JP7716202B2 - Information processing device, information processing method, and program - Google Patents

Description

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

When conducting commercial or financial transactions with customers, it is necessary to evaluate the customer's creditworthiness. As a result, various methods for evaluating customer creditworthiness have been proposed.

For example, Patent Document 1 discloses a determination device that estimates asset information of a second user who has a similar behavioral history to a first user based on the user's asset information and the user's behavioral history on a network, and determines the trustworthiness of the second user.

JP 2018-110017 A

The invention described in Patent Document 1 extracts other users (first users) who are similar to the user to be evaluated (second user) from their behavioral history on a network, using search queries entered into a search engine and articles posted to a social networking service (SNS) as the behavioral history. However, in order to collect this behavioral history, user information (such as the device used by the user and the SNS account) must be registered in advance, which is not easy.

One aspect is to provide an information processing device or the like that can appropriately evaluate users.

An information processing device according to one aspect includes a memory unit that stores transaction information indicating a transaction history of a first user and behavioral history information indicating the behavioral history of the first user on a network, the behavioral history information including access information that can be obtained when the first user accesses a website; a generation unit that generates a trained model based on the transaction information and behavioral history information of the first user to determine an evaluation value of a user when the behavioral history information is input; a first acquisition unit that acquires identification information assigned to a second user through a session to the website; a second acquisition unit that acquires the behavioral history information including the access information of the second user based on the identification information ; and an evaluation unit that determines an evaluation value of the second user by inputting the behavioral history information of the second user into the model .

On the one hand, it allows users to be appropriately evaluated.

FIG. 1 is a schematic diagram illustrating an example of the configuration of a user evaluation system. FIG. 2 is a block diagram illustrating an example of the configuration of a server. 10 is an explanatory diagram showing an example of a record layout of a behavior history DB and a customer DB. FIG. 10A and 10B are explanatory diagrams showing an example of the record layout of a first user DB and a determination table. 10 is a timing chart showing the procedure of the behavior history analysis process. FIG. 10 is an explanatory diagram regarding the analysis process of behavior history information. FIG. 10 is an explanatory diagram regarding a keyword extraction process. 10 is a timing chart showing the procedure of a user evaluation process. FIG. 1 is an explanatory diagram illustrating an example of processing of a user evaluation system. 10 is a flowchart showing a processing procedure executed by a server. FIG. 1 is an explanatory diagram showing an application example of a user evaluation system. FIG. 11 is a block diagram illustrating an example of the configuration of a server according to Modification 2. 10 is a flowchart showing a procedure for generating a determination model.

The present invention will be described in detail below with reference to the drawings showing embodiments thereof.
(Embodiment)
1 is a schematic diagram showing an example of the configuration of a user evaluation system. In this embodiment, a user evaluation system is described that evaluates the creditworthiness of an arbitrary second user based on the financial transaction history of a large number of first users (other users) who are customers of a financial institution and the behavioral history of each first user on the Internet. The user evaluation system includes an information processing device 1, an analysis server 2, a user terminal 3, an affiliated server 4, and a financial institution server 5. Each device is connected to each other via a network N such as the Internet for communication.

Information processing device 1 is an information processing device capable of various information processing and information transmission and reception, such as a server computer or personal computer. In this embodiment, information processing device 1 is assumed to be a server computer, and for simplicity's sake will be referred to as server 1 below. Preferably, server 1 determines (evaluates) the creditworthiness of an arbitrary second user based on transaction information (e.g., financial product transaction history) indicating the transaction history of a first user who is a customer of a financial institution (e.g., a bank) and behavioral history information (e.g., words used on websites viewed by the first user) indicating the behavioral history of the first user on network N.

In this embodiment, the user who is the basis for the credit evaluation is referred to as the "first user," and the user who is the subject of the evaluation is referred to as the "second user." However, these names are used for convenience to distinguish between the two, and for example, the first user and the second user may be the same person. Furthermore, when there is no distinction between the two, they are simply referred to as "users." Furthermore, users are not limited to individuals, and may also be corporations.

Furthermore, in this embodiment, the creditworthiness of the second user is determined from the transaction history of financial products, etc., but this embodiment is not limited to this. For example, the server 1 may determine the creditworthiness from the purchase history (commercial transaction history) of products at a physical store or an e-commerce site, etc. In other words, the transaction information is not limited to financial transaction history.

Furthermore, while this embodiment primarily determines creditworthiness, this embodiment is not limited to this. For example, as described below, server 1 may determine the purchase probability of the second user purchasing a product. Server 1 only needs to be able to evaluate users based on transaction information and behavioral history information, and the scale of this evaluation is not particularly limited.

The analysis server 2 is a server computer that analyzes the behavioral history of users on the network N, and analyzes the web pages viewed by users on multiple affiliated websites (hereinafter referred to as "affiliated sites"). In this embodiment, the analysis server 2 uses a method known as a DMP (Data Management Platform) to collect data on viewed pages and analyze the behavioral history information. The analysis server 2 stores the analyzed behavioral history information in a behavioral history DB (Database) 201.

A DMP is a platform for integrating and managing data stored on various web servers and optimizing ad delivery. In this embodiment, the DMP is applied to credit assessment, and creditworthiness is determined using the analysis results from analysis server 2, i.e., user behavior history information.

The user terminal 3 is a terminal device operated by a user, such as a personal computer, smartphone, or tablet terminal. The analysis server 2 analyzes the web pages of affiliated sites accessed by the user terminal 3.

Affiliate server 4 is a server computer of the operator who runs the affiliated site, and functions as a web server. Examples of affiliated sites include EC (electronic commerce) sites, news article sites, and search sites, but the content is not particularly limited.

Financial institution server 5 is a server computer of a financial institution used by the first user as a customer, and provides server 1 with customer transaction information stored in customer DB 501. In this embodiment, financial institution server 5 is also affiliated with analysis server 2, and provides financial institution websites (e.g., electronic remittance sites, credit statement confirmation sites, etc.). For convenience, in the following explanation, this will be referred to as the "financial institution site" to distinguish it from other affiliated sites.

In this embodiment, analysis server 2 centrally manages identification information assigned to users at each affiliated site, collects data on web pages viewed by users linked to that identification information, and analyzes their behavioral history.

Generally, when a website accepts a session from a user, it assigns a session ID to each session and manages the data. This ID is stored on the user terminal 3 as a cookie data value, and the web server (affiliated server 4) identifies the terminal that originated the session based on this ID.

However, because each website assigns a separate ID, even if you try to link session data from multiple websites for a single user, it is not possible to identify which session data from one website corresponds to which session data from another website.

For example, affiliated sites AA, BB, CC, etc. that are affiliated with analysis server 2 have a program called a DMP tag embedded in them. When user terminal 3 accesses affiliated site AA, it executes the DMP tag to access analysis server 2 and is issued an ID that is common to affiliated sites AA, BB, CC, etc. When other affiliated sites BB, CC, etc. are accessed, user terminal 3 sends the ID to initiate a session, making it possible to link the session data at each affiliated site.

In the following explanation, the ID assigned by the analysis server 2 will be referred to as the "common ID."

Note that the above linking method is just one example, and the method is not limited to what is called DMP. The analysis server 2 only needs to be able to collect data based on identification information (such as the above-mentioned common ID) assigned to users on a certain affiliated site, and the specific method is not particularly limited. Furthermore, the data collection method is not limited to using cookies, and may also be a method that uses access information such as IP addresses and user agents.

DMP tags are also embedded in financial institution websites, and when the first user's user terminal 3 accesses the financial institution website, the common ID assigned to the first user is sent to the analysis server 2. This makes it possible to link the first user's transaction information held by the financial institution with the first user's behavioral history information analyzed by the analysis server 2.

Server 1 determines the creditworthiness of the second user based on the transaction information and behavioral history information of the first user. Server 1 then provides the determination result (evaluation result) to the affiliated server 4 of the business that conducts transactions with the second user. For example, when a business accepts a product purchase application from the second user via an affiliated site, such as an e-commerce site, Server 1 receives application information indicating the purchased product, price, etc., as well as the second user's common ID from the affiliated server 4, and accepts a request to output the creditworthiness. Server 1 determines the creditworthiness in response to the request and outputs the determination result.

Note that the following explanation will be primarily based on the assumption that a product is purchased on an e-commerce site, but the services (transactions) provided by the affiliated site are not limited to purchasing products and may include, for example, financial loans, insurance, real estate sales, etc. Furthermore, server 1 does not need to provide the credit assessment results to all affiliated site operators (affiliated servers 4), but may only provide them to some of those operators. Furthermore, the affiliated site that collects behavioral history information may be separate from the affiliated site of the operator that conducts transactions with the second user.

Furthermore, although the server 1 that performs the credit rating determination process, the analysis server 2 that analyzes the user's behavioral history, and the financial institution server 5 that stores the user's transaction information are described as being separate server computers, each device may also be the same server computer.

Furthermore, in the following explanation, the user's credit rating is provided to the operator of the affiliated site (affiliated server 4), but it can of course also be provided to a financial institution (financial institution server 5).

2 is a block diagram showing an example of the configuration of the server 1. The server 1 includes a control unit 11, a main memory unit 12, a communication unit 13, and an auxiliary memory unit .
The control unit 11 has one or more arithmetic processing devices such as a central processing unit (CPU), a micro-processing unit (MPU), a graphics processing unit (GPU), etc., and performs various information processing, control processing, etc. by reading and executing a program P stored in the auxiliary storage unit 14. The main storage unit 12 is a temporary storage area such as a static random access memory (SRAM), a dynamic random access memory (DRAM), or a flash memory, and temporarily stores data necessary for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing communication-related processing, and transmits and receives information to and from the outside.

The auxiliary memory unit 14 is a non-volatile storage area such as a large-capacity memory or hard disk, and stores the program P and other data required for the control unit 11 to execute processing. The auxiliary memory unit 14 also stores a first user DB 141 and a judgment table 142. The first user DB 141 is a database that stores transaction information and behavioral history information of the first user. The server 1 associates transaction information of the first user, who is a customer of the financial institution, with behavioral history information obtained by analyzing web pages of affiliated sites viewed by the first user, and stores these in the first user DB 141. The judgment table 142 is a table used to judge the creditworthiness of the second user, and stores keywords (reference strings) referenced during the judgment process.

The auxiliary storage unit 14 may be an external storage device connected to the server 1. The server 1 may also be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed using software.

In addition, in this embodiment, the server 1 is not limited to the above configuration and may include, for example, an input unit that accepts operational input, a display unit that displays images, etc. The server 1 may also be equipped with a reading unit that reads portable storage media 1a such as CD (Compact Disk)-ROM or DVD (Digital Versatile Disc)-ROM, and may read and execute the program P from the portable storage medium 1a. Alternatively, the server 1 may read the program P from semiconductor memory 1b.

FIG. 3 is an explanatory diagram showing an example of the record layout of the behavior history DB 201 and the customer DB 501. As shown in FIG.
The behavior history DB 201 includes a common ID string and a word string. The common ID string stores a common ID assigned to a first user. The word string stores words (character strings) described in web pages viewed by the first user in association with the common ID.

Customer DB501 includes a customer ID column, a name column, a common ID column, a deposit balance column, a card balance column, and a transaction column. The customer ID column is an ID used to identify a first user who is a customer of the financial institution. The name column, common ID column, deposit balance column, card balance column, and transaction column each store, in association with the customer ID, the first user's name, the common ID assigned to the first user, the deposit balance, the available balance on the credit card, and the transaction history of financial products.

FIG. 4 is an explanatory diagram showing an example of the record layout of the first user DB 141 and the determination table 142.
The first user DB141 includes a data number string, a deposit balance string, a card balance string, a transaction string, and a word string. The data number string stores data numbers for linking transaction information and behavioral history information. The deposit balance string, card balance string, transaction string, and word string store the first user's deposit balance, credit card balance, financial product transaction history, and viewed page words, respectively, in association with the data number. The data stored in the deposit balance string, card balance string, and transaction string corresponds to transaction information, and the data stored in the word string corresponds to behavioral history information.

The judgment table 142 includes a transaction type column and a keyword column. The transaction type column stores the type of financial transaction. The keyword column stores keywords that are referenced when judging creditworthiness, in association with the transaction type.

5 is a timing chart showing the procedure of the behavior history analysis process, which will be described with reference to FIG.
The user terminal 3 of the first user accesses the affiliated site in response to an operation input from the first user (step S101). In response to the access from the user terminal 3, the affiliated server 4 transmits the web page data (HTML file) of the affiliated site (step S102). The data transmitted in step S102 includes a DMP tag for assigning a common ID.

The user terminal 3 calls the DMP tag to redirect to the analysis server 2, and sends the web page data of the affiliated site accessed in step S101 to the analysis server 2 (step S103). For example, the user terminal 3 sends text data such as the title and body of the web page. If a common ID has not been assigned to the user terminal 3, the analysis server 2 assigns a common ID and stores the web page data in association with the common ID (step S104). The user terminal 3 saves the common ID assigned by the analysis server 2.

The user terminal 3 accesses the affiliated site in response to operation input from the first user (step S105). Note that the affiliated site accessed in step S105 may be the same as the affiliated site accessed in step S101, or it may be different. In response to the access from the user terminal 3, the affiliated server 4 transmits data of the affiliated site including the DMP tag (step S106). The user terminal 3 retrieves the DMP tag and transmits the saved common ID and the data of the web page accessed in step S105 to the analysis server 2 (step S107). The analysis server 2 stores the web page data in association with the common ID.

The analysis server 2 analyzes the first user's behavioral history from the web page data collected based on the common ID, for example, by batch processing (step S108). For example, the analysis server 2 extracts words contained in the web pages viewed by the user, such as the title of the viewed page and meta keywords (keywords specified as important words by tags in the HTML file).

Figure 6 is an explanatory diagram regarding the analysis process of behavioral history information. As described above, when the user terminal 3 accesses an affiliated site, a common ID is assigned by the analysis server 2 by calling the DMP tag. When the user terminal 3 accesses an affiliated site, it sends the common ID and data on the viewed page to the analysis server 2.

The analysis server 2 extracts words written within pages, such as titles and meta keywords, from the viewed pages collected based on the common ID. The analysis server 2 associates the extracted words with the common ID assigned to the user terminal 3 and stores (stores) them in a database.

In this embodiment, words within a viewed page are extracted as behavioral history information, but this embodiment is not limited to this. For example, the analysis server 2 may analyze session data between the user terminal 3 and the affiliated server 4, and extract the URL of the page of the affiliated site viewed by the first user, the referrer (the page from which the link originated when the first user accessed the affiliated site via a link from an external site), advertisements clicked by the first user, etc. Alternatively, if the affiliated site is a search site, the search query entered by the first user, etc. may be analyzed. In this way, behavioral history information is not limited to words within pages viewed by the first user.

Furthermore, while the above examples of behavioral history information include words, URLs, referrers, etc. on pages viewed by the user, this embodiment is not limited to this. For example, if the page viewed by the user is a product listing page on an e-commerce site, information about the listed products may be extracted as behavioral history information. In this way, behavioral history information may be any data that indicates the user's behavior on network N, and its content is not particularly limited.

Also, while the above example extracts character strings from the viewed page on a word-by-word basis, it is also possible to extract character strings on a phrase, sentence, etc. basis.

In addition, in the above example, the text data of the viewed page is sent directly from the user terminal 3 to the analysis server 2. However, for example, the analysis server 2 may record only the URL and session ID of the viewed page when the user terminal 3 accesses an affiliated site, and later access the affiliated server 4 to obtain the text data of the page viewed by the first user from the affiliated server 4.

Returning to Figure 5, we will continue the explanation. The user terminal 3 accesses the financial institution website in accordance with operational input from the first user (step S109). The financial institution server 5 responds to the access from the user terminal 3 by transmitting web page data for the financial institution website (step S110). The data transmitted in step S110 includes a DMP tag.

The user terminal 3 calls the DMP tag and sends the common ID to the financial institution server 5 (step S111). The financial institution server 5 stores the common ID in association with the customer ID. Through the above process, the analysis server 2 accumulates the first user's behavioral history information, and the financial institution server 5 associates the first user, who is a customer, with the common ID.

In the above, it was explained that the first user was already assigned a common ID when he accessed the financial institution's site, but if the first affiliated site he accessed was a financial institution's site, he would of course be assigned a common ID at the time he accessed the financial institution's site.

Based on the common ID acquired in step S111, the financial institution server 5 requests the analysis server 2 to output the behavioral history information of the first user corresponding to that common ID (step S112). The analysis server 2 transmits the behavioral history information of the first user to the financial institution server 5 (step S113). Specifically, as described above, the analysis server 2 transmits word data extracted from the first user's browsed pages to the financial institution server 5. The financial institution server 5 transmits the behavioral history information received from the analysis server 2 and the corresponding transaction information of the first user to the server 1 (step S114).

As described above, transaction information is data indicating the first user's transaction history at a financial institution, including the first user's deposit balance, available credit card balance, etc., as well as data indicating whether a contract or cancellation has been made for each type of transaction (e.g., type of financial product) (see Figure 3). The financial institution server 5 provides the transaction information and the corresponding behavioral history information of the first user to the server 1 as a series of data sets that serve as the basis for evaluating the second user. The server 1 associates the first user's transaction information with the behavioral history information linked to the first user's common ID and stores them in the first user DB 141.

When the financial institution server 5 sends transaction information, it sends anonymized data to prevent individuals from being identified from the transaction history. For example, the financial institution server 5 does not send data that could identify an individual, such as a common ID or the name of the first user, and only sends the transaction details. Also, to prevent individuals from being identified from the transaction history, only a portion of the various data included in the transaction information is sent to the server 1. Alternatively, when sending numerical values such as a deposit balance, the financial institution server 5 sends rounded values rather than raw values. This prevents the first user from being identified from the transaction information.

In this embodiment, anonymized data is sent to server 1 so that individuals cannot be identified, but it goes without saying that raw data including personal information may also be provided to server 1.

Furthermore, in this embodiment, the financial institution server 5 transmits the first user's transaction information and behavior history information together to server 1, but this embodiment is not limited to this. For example, the financial institution server 5 may provide the first user's transaction information to the analysis server 2, and the analysis server 2 may transmit the transaction information and behavior history information to server 1. Alternatively, if server 1 is able to obtain the first user's common ID, server 1 may request the analysis server 2 and financial institution server 5 to output behavior history information and transaction information based on the common ID, and obtain the data from each. In this way, there is no particular restriction on the route by which server 1 obtains transaction information and behavior history information.

The server 1 extracts keywords (reference strings) to be referenced when evaluating the second user from the transaction information and behavior history information of the first user (step S115). The server 1 stores the extracted keywords in the judgment table 142.

7 is an explanatory diagram of the keyword extraction process, and the processing content when extracting keywords in step S115 will be described with reference to FIG.
As described above, the behavior history information and transaction information of the first user are linked based on the common ID assigned to the first user. The server 1 extracts keywords from the behavior history information to be used when determining the creditworthiness of the second user, based on the transaction information linked to the behavior history information based on the common ID.

For example, as shown in the lower left of Figure 7, server 1 extracts words from the page viewed by the first user as keywords and stores them in judgment table 142. Specifically, server 1 extracts keywords for each type of transaction (e.g., type of financial product) depending on whether or not the transaction has occurred, and stores them in judgment table 142. For example, if a first user has a "card loan" and "overdue," server 1 associates the words extracted from the page viewed by the first user with "card loan" and "overdue" and stores them in judgment table 142.

Server 1 extracts keywords for each type of transaction as described above. When determining the trustworthiness of the second user, server 1 compares the words in the pages viewed by the second user with the keywords to determine the user's trustworthiness when conducting each type of transaction.

8 is a timing chart showing the procedure of the user evaluation process, which will be described with reference to FIG. 8. The process for determining the trustworthiness of the second user will be described with reference to FIG.
The user terminal 3 of the second user accesses an affiliated site (e.g., an e-commerce site) where commercial transactions, financial transactions, etc. are conducted in accordance with operation input from the second user (step S201). The affiliated server 4 responds to the access from the user terminal 3 and transmits web page data of the affiliated site (step S202). The data transmitted in step S202 includes a DMP tag in addition to information related to the transaction (e.g., product information). The user terminal 3 calls the DMP tag and transmits the common ID to the analysis server 2 (step S203).

In response to operational input from the second user, the user terminal 3 transmits application information from the second user to the affiliated server 4, requesting a transaction with the affiliated site operator (step S204). If the affiliated site is an e-commerce site, the user terminal 3 transmits an application to purchase a product. In step S204, the user terminal 3 transmits a common ID in addition to the application information to the affiliated server 4.

When application information is acquired from the user terminal 3, the affiliated server 4 transmits the application information and common ID to the server 1 and requests output of the credit rating (step S205). When the request to output the credit rating is accepted, the server 1 transmits the common ID acquired from the affiliated server 4 to the analysis server 2 and requests output of the behavioral history information of the second user corresponding to the common ID (step S206). Upon receiving the request from the server 1, the analysis server 2 outputs the behavioral history information of the second user (step S207).

Server 1 determines (evaluates) the creditworthiness of the second user based on the behavioral history information of the second user acquired from analysis server 2 and the transaction information and behavioral history information of the first user, and outputs the determination result (evaluation result) to affiliated server 4 (step S208). Affiliated server 4 determines whether or not to permit the transaction based on the creditworthiness output from server 1, and notifies the user terminal 3 of the determination result (step S209).

For example, server 1 compares the words in the page viewed by the second user with keywords for each type of transaction stored in determination table 142 to determine the creditworthiness for each type of transaction. For example, in the case of a "home loan," server 1 compares the words in the viewed page with keywords related to "home loan," such as "contract," "cancellation," and "bankruptcy," and determines the creditworthiness of the second user when conducting a transaction related to a home loan.

There is no particular restriction on the specific method for determining (calculating) creditworthiness, but for example, positive and negative points may be assigned to each keyword, and creditworthiness may be calculated by accumulating these points. Server 1 calculates the overall creditworthiness of the second user by adding up the creditworthiness determined for each type of transaction.

Note that the above determination method is just one example, and the credibility may also be determined by referring to the URL, referrer, etc. of the page viewed by the second user in addition to the words in the viewed page.

Furthermore, for example, the server 1 may also weight the creditworthiness by referring to the deposit balance, available credit card balance, etc. of a first user whose viewed page is the same as or similar to that of a second user. The similarity of viewed pages may be determined, for example, according to the frequency of appearance of common words (keywords).

Furthermore, for example, the server 1 may refer to the behavioral history information for a predetermined period (e.g., every month) and determine the trustworthiness for each period. This makes it possible to output changes in the trustworthiness of the second user.

Furthermore, for example, server 1 may also output a breakdown of the factors that determine creditworthiness (for example, scores for each transaction type) based on the score of the behavioral history information that was used as the basis for the creditworthiness determination (calculation). This allows external businesses to understand the factors that determine creditworthiness.

In addition, while creditworthiness was determined on a rule-based basis in the above, as in Variation 2 described below, machine learning can be performed based on the behavioral history information and transaction information stored in the customer DB 501 and behavioral history DB 201 to construct a model (e.g., a neural network) that takes behavioral history information as input and outputs creditworthiness. This allows the system administrator to input the user's behavioral history information into the model and determine creditworthiness without having to set rules.

9 is an explanatory diagram showing an example of processing of the user evaluation system, which illustrates an assumed example of the processing explained in the timing chart of FIG.
As described above, the business operator associated with the affiliated server 4 conducts a transaction with the second user via the affiliated site. In the example of Fig. 9, it is assumed that the affiliated site is an e-commerce site, and the second user later pays the purchase price to the business operator using a payment method such as a credit card.

When application information is acquired from the user terminal 3, the affiliated server 4 transmits the application information indicating the details of the transaction requested by the second user and the second user's common ID to the server 1, and requests output of the credit rating. The application information includes, for example, information such as the product the second user wishes to purchase and the purchase price, as well as information on the second user's payment method (credit card, etc.).

When the application information is acquired, Server 1 transfers the common ID to Analysis Server 2 and acquires the second user's behavioral history information. Server 1 then references the judgment table 142 and determines the second user's credibility from the behavioral history information.

In this embodiment, the second user's creditworthiness is determined from the first user's financial transaction history. However, the server 1 may also determine the second user's creditworthiness by referring to the history of past commercial transactions between operators of each affiliated site and the first user. Specifically, when the server 1 acquires the first user's common ID and application information from each operator (affiliated server 4), it stores the acquired application information in the first user DB 141 as transaction information. The server 1 also requests output of the first user's behavioral history information based on the common ID, acquires the behavioral history information from the analysis server 2, and stores it in the first user DB 141 in association with the first user's transaction information.

When server 1 receives a request from a business operator to output the creditworthiness of a second user, it determines the creditworthiness by referencing not only the first user's financial transaction information obtained from financial institution server 5, but also the first user's commercial transaction information obtained from affiliated server 4. In the example of Figure 9, when a request to output the creditworthiness is received from "EC site XX," server 1 determines the creditworthiness by referencing the first user's transaction history on "EC site XX." Note that server 1 may also use the transaction history of an affiliated site other than the affiliated site where the second user applied for the transaction when determining the second user's creditworthiness. By referencing the transaction history of the first user who previously used the same affiliated site, it is possible to appropriately determine the second user's creditworthiness based on the characteristics of users of that affiliated site.

The method for determining creditworthiness, including transaction history on affiliated sites, is not particularly limited, but one possible method is to identify a first user with similar transaction trends based on the products and purchase amounts applied for by the second user, and if a first user similar to the second user has a transaction history of non-payment, determine that there is a high possibility of default (bad debt) and lower the creditworthiness. In this way, server 1 may determine creditworthiness by referring to the transaction history of the first user's past transactions with operators on affiliated sites.

Furthermore, for example, if server 1 acquires application information from the same second user from multiple affiliated sites (EC sites), it may lower the creditworthiness of the second user. Specifically, if the transactions (post-payment) indicated in the application information acquired from each affiliated site occur at the same time, server 1 lowers the creditworthiness of the second user. This allows the creditworthiness to be appropriately determined taking into account the current transaction status (application status) at multiple affiliated sites.

Server 1 outputs the creditworthiness assessment result to affiliated server 4. It is preferable that the creditworthiness is output as a numerical value (for example, a value between 0 and 1), but it may also be output as a ranked result, for example, rank A, rank B, or rank C. Affiliated server 4 determines whether to approve the application based on the creditworthiness output from server 1 and notifies the user terminal 3 of the assessment result.

Note that while the above describes determining the second user's creditworthiness in commercial and financial transactions, this embodiment is not limited to this. For example, the server 1 may estimate (evaluate) the purchase probability of the second user purchasing a product on an EC site based on the first user's behavioral history information and transaction information, and provide the estimation result to the EC site. In this case, for example, the server 1 determines the similarity between the first user and the second user based on their behavioral history information, and extracts first users who are similar to the second user. The server 1 estimates the second user's purchasing power and products of interest from the extracted transaction information of the first user, and outputs the estimation result to the affiliated server 4. In this way, the server 1 only needs to be able to evaluate the second user based on behavioral history information, transaction information, etc., and the evaluation measure is not limited to transactional creditworthiness.

10 is a flowchart showing the processing procedure executed by the server 1. The processing contents executed by the server 1 will be described with reference to FIG.
The control unit 11 of the server 1 obtains application information indicating the content of the transaction proposed by the second user to the business operator and a common ID (identification information) assigned to the second user through a session to the affiliated site from the business operator's affiliated server 4 (step S11).

The control unit 11 sends the common ID acquired in step S11 to the analysis server 2 and requests the output of behavioral history information indicating the second user's behavioral history on the network N (step S12). For example, the control unit 11 requests the output of words (character strings) written on pages of affiliated sites viewed by the second user, as well as URLs, referrers, click advertisements, etc. The control unit 11 acquires the behavioral history information from the analysis server 2 (step S13).

The control unit 11 determines the creditworthiness of the second user based on the behavioral history information of the second user acquired in step S13 and the behavioral history information and transaction information of the first user, who is a customer of the financial institution (step S14). Specifically, keywords (reference strings) are stored for each transaction type in the determination table 142, and the control unit 11 determines the creditworthiness for each transaction type by comparing the keywords with words in the pages viewed by the user. The control unit 11 also determines the creditworthiness by referencing the transaction history (transaction information) for each transaction type. The control unit 11 may also determine the creditworthiness based on the URL, referrer, click advertisement, etc. The control unit 11 may also determine the creditworthiness by referencing the commercial transaction history (transaction information) of the first user acquired from each business operator. The control unit 11 outputs the determination result to the affiliated server 4 (step S15) and ends the series of processes.

In the above, the transaction information and behavioral history information of each user are linked based on a common ID (DMP tag), but this embodiment is not limited to this, and the information may be linked based on other identification information that can serve as a user identifier. The other identification information is, for example, access information of the user terminal 3 that can be obtained when accessing a website, specifically an IP address, user agent, timestamp, etc.

For example, when analyzing each user's behavioral history information, analysis server 2 associates access information when the user accesses an affiliated site with the behavioral history information and stores it in behavioral history DB 201. When analysis server 2 receives a request to output the second user's behavioral history information from server 1 and/or financial institution server 5, it acquires this access information in addition to or instead of the common ID. Analysis server 2 compares the acquired access information with the access information of each user stored in behavioral history DB 201, infers the same or similar users, and outputs the behavioral history information of those users. In this way, it is sufficient to link behavioral history information and transaction information based on identification information related to the user that can be acquired when accessing a website (affiliated site), and the identification information used as the basis is not limited to the common ID.

As described above, according to this embodiment, the second user's behavioral history information can be obtained based on the identification information assigned in response to the session to the affiliated site (website), and the second user can be appropriately evaluated (preferably credit evaluated) by referring to the first user's behavioral history information and transaction information.

Furthermore, according to this embodiment, by obtaining behavioral history information from an analysis server 2 (management device) that manages the identification information (common ID) of second users on each affiliated site, it is possible to appropriately grasp the behavioral history of second users on the network N.

Furthermore, according to this embodiment, by acquiring or assigning identification information according to a session to a financial institution site (first website), it is possible to efficiently acquire the first user's behavioral history information from the analysis server 2.

Furthermore, according to this embodiment, by comparing the words (character strings) written on the pages of the affiliated site viewed by the second user with keywords (reference character strings), the second user can be more appropriately evaluated.

Furthermore, according to this embodiment, by referencing keywords and transaction history for each type of transaction, it is possible to more appropriately evaluate users by combining behavioral history information and transaction information.

Furthermore, according to this embodiment, the evaluation results (e.g., credit rating) of the second user can be provided to an external business.

Furthermore, according to this embodiment, by also referring to the details of transactions (application information) that the first user has previously conducted with the business, the second user can be more appropriately evaluated.

(Variation 1)
In the above embodiment, the case where the creditworthiness of the second user is determined (evaluated) has been mainly described. However, the evaluation measure of the second user does not have to be the creditworthiness.

Figure 11 is an explanatory diagram showing an example application of the user evaluation system. Figure 11 provides a schematic illustration of cases where the system is applied to a variety of uses. Specifically, it conceptually illustrates keywords (behavioral history information) that serve as criteria for evaluating a target second user, divided into target user (individual or corporation), target service ("finance," "non-finance," etc.), and use ("credit," "sales promotion," "fraud detection," "other"). In the above-described embodiment, we have described the case of determining creditworthiness (the "credit" column in Figure 11), but as shown in Figure 11, server 1 may also evaluate second users for uses such as "sales promotion," "fraud detection," and "other."

"Sales promotion" refers to cases where the sale of a target object, such as a product, is promoted to the second user. Note that "products" here are not limited to physical products, but can also include financial products, insurance products, etc. Also, "target objects" are not limited to products, but can also include the provision of services.

For example, server 1 stores the transaction history of various objects by first users (whether a transaction was made, etc.) as transaction information in first user DB 141. Server 1 then obtains keywords from web pages viewed by each first user from analysis server 2 and stores them in determination table 142 in association with the transaction history of the objects. When server 1 receives a request from affiliated server 4 to output behavioral history information of a second user, it obtains the behavioral history information (keywords) of the second user from analysis server 2 and compares it with the keywords stored in determination table 142 to identify first users whose keywords are the same as or similar to those of the second user. Server 1 then determines objects whose sales should be promoted to the second user based on the transaction history of the first user and outputs the determination result to affiliated server 4.

Referring to Figure 11, for example, in the case of promoting sales of "finance"-related objects (financial products) to "individual" users ("Understanding necessary products"), server 1 stores keywords such as "Introduction to investing" and "New home" in determination table 142. Server 1 compares the keywords of the web page viewed by the second user with these keywords and determines the objects (financial products) that should be promoted for sale to the second user.

"Fraud detection" refers to a case where the server 1 detects (determines) the possibility that a second user may commit fraud, such as making a false declaration, during a transaction. The server 1 may determine the possibility of fraud (hereinafter referred to as "fraud risk") from the user's behavioral history information on the network N.

For example, server 1 stores in first user DB 141, as transaction information for the first user, whether or not the first user engaged in fraudulent activity at the time of the transaction. Server 1 then obtains from analysis server 2 the behavioral history information of first users who have engaged in fraudulent activity in the past, and stores keywords for each type of fraudulent activity (false declaration, multiple contracts using a false user name from the same terminal, fraudulent use of loans, etc.) in judgment table 142. When server 1 receives a request from affiliated server 4 to output the behavioral history information of a second user, it obtains the behavioral history information of the second user from analysis server 2 and compares it with the keywords stored in judgment table 142 to determine the risk of fraud by the second user. Server 1 outputs the determination result to affiliated server 4 and issues a warning.

Examples of "other" applications include predicting the churn of second users (customers), making calls to second users (sales calls via phone, email, etc.), and predicting relocation. For example, when the second user is a customer of a financial institution, server 1 stores keywords related to first users who have churned from the financial institution (users who have closed their accounts, etc.) in determination table 142 and determines the likelihood of the second user churn. For example, when the second user is a customer of a financial institution and a call is made to the second user to introduce a financial product, server 1 extracts the time periods when each user accesses a web page (browser) from the access information and determines the time period when the second user should be called. For example, when the second user is a customer of a financial institution and a call is made to the second user to introduce a financial product, server 1 extracts the time periods when each user accesses a web page (browser) from the access information and determines the time period when the second user should be called. For example, when the second user is a customer of a financial institution and a call is made to the second user to introduce a financial product, server 1 determines the time period when the second user will change their residence (move) based on changes in the second user's (customer's) access information (IP address, user agent, etc.) at multiple points in time.

In this way, server 1 only needs to be able to evaluate the second user based on the second user's behavioral history information, and a wide range of evaluation applications are envisioned, including determining creditworthiness, sales promotion, and fraud detection. This allows the system to be applied to a variety of measures, including narrowing down targets for sales promotions, strengthening identity verification, implementing campaigns to prevent churn, and creating customer call lists, in addition to credit evaluation.

Note that the server 1 may combine these evaluation methods. For example, when a second user accesses an affiliated site, the server 1 first determines the risk of fraud by the second user. If it determines that there is no risk of fraud, the server 1 then determines the creditworthiness of the second user. If it determines that there is no problem with the creditworthiness (for example, the creditworthiness is above a predetermined threshold), the server 1 further determines the objects that should be promoted for sale. The server 1 outputs each determination result to the affiliated server 4. In this case, for example, the server 1 may store keywords for each determination in multiple determination tables 142, 142, 142... and use the corresponding determination table 142 when making each determination. In this way, each evaluation method may be combined. Furthermore, for example, the server 1 may allow the affiliated business, to which the creditworthiness is output, to select one of these evaluation methods.

This variant is the same as the above-mentioned embodiment except for the method of evaluating the second user, so detailed explanations of the flowchart, etc. will be omitted.

As described above, according to this variant example 1, second users can be evaluated in various ways.

(Variation 2)
In the above embodiment, the second user is evaluated based on rules. Alternatively, a machine learning model may be constructed using the behavioral history information and transaction information of the first user as training data, and the second user may be evaluated using the model.

Figure 12 is a block diagram showing an example configuration of a server 1 according to Modification 2. The auxiliary memory unit 14 of the server 1 according to this modification stores a judgment model 50 instead of a judgment table 142. The judgment model 50 is a machine learning model that has learned the behavioral history information and transaction information of a first user, and is a model that, when the behavioral history information of a second user is input, judges the second user's evaluation value regarding credit, etc. It is expected that the judgment model 50 will be used as a program module that constitutes part of artificial intelligence software.

For example, the determination model 50 is a neural network generated by deep learning using the behavioral history information and transaction information of the first user as training data. Note that the determination model 50 may be a model based on other learning algorithms, such as a decision tree, random forest, or SVM (Support Vector Machine), in addition to a neural network.

The evaluation value output from the judgment model 50 is, for example, the second user's creditworthiness (for example, the probability of arrears in a deferred payment settlement). Note that the evaluation value is not limited to creditworthiness, but may also be the probability of purchasing a promotional item, the probability of fraud risk, the probability of defection, etc., as described in Variant 1. Furthermore, the evaluation value may be a continuous regression estimate (for example, a value between 0 and 1), or a classification value (for example, a rank that divides creditworthiness into stages).

The server 1 generates a judgment model 50 by learning the behavioral history information of the first user collected based on the common ID and the transaction information of the first user. Specifically, the server 1 assigns a correct value of creditworthiness to each first user according to the behavioral history information of each first user. For example, the server 1 refers to the first user DB 141 (see Figure 4) and adjusts the creditworthiness according to the presence or absence of a contract, late payment, cancellation, etc. with each first user, and assigns a correct value. The server 1 creates training data by labeling the behavioral history information of each first user with the correct value of creditworthiness assigned to each first user. The server 1 generates a judgment model 50 using the created training data.

Note that the correct value of the credibility (evaluation value) may not be assigned automatically by server 1, but may be assigned manually.

Server 1 inputs behavioral history information for the teacher into the judgment model 50 and obtains the credibility as an output from the judgment model 50. Server 1 compares the credibility output from the judgment model 50 with the correct answer value and updates parameters such as the weights between neurons so that the two are close to each other. Server 1 learns the teacher data for each first user and ultimately generates a judgment model 50 with optimized parameters.

In this embodiment, the behavioral history information input to the determination model 50 includes not only keywords extracted from affiliated sites viewed by the first user, but also access information that can be obtained when the affiliated site is accessed. The access information is log data when the user terminal 3 accesses an affiliated site, such as an IP address and user agent. In addition to keywords, the server 1 provides the access information to the determination model 50 as training data, allowing it to learn.

In particular, the access information to be learned in this embodiment includes the domain of the web page viewed by the user, the language setting on the user terminal 3, the user agent, and the IP address. According to the developments of the present inventors, including this access information in the features can improve the accuracy of determining credibility.

When determining the trustworthiness of a second user, the server 1 determines the trustworthiness of the second user by inputting the behavioral history information (including access information) of the second user obtained from the affiliated server 4 into the determination model 50. The server 1 outputs the determined trustworthiness to the affiliated server 4.

13 is a flowchart showing the procedure of the process of generating the determination model 50. The process of generating the determination model 50 by machine learning will be described with reference to FIG.
The control unit 11 of the server 1 acquires behavior history information and transaction information of each first user from the first user DB 141 (step S21). The behavior history information includes access information as well as keywords extracted from affiliated sites (websites) accessed by the first user. The access information is log data when the user terminal 3 accesses an affiliated site, and includes the domain of the web page viewed (accessed) by the user, the language setting on the user terminal 3, the IP address, and the user agent.

The control unit 11 assigns a correct value for the first user's credit rating based on the first user's transaction information (step S22). Note that the correct value may be assigned automatically by the server 1 or manually.

The control unit 11 generates a judgment model 50 that judges the trustworthiness when behavioral history information is input, based on the behavioral history information of the first user and the correct value of the trustworthiness assigned in step S202 (step S23). For example, the control unit 11 generates a neural network as the judgment model 50. The control unit 11 inputs the behavioral history information of the first user into the judgment model 50, thereby obtaining the trustworthiness of the first user from the judgment model 50. The control unit 11 optimizes parameters such as the weights between neurons so that the obtained trustworthiness approximates the correct value. In this way, the control unit 11 generates the judgment model 50. The control unit 11 ends the series of processes.

This variant is the same as the above-described embodiment except that the trustworthiness of the second user is determined using the determination model 50, so detailed explanations of the flowchart for determining trustworthiness (see Figure 10) and other steps will be omitted.

In addition, access information such as the domain of the viewed page, language setting, IP address, and user agent may also be used when evaluating the second user (determining their credibility) on a rule-based basis, as in the above-described embodiment. For example, when storing keywords of a web page viewed by the first user in the determination table 142, the server 1 simultaneously stores the domain of the viewed page of the first user, language setting, IP address, and user agent. The server 1 then acquires each address information along with the keywords as the second user's behavior history information, and compares it with the keywords and address information stored in the determination table 142 to determine their credibility. In this way, the server 1 only needs to be able to evaluate the second user based on behavior history information including each access information, and the evaluation method based on access information is not limited to evaluation using a machine learning model.

As described above, according to this variant example 2, it is also possible to evaluate second users using a machine learning model.

The embodiments disclosed herein are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the claims, not by the meaning described above, and is intended to include all modifications within the meaning and scope of the claims.

1. Server (information processing device)
11 Control unit 12 Main memory unit 13 Communication unit 14 Auxiliary memory unit P Program 2 Analysis server 3 User terminal 4 Affiliated server 5 Financial institution server

Claims (13)

a storage unit that stores transaction information indicating a transaction history of a first user and behavior history information indicating a behavior history of the first user on a network, the behavior history information including access information that can be obtained when the first user accesses a website;
a generation unit that generates a trained model based on transaction information and behavior history information of the first user to determine a user's evaluation value when the behavior history information is input;
a first acquisition unit that acquires identification information assigned to the second user through a session to the website;
a second acquisition unit that acquires the behavior history information including the access information of the second user based on the identification information;
an evaluation unit that determines an evaluation value of the second user by inputting the behavior history information of the second user into the model . The information processing apparatus according to claim 1 , wherein the access information includes a browsing domain, a language setting, a user agent, or an IP address of the user. The information processing device according to claim 1 , wherein the evaluation unit determines the creditworthiness of the second user. The transaction information includes a transaction history of a predetermined object by the first user;
4. The information processing device according to claim 1, wherein the evaluation unit determines the object to be promoted for sale to the second user. The transaction information includes whether or not the first user engaged in fraudulent activity at the time of the transaction,
The information processing device according to claim 1 , wherein the evaluation unit determines the possibility of fraudulent conduct by the second user during a transaction. a request unit that requests a management device that manages the identification information commonly assigned to the plurality of websites to output the behavior history information based on the identification information acquired by the first acquisition unit;
The information processing device according to any one of claims 1 to 5 , wherein the second acquisition unit acquires the behavior history information indicating a behavior history of the second user on the plurality of websites. the transaction information includes information indicating a financial transaction history of the first user who is a customer of a financial institution;
a third acquisition unit that acquires, in response to a session from the first user to the website of the financial institution, the transaction information of the first user and the behavior history information of the first user on the plurality of websites linked to the identification information of the first user;
The information processing device according to claim 6 , wherein the storage unit stores the transaction information and the behavior history information of the first user in association with each other. The behavior history information includes data indicating character strings written on pages of the website.
8. The information processing device according to claim 1, wherein: the first acquisition unit acquires the identification information of the second user from a business that conducts a transaction with the second user via the website;
The information processing device according to any one of claims 1 to 8 , further comprising an output unit that outputs the evaluation result of the second user to the business operator. generating a trained model to determine a user's evaluation value when the behavioral history information is input based on transaction information indicating a transaction history of a first user stored in a storage unit and behavioral history information indicating a behavioral history of the first user on a network, the behavioral history information including access information obtainable when the first user accesses a website;
acquiring identification information assigned to the second user through a session to the website;
acquire the behavior history information including the access information of the second user based on the identification information;
An information processing method, characterized in that a computer executes a process of determining an evaluation value of the second user by inputting the behavior history information of the second user into the model. generating a trained model to determine a user's evaluation value when the behavioral history information is input based on transaction information indicating a transaction history of a first user stored in a storage unit and behavioral history information indicating a behavioral history of the first user on a network, the behavioral history information including access information obtainable when the first user accesses a website;
acquiring identification information assigned to the second user through a session to the website;
acquire the behavior history information including the access information of the second user based on the identification information;
A program causing a computer to execute a process of determining an evaluation value of the second user by inputting the behavior history information of the second user into the model . a storage unit that stores transaction information indicating a transaction history of a first user and behavior history information indicating a behavior history of the first user on a network, the behavior history information including access information that can be obtained when the first user accesses a website;
a generation unit that generates a trained model based on transaction information and behavior history information of the first user to determine a user's evaluation value when the behavior history information is input;
a first acquisition unit that acquires identification information related to a second user who accesses the website;
a second acquisition unit that acquires the behavior history information including the access information of the second user based on the identification information;
an evaluation unit that determines an evaluation value of the second user by inputting the behavior history information of the second user into the model . a storage unit that stores transaction information indicating a transaction history of a first user and behavior history information indicating a behavior history of the first user on a network, the behavior history information including data indicating character strings described on pages of a website viewed by the first user;
an extraction unit that extracts a reference character string to be referred to when evaluating a second user from the behavior history information of the first user;
a table generating unit that stores the extracted reference character string in a table in association with a type of transaction previously performed by the first user based on the transaction information;
a first acquisition unit that acquires identification information assigned to the second user through a session to a website;
a second acquisition unit that acquires the behavior history information of the second user based on the identification information, the behavior history information including data indicating a character string described on a page of a website visited by the second user;
an evaluation unit that evaluates the second user for each type of transaction by comparing the character string indicated by the behavioral history information of the second user with the reference character string stored in the table for each type of transaction.