. The CPU 81 loads this program from the recording medium 88 onto the RAM 82 via the media interface 87, and executes the loaded program. The recording medium 88 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. etc.

For example, when the computer 80 functions as the information processing device 1 according to the embodiment, the CPU 81 of the computer 80 realizes the functions of the processing unit 12 by executing a program loaded onto the RAM 82. Furthermore, data in the storage unit 11 is stored in the HDD 84 . The CPU 81 of the computer 80 reads these programs from the recording medium 88 and executes them, but as another example, these programs may be acquired from another device via the network N.

[7. others〕
Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed manually. All or part of the process can also be performed automatically using known methods. In addition, information including the processing procedures, specific names, and various data and parameters shown in the above documents and drawings may be changed arbitrarily, unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Furthermore, each component of each device shown in the drawings is functionally conceptual, and does not necessarily need to be physically configured as shown in the drawings. In other words, the specific form of distributing and integrating each device is not limited to what is shown in the diagram, and all or part of the devices can be functionally or physically distributed or integrated in arbitrary units depending on various loads and usage conditions. Can be integrated and configured.

For example, the information processing device 1 described above may be realized by a plurality of server computers, or depending on the function, it may be realized by calling an external platform using an API (Application Programming Interface) or network computing. can be changed flexibly.

Furthermore, the embodiments and modifications described above can be combined as appropriate within a range that does not conflict with the processing contents.

[8. effect〕
As described above, the information processing device 1 according to the embodiment includes the exclusion section 33 and the learning section 34. The exclusion unit 33 excludes history information that satisfies the second condition from the history information of the user U narrowed down based on the first condition. The learning unit 34 generates a learning model using the history information of the user U from which the history information satisfying the second condition has been excluded by the exclusion unit 33. Thereby, the information processing device 1 can appropriately generate a learning model generated using the user U's history information.

Further, the history information is history information of a search query. Thereby, the information processing device 1 can appropriately generate a learning model generated using the history information of the user U's search query.

The information processing device 1 also includes a narrowing down unit 32 that narrows down a search query that is highly common among a plurality of users U having a specific attribute as history information of the users U who satisfy the first condition. Thereby, the information processing device 1 can appropriately generate a learning model generated using the history information of the user U's search query.

Further, the learning unit 34 uses the history information of the user U from which the history information satisfying the second condition has been excluded by the exclusion unit 33, inputs information including the history information of the user U, and outputs an attribute score. It includes an estimated model generation unit 40 that generates an attribute estimation model. Thereby, the information processing device 1 can more appropriately generate a learning model generated using the history information of the user U's search query.

Furthermore, the exclusion unit 33 excludes history information that satisfies a second condition that differs for each attribute from the history information of the user U that has been narrowed down using a first condition that differs for each attribute. The estimation model generation unit 40 generates an attribute estimation model that outputs an attribute score for each attribute using the history information of the user U from which the history information that satisfies the second condition has been excluded by the exclusion unit 33 for each attribute. . The learning unit 34 includes a predictive model generating unit 41 that receives information including the attribute score for each attribute output from the attribute estimation model and generates a predictive model that outputs a predictive score regarding a specific behavior of the user U. . Thereby, the information processing device 1 can more appropriately generate a prediction model generated using the history information of the user U's search query.

Further, the prediction model is a loan loss prediction model that takes as input the attribute score output from the attribute estimation model and outputs the score related to loan loss risk. Thereby, the information processing device 1 can appropriately generate a bad debt prediction model that is generated using the history information of the user U's search query.

Furthermore, the learning unit 34 excludes the attribute score obtained by the attribute estimation model whose feature is information included in the history information after excluding the history information that satisfies the second condition, and the history information that satisfies the second condition. Calculates the difference for each attribute from the attribute score obtained by an attribute estimation model whose feature is information contained in previous history information, and generates a prediction model without using attribute scores for which the difference is greater than a predetermined value. do. Thereby, the information processing device 1 can appropriately generate a learning model generated using the history information of the user U's search query.

Furthermore, the exclusion unit 33 excludes history information that satisfies the second condition from the history information of the user U narrowed down based on the first condition, using a rejection list that lists information to be excluded. Thereby, the information processing device 1 can easily exclude history information that satisfies the second condition.

Furthermore, the exclusion unit 33 excludes history information that satisfies the second condition from the history information of the user U narrowed down based on the first condition, using a permission list that lists non-exclusion target information. Thereby, the information processing device 1 can easily exclude history information that satisfies the second condition.

Further, in the rejection list, the information to be excluded is vectorized, and the exclusion unit 33 excludes information whose vector is similar to the information to be excluded as history information that satisfies the second condition. Thereby, the information processing device 1 can easily exclude history information that satisfies the second condition.

Further, in the permission list, non-exclusion target information is vectorized, and the exclusion unit 33 excludes information other than information whose vector is not similar to the non-exclusion target information as history information that satisfies the second condition. Thereby, the information processing device 1 can easily exclude history information that satisfies the second condition.

Further, the information that is excluded by the exclusion unit as satisfying the second condition is consideration-required personal information that is information that requires consideration for the user U. Thereby, the information processing device 1 can appropriately generate a learning model that suppresses discriminatory judgments.

Further, the information that is excluded by the exclusion unit as satisfying the second condition is consideration-required personal information that does not have a specific relationship with bad debt and that requires consideration for user U. Thereby, the information processing device 1 can appropriately generate a learning model that suppresses discriminatory judgments.

Although the embodiments of the present application have been described above in detail based on the drawings, this is merely an example, and various modifications and improvements can be made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure section of the invention. It is possible to implement the invention in other forms.

Further, the above-mentioned "section, module, unit" can be read as "means", "circuit", etc. For example, the acquisition unit can be read as an acquisition means or an acquisition circuit.

1 Information processing device 2, _{2 1} , 2 ₂ ,..., 2 _n terminal device 3 External device 10 Communication section 11 Storage section 12 Processing section 13 Input section 14 Display section 20 User information storage section 30 Acquisition section 31 Display processing Section 32 Narrowing down section 33 Exclusion section 34 Learning section 40 Estimated model generation section 41 Prediction model generation section 100 Information processing system N network

Claims (15)

an exclusion unit that excludes history information that satisfies a second condition from the history information of users narrowed down by the first condition;
An information processing device comprising: a learning unit that generates a learning model using history information of the user from which history information that satisfies the second condition has been excluded by the exclusion unit. The history information is
The information processing device according to claim 1, wherein the information is history information of a search query. The information processing apparatus according to claim 2, further comprising a narrowing-down unit that narrows down search queries that are highly common among a plurality of users having a specific attribute as history information of users who satisfy the first condition. The learning department is
Estimation that uses the user's history information from which history information that satisfies the second condition has been excluded by the exclusion unit to generate an attribute estimation model that receives information including the user's history information as input and outputs an attribute score. The information processing device according to claim 3, further comprising a model generation section. The exclusion part is
Excluding history information that satisfies the second condition, which differs for each attribute, from the user's history information narrowed down by the first condition, which differs for each attribute;
The estimated model generation unit includes:
generating the attribute estimation model that outputs an attribute score for each attribute using the history information of the user from which history information that satisfies the second condition has been excluded by the exclusion unit for each attribute;
The learning department is
It is characterized by comprising a prediction model generation unit that receives information including an attribute score for each attribute output from the attribute estimation model and generates a prediction model that outputs a prediction score regarding a specific behavior of the user. The information processing device according to claim 4. The prediction model is
The information processing device according to claim 5, wherein the information processing device is a loan loss prediction model that receives an attribute score output from the attribute estimation model as an input and outputs a score related to loan loss risk. The learning department is
An attribute score obtained by the attribute estimation model whose feature is information included in the history information after excluding the history information that satisfies the second condition, and history information before excluding the history information that satisfies the second condition. For each attribute, calculate the difference between the attribute score obtained by the attribute estimation model whose feature is information included in the attribute, and generate the prediction model without using the attribute score for which the difference is a predetermined value or more. The information processing device according to claim 5 or 6, characterized in that: The exclusion part is
According to claim 1, the history information that satisfies the second condition is excluded from the history information of the users narrowed down by the first condition using a rejection list that lists information to be excluded. The information processing device described. The exclusion part is
Claim 1 characterized in that history information that satisfies the second condition is excluded from the history information of users narrowed down by the first condition using a permission list that lists information that is not to be excluded. The information processing device described in . In the rejection list,
The information to be excluded is vectorized,
The exclusion part is
The information processing apparatus according to claim 8, wherein information whose vector is similar to the information to be excluded is excluded as history information that satisfies the second condition. In the permission list,
The information on the non-excluded items is vectorized,
The exclusion part is
The information processing apparatus according to claim 9, wherein information other than information whose vector is not similar to the non-exclusion target information is excluded as history information that satisfies the second condition. Information excluded by the exclusion unit as satisfying the second condition is:
The information processing device according to any one of claims 8 to 11, characterized in that the information is consideration-required personal information that is information that requires consideration for the user. Information excluded by the exclusion unit as satisfying the second condition is:
The information processing device according to claim 6, wherein the information processing device is consideration-required personal information that does not have a specific relationship with bad debt and requires consideration for the user. An information processing method performed by a computer, the method comprising:
an exclusion step of excluding history information that satisfies a second condition from the user's history information narrowed down by the first condition;
An information processing method comprising: a learning step of generating a learning model using history information of the user from which history information satisfying the second condition has been excluded in the exclusion step. an exclusion procedure for excluding history information that satisfies a second condition from the history information of users narrowed down by the first condition;
An information processing program that causes a computer to execute a learning procedure for generating a learning model using history information of the user from which history information satisfying the second condition has been excluded by the exclusion procedure.

Images