TW202405723A - Information processing system, method and program - Google Patents

Abstract

To realize evaluation such as calculating a user score or to improve evaluation accuracy even when information of a target user is missing or information reliability is low. An information processing system includes: a reference user identification unit 22 that identifies a reference user who is mutually related to a target user; an attribute generation unit 26 that generates corresponding attribute data of the target user based on attribute data of the reference user identified for the target user; an attribute complementing unit 27 that complements a corresponding attribute data group of the target user based on at least a part of the generated corresponding attribute data of the target user; and a user score estimation unit 28 that estimates a user score to be set for the target user based on the complemented corresponding attribute data group of the target user.

Description

Information processing systems, information processing methods and program products

This disclosure relates to technologies required to support evaluation such as calculation of a user's score.

Previously, a judgment device was provided with: a user information acquisition unit that acquires behavior information indicating the user's behavior; and a credit rating determination unit that determines a future user's repayment ability for financing based on the behavior information. Credibility has been proposed (see Patent Document 1). Furthermore, a system has been proposed that determines whether or not user scores can be displayed in accordance with the degree of intimacy between users (for example, refer to Patent Document 2). [Prior technical literature] [Patent Document]

[Patent Document 1] Japanese Patent Application Publication No. 2021-174039 [Patent Document 2] Japanese Patent Application Publication No. 2020-129228

[Problem to be solved by the invention]

Previously, technology has been proposed that calculates a user score indicating the user's creditworthiness based on the user's action history. However, if all the information about the target user is missing or the reliability of the information is low, there is a problem that the user score cannot be calculated or the accuracy of the calculated user score is insufficient.

The present disclosure is based on the above problem, and its subject is to realize evaluation such as calculation of user scores or to improve evaluation accuracy even when all the information about the target user is defective or the reliability of the information is low. [Means used to solve problems]

An example of this disclosure is an information processing system that includes: a reference user specifying means for specifying a reference user that is related to a target user; and an attribute generation means for specifying a reference user based on the foregoing information. The object user is specified by referring to the attribute data of the user, and the corresponding attribute data of the object user is generated; and the attribute completion means is used based on the corresponding attributes of the user of the object that has been generated. At least part of the data is used to complete the corresponding attribute data group of the previous target user; and the user score estimation means is used to complete the corresponding previous attribute data group of the previous target user based on the completion. The user score that is presumed to be set to this target user.

This disclosure may be defined as a method executed by an information processing device, a system, a computer, or a program executed by a computer. In addition, the present disclosure can also be defined as recording such a program on a recording medium that can be read by a computer or other device or machine. Here, the so-called recording medium that can be read by a computer or the like refers to a storage medium in which information such as data or programs is stored electrically, magnetically, optically, mechanically or chemically and can be read from a computer or the like. recording media. [Effects of the invention]

According to this disclosure, even if all the information about the target user is defective or the reliability of the information is low, evaluation such as calculation of user scores can still be realized, or evaluation accuracy can be improved.

Below, the implementation forms of the information processing system, information processing method and program product involved in the present disclosure are explained based on the diagrams. However, the embodiments described below are only exemplary embodiments, and the information processing system, information processing method, and program product involved in the present disclosure are not limited to the specific configurations described below. During implementation, specific configurations may be appropriately adopted depending on the implementation aspect, and various improvements or modifications may be made.

This embodiment describes a case where the technology according to the present disclosure is implemented in a user score management system that manages user scores indicating certain criteria (such as credit, etc.) related to the user. Implementation form. However, the technology involved in the present disclosure can be widely used as a technology required for estimating a user's score, and the applicable objects of the present disclosure are not limited to the examples shown in the embodiments.

＜System Structure＞ FIG. 1 is a schematic diagram showing the structure of the information processing system according to this embodiment. In the system described in this embodiment, the information processing device 1 and 1 or a plurality of service providing systems 5 are connected so as to communicate with each other. The user is a user of the service provided by the service providing system 5, and receives the provision of the service by accessing the service providing system 5 from the user terminal.

The information processing device 1 is equipped with: CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, EEPROM (Electrically Erasable and Programmable Read Only Memory) or HDD (Hard Disk Drive) A computer such as a memory device 14, a communication unit 15 such as a NIC (Network Interface Card), etc. However, the specific hardware configuration of the information processing device 1 can be appropriately omitted, replaced, or added depending on the implementation aspect. In addition, the information processing device 1 is not limited to a device composed of a single housing. The information processing device 1 can be implemented by a plurality of devices using so-called cloud or distributed computing technology.

The information processing device 1 manages user scores for each user, and provides the user scores to the service providing system 5 . The service providing system 5 can customize the service for the target user in accordance with the user score provided from the information processing device 1 .

The service providing system 5 is a computer equipped with a CPU, ROM, RAM, memory device, communication unit, input device, output device, etc. (illustration omitted). Furthermore, these systems and terminals are not limited to devices composed of a single frame. These systems and terminals can be implemented by multiple devices using so-called cloud or distributed computing technologies.

In the system described in this embodiment, the service providing system 5 includes an e-commerce transaction system 40, a golf course reservation system 42, a travel reservation system 44, and a card management system 46, which are connected to communicate with each other. However, the services provided by the service providing system 5 are not limited to the examples in this embodiment. The services provided by the service providing system 5 may be, for example: map information services or credit card/postpaid billing services, electronic money billing services, online shopping services, online reservation services, customer service center services, etc. In addition, "postpaid checkout" is not limited to services called so-called Buy Now, Pay Later (BNPL), etc., and can include the purchase of any postpaid goods/services.

The service providing system 5 notifies the information processing device 1 of the user's attribute data group obtained from the user when the service is provided. In addition, the information processing device 1 can access the service providing system 5, acquire the user attribute data registered in the system for a plurality of users including the target user, and include it in the attribute data. in the group. Here, the user's attribute data includes information about the user who uses the system, that is, account data, and usage history data of services provided by the user. The content of service usage history data will vary depending on the content of the service, such as history data of the user's location information, payment history data of credit card usage/postpaid bill usage, and electronic money usage. History data, transaction history data, reservation history data, customer service history data for the user from the customer service center, and frequently visited places of stay specified based on history data based on location information, etc. In addition, the account information includes, for example: user ID, name information, address information, age information, gender information, phone number information, mobile phone number information, credit card number information, IP address information, school location information, and work location information. wait.

User ID is, for example, the identification information of the user in the computer system. Name information is, for example, information indicating the name (surname and first name) of the user. Address information is, for example, information indicating the user's residential address. When the computer system is an e-commerce transaction system 40, the address information may also be the address used to indicate the shipping location of the goods purchased by the user. Age information is, for example, information indicating the age of the user. Gender information is, for example, information indicating the gender of the user. Telephone number information is, for example, information indicating the telephone number of the user. Mobile phone number information is, for example, data indicating the mobile phone number of the user. Credit card number information is, for example, information indicating the card number of the credit card used by the user when checking out in the computer system. IP address information is, for example, information indicating the IP address of the computer used by the user (such as the IP address of the source of the message). The school place information is, for example, when the user is a student, it is data indicating the school place (name or address of an educational institution, etc.) of the user. The workplace information is, for example, information indicating the workplace (company name or address, etc.) of the user when the user is a member of the public.

FIG. 2 is a schematic diagram showing the functional configuration of the information processing device 1 according to this embodiment. The information processing device 1 reads the program recorded in the memory device 14 into the RAM 13 and executes it by the CPU 11 to control each hardware included in the information processing device 1, thereby It includes a graphic data generating unit 21, a reference user specifying unit 22, a relationship specifying unit 23, a relationship strength determining unit 24, an attribute selecting unit 25, an attribute generating unit 26, an attribute complementing unit 27, a user score estimating unit 28, and the information processing device of the machine learning unit 29 to function. In addition, in this embodiment and other embodiments described below, each function of the information processing device 1 is executed by the general-purpose processor, that is, the CPU 11, but some or all of these functions may also be executed by the CPU 11. or multiple special-purpose processors.

The graphic data generating unit 21 specifies a pair of users who are related to each other based on the attribute data group of each of the plurality of users, thereby generating graphic data (social graph network) indicating the relationship between the users. road). More specifically, the graphic data generating unit 21 generates, for example, node data 50 corresponding to a plurality of users including the target user, and link data 52 corresponding to pairs of users having a relationship with each other. Graphical data (refer to Figure 4, Figure 6, Figure 8, and Figure 9). In addition, the graph generation unit 21 performs learning (expression learning, relationship learning, embedding learning, knowledge graph embedding) of a relationship graph between users composed of nodes (users) connected by explicit connections, so as to Predict and create suggestive connections between users. At this time, the graph generation unit 21 can appropriately perform the necessary learning based on a known embedding model or its extension.

For example, as shown in FIG. 3 , assume that the attribute data group of user A is registered in the e-commerce transaction system 40 . Furthermore, it is assumed that the attribute data group of user B is registered in the golf course reservation system 42. Furthermore, it is assumed that the attribute data of user C is registered in the travel reservation system 44. Then, assume that the value of the IP address data of user A logged in the e-commerce transaction system 40, the value of the IP address data of user B logged in the golf course reservation system 42, and the travel reservation system 44 The value of the IP address data of user C logged in is the same.

In this case, the graphic data generating unit 21 generates, as shown in FIG. 4 , node data 50a associated with the user A, node data 50b associated with the user B, and node data associated with the user C. Data 50c, link data 52a indicating that user A has a relationship with user B, link data 52b indicating that user A has a relationship with user C, and link data 52b indicating that user B has a relationship with user C. Graphical data linking data 52c. The IP address belongs to the same user, presumably using the same computer or sharing a global address in the same residence or workplace. Therefore, in this embodiment, such users are related to each other.

For example, as shown in FIG. 5 , it is assumed that the attribute data groups of user D, user E, and user F are registered in the e-commerce transaction system 40 . Then, it is assumed that the values of the address data of user D, the address data of user E, and the address data of user F registered in the e-commerce transaction system 40 are the same.

In this case, the graphic data generating unit 21 generates, as shown in FIG. 6 , node data 50d associated with the user D, node data 50e associated with the user E, and nodes associated with the user F. Data 50f, link data 52d indicating that user D is related to user E, link data 52e indicating that user D is related to user F, and link data 52e indicating that user E is related to user F. Link data 52f graphic data. Users with the same address are presumed to be living together. Therefore, in this embodiment, such users are related to each other.

For example, as shown in FIG. 7 , it is assumed that the attribute data group of user G is registered in the e-commerce transaction system 40 . Furthermore, it is assumed that the attribute data group of user H is registered in the golf course reservation system 42. Furthermore, it is assumed that the attribute data group of user I is registered in the travel reservation system 44. Then, assume that the value of the credit card number data of user G registered in the e-commerce transaction system 40, the value of the credit card number data of user H registered in the golf course reservation system 42, and the value of the credit card number data of the user H registered in the travel reservation system 44. The values of the credit card number data of the logged-in user I are the same.

In this case, the graphic data generation unit 21 generates, as shown in FIG. 8 , node data 50g associated with the user G, node data 50h associated with the user H, and node data associated with the user I. Data 50i, link data 52g indicating that user G is related to user H, link data 52h indicating that user G is related to user I, and link data 52h indicating that user H is related to user I. Graphical data for link data 52i. The credit card numbers belong to the same user, and they are presumed to be family members such as parent and child. Therefore, in this embodiment, such users are related to each other.

In addition, the criteria for judging whether the matching of users who are related to each other is met is not limited to the above explanation. User matching can be judged based on various criteria such as location information history or action history.

In addition, as described above, a link represented by the link data 52 that is related to users who are specified to have a relationship with each other is called an explicit link. It is assumed here that, for example, there are more than a certain number of users (for example, 3 or more people) who are explicitly linked to the first user and users who are explicitly linked to the second user. Common. In this case, for example, in the present embodiment, the graphic data generating unit 21 generates link data 52 indicating that the first user is related to the second user. The link represented by the link data 52 generated in this way is called a suggestive link.

For example, as shown in FIG. 9 , it is assumed that node data 50j associated with user J and node data 50k associated with user K are connected by link data 52j indicating an explicit link. Furthermore, it is assumed that the node data 50j associated with the user J and the node data 50l associated with the user L are connected by the link data 52k indicating an explicit link. Furthermore, it is assumed that the node data 50j associated with the user J and the node data 50m associated with the user M are connected by the link data 52l indicating an explicit link.

Furthermore, it is assumed that the node data 50k associated with the user K and the node data 50n associated with the user N are connected by the link data 52m indicating an explicit link. Furthermore, it is assumed that the node data 50l associated with the user L and the node data 50n associated with the user N are connected by the link data 52n indicating an explicit link. Furthermore, it is assumed that the node data 50m associated with the user M and the node data 50n associated with the user N are connected by the link data 52o indicating an explicit link.

In this case, the graphic data generation unit 21 generates link data 52p indicating that user J is related to user N (link data 52p indicating a suggestive link). In this way, user N will be specified as a user who has a relationship with user J.

For example, it is also assumed that there are more than a certain number of users who are explicitly linked or implicitly linked to the first user, and users who are explicitly or implicitly linked to the second user. (e.g. 3 or more people) are common. In this case, the graphic data generating unit 21 may generate link data 52 indicating that the first user is related to the second user (link data 52 indicating a suggestive link).

The reference user specifying unit 22 refers to the graphic data generated by the graphic data generating unit 21 and specifies other users who are related to the target user among the users included in the graphic data. The reference user of the object user. Here, referring to the user specifying section 22, the user specified as the user having a relationship with the target user, and the user specified as the user having the relationship may be a predetermined number of users. The above target users are common users and specifically become reference users. Furthermore, the reference user specifying unit 22 may specify a reference user from among the plurality of users based on the attributes of the target user and the attributes of the plurality of users.

The reference user specifying part 22 may also associate, for example, node data 50 associated with the target user and use associated with node data 50 connected by link data 52 indicating an explicit link or an implicit link. , specifically becomes the reference user for the target user.

The relationship specifying part 23 specifies the relationship between users. The relationships between users specified here are, for example: (1) parent-child relationship or husband-wife relationship living in the same household, (2) friend relationship, (3) relationship working in the same workplace, etc. However, the specified relationship is not limited to the examples in this disclosure. In this embodiment, the relationship specifying unit 23 specifies the relationship between users based on the result of clustering based on the values corresponding to the relationship between the users. Here, the types of values that can be used as values to establish relationships between users are not limited, but may include, for example: the user's name, IP address, address, credit card number, age, gender, At least one of the place of study, place of work and place of stay.

The relationship specifying unit 23 specifies the relationship between the target user and the reference user. Here, the relationship specifying unit 23 may specify the relationship between the target user and the reference user based on the attribute data group of the target user and the attribute data group of the reference user. In addition, the computer system in which the attribute data group of the target user is registered may be different from the computer system in which the attribute data group of the reference user is registered. For example, the target user and the reference user may also be combined based on the attribute data group of the target user registered in the e-commerce transaction system 40 and the attribute data group of the reference user registered in the golf course reservation system 42. The relationship between the two is specified.

The relationship specifying unit 23 specifies pairs of node data 50 connected by the link data 52 , for example. Then, the relationship specifying unit 23 generates pairing attribute data corresponding to the pairing based on the user attribute data groups of the two users associated with the pairing. Here, the matching attribute data includes, for example: a common IP flag, a common address flag, a common credit card number flag, a same last name flag, age difference data, matched gender data, a common school place flag, and a common work place flag. flags, common flags for detention areas, etc.

The IP common flag is, for example, a flag indicating whether the value of the IP address data contained in the attribute data of one party in the pair is the same as the value of the IP address data contained in the attribute data of the other party. For example, when the values of the IP address data are the same, the value of the IP common flag is set to 1. When the values of the IP address data are different, the value of the IP common flag is set. The value is set to 0.

The common address flag, the common school place flag, the work place common flag and the stay place common flag are, for example, the address data/school place data/work place included in the attribute data group of one of the matched parties. Whether the value of the data/stay place data is the same flag as the value of the address data/school place data/work place data/stay place data included in the attribute data group of the other party. For example, when the values of the address data are the same, the value of the address common flag is set to 1; when the values of the address data are different, the value of the address common flag is set to 0.

The credit card number common flag is, for example, a flag indicating whether the value of the credit card number data contained in the attribute data group of one party in the pair is the same as the value of the credit card number data contained in the other party's attribute data group. . For example, if the values of the credit card number data are the same, set the value of the credit card number common flag to 1; if the values of the credit card number data are different, set the value of the credit card number common flag to 0.

The same last name flag is, for example, a flag indicating whether the last name represented by the name data contained in the attribute data group of one party in the pair is the same as the last name represented by the name data contained in the attribute data group of the other party. mark. For example, if the last names represented by the name data are the same, the value of the same last name flag is set to 1. If the last names represented by the name data are different, the value of the same last name flag is set to 0.

The age difference data is, for example, data that represents the difference between the value of the age data included in the attribute data group of one of the paired parties and the value of the age data included in the attribute data group of the other party.

The paired gender data is, for example, data that represents a combination of the value of the gender data included in the attribute data group of one of the paired parties and the value of the gender data included in the attribute data group of the other party.

Then, the relationship specifying unit 23 performs clustering using a general clustering technique based on the value of the pairing attribute data group corresponding to each of the plurality of pairs, so as to classify the plurality of pairs as shown in Figure 10 A plurality of clusters 54 are shown.

FIG. 10 is a schematic diagram showing an example of how a plurality of pairs are classified into five clusters 54 (54a, 54b, 54c, 54d, and 54e). The fork shown in Figure 10 corresponds to the pairing. Then, each of the plurality of crosses is arranged at a position corresponding to the value of the pairing attribute data of the pair corresponding to the cross. In the example of FIG. 10 , although the plurality of pairs are classified into five clusters 54 , the number of clusters 54 into which the plurality of pairs are classified is not limited to five. For example, the plurality of pairs can be classified into 4 clusters 54.

FIG. 11 is an illustration of an example of visualization that should be performed when a plurality of pairs are classified into four clusters 54 . As shown in Figure 11, pairs with the same address, the same gender, the age difference is greater than X years, and the same surname can also be classified into the first cluster. In addition, pairs with the same address, the same gender, the age difference is less than X years, and the same surname can also be classified into the second cluster. In addition, pairs with the same address, different genders, an age difference greater than Y years, and the same surname can also be classified into the third cluster. In addition, pairs with the same address, different genders, an age difference of Y years or less, and the same surname can also be classified into the fourth cluster.

In this case, the first cluster system can be inferred to be, for example, cluster 54 associated with parents and children of the same sex. Furthermore, the second cluster system can be speculated to be a cluster 54 corresponding to brothers and sisters of the same sex. In addition, the third cluster system can be speculated to be a cluster 54 corresponding to parents and children of the opposite sex. In addition, the fourth cluster system can be inferred to be cluster 54 corresponding to the couple.

As explained above, the relationship specifying unit 23 may specify the relationship between the target user and the reference user based on the result of clustering based on the values corresponding to the relationship between the users. A specific example of a clustering of friendships or relationships working in the same workplace by clustering based on a common flag of school place, a common flag of work place, and a common flag of residence is as described above. The explanatory examples are basically the same, so the explanation is omitted. In addition, the relationship specific part 23 may be based on the result of clustering based on at least one of the last name, IP address, address, credit card number, age difference, gender, school place, work place, and stay place, The relationship between the target user and the reference user is specified.

The relationship strength determination unit 24 determines the target user based on an index indicating the strength of the relationship between the target user and the reference user based on the judgment criteria corresponding to the relationship between the target user and the reference user. The strength of the relationship (hereinafter also referred to as the "proximity score") to the user that should be referenced. In this embodiment, the relationship strength determination unit 24 determines the representation object based on the output when the data representing the index is input to the learned machine learning model corresponding to the relationship between the target user and the reference user. The strength of the relationship between the user and the reference user (proximity score).

Here, the relationship strength determining unit 24 may also include learned machine learning models corresponding to the above-mentioned clusters 54 respectively. For example, when a plurality of pairs are classified into five clusters 54, the relationship strength determination unit 24 may also include five machine learning models. Then, the relationship strength determination unit 24 may also input the strength of the relationship between the target user and the reference user based on the learned machine learning model corresponding to the relationship between the target user and the reference user. The output of the indicator data is used to determine the proximity score that represents the distance between the target user and the reference user. In this case, the input-output relationship implemented in the learned machine learning model is equivalent to the above-mentioned judgment standard.

As shown in FIG. 12 , the relationship strength determination unit 24 may input the pairing information classified into the cluster 54 corresponding to the n-th machine learning model for the n-th machine learning model. Corresponding input data. For example, when the relationship strength determination unit 24 includes five machine learning models, the above-mentioned value n is any one of the integers from 1 to 5. Then, the relationship strength determination unit 24 may determine the value of the output data output from the n-th machine learning model in response to the input of the corresponding input data as the value of the proximity score for the corresponding pairing.

The input data corresponding to the pair may also include, for example, part or all of the pairing attribute data corresponding to the pair. In addition, the input data may also contain data that is not included in the matching attribute data. For example, the input data may include data indicating the usage history of the e-commerce transaction system 40 or data obtained from other information sources such as SNS by the relationship strength determination unit 24. More specifically, for example, the input data may also include information indicating the number of calls or messages exchanged between the pairings per unit period, the number of gifts given by one party to the other party, the number of common friends in the pairing, etc. information.

In addition, the type of data contained in the input data corresponding to the matching may also be the same or different depending on the cluster 54 to which the matching belongs. For example, the type of data included in the input data input to the first machine learning model may be different from the type of data included in the input data input to the second machine learning model.

In this embodiment, for example, before the determination of the proximity score by the relationship strength determination unit 24, the n-th machine learning model is executed in advance using a given plurality of training data associated with the n-th machine learning model. of learning. The training data is prepared in advance, for example, in such a way that the determination of the proximity scores in the cluster 54 corresponding to the nth machine learning model will be appropriate. Here, the proximity score set to the training data may also be a proximity score set based on rules (annotated). Alternatively, it may be a proximity score that has been corrected by a manager or the like after being outputted in the past using a machine learning model.

Here, learning by weakly supervised learning can also be performed on the nth machine learning model. For example, the training data may also include: learning input data containing the same type of data as the input data input to the n-th machine learning model, and learning input data for learning the model from the n-th machine corresponding to the input of the learning input data. The output data being output is compared to the training data.

Here, for example, the above-mentioned proximity score takes any value of 0 or 1. For example, when a pair is in a close relationship, the value of the proximity score of the corresponding pair is determined to be 1. In other cases, the value of the proximity score of the corresponding pair is determined to be 0. In this case, the teacher information system may also include the appropriate proximity score value in the corresponding learning input data, and data indicating the probability that the value is appropriate. Then, it may also be based on, for example, the value of the output data output from the n-th machine learning model in response to the input of the learning input data contained in the training data, and the value of the training data contained in the training data, To perform weakly supervised learning that updates the parameter values of the nth machine learning model.

In addition, the above-mentioned proximity score does not necessarily need to be binary data that can only take any value of 0 or 1. For example, the above-mentioned proximity score system can also be a real value (for example, a real value between 0 and above 10), or a multi-stage integer value (for example, a real value above 1) that takes on a larger value when the pair is in a closer relationship. an integer value below 10).

In addition, the learning method of the machine learning model is not limited to weakly supervised learning. As a specific example, consider a pair having a brother-sister relationship. In this case, the input data corresponding to the pairing is input to the learned machine learning model corresponding to the relationship between siblings. Then, for example, if the values of the address data for the pair are the same, the number of gifts given by one party to the other party is 50, and the number of calls so far for the pair is 1200, then the output value is 1. information, such learning can also be performed. For another example, if the values of the address data for the pair are different, the number of gifts given by one party to the other party is 2, and the number of calls so far for the pair is 30, the output value is 0. information, such learning can also be performed. Then, the criterion (such as a threshold) for judging whether the value of the output data corresponding to the proximity score is 1 or 0 can also vary with the machine learning model.

The attribute selection unit 25 selects the type of attribute data generated by the attribute generation unit 26 (the type of attribute data that complements the object) in accordance with the type of relationship between the target user and the reference user. Specific examples of the types of relationships between users and the types of attribute data selected in accordance with the types of relationships include relationships and attribute data exemplified below.

(1) Parent-child relationship or husband-wife relationship living in the same household When the relationship between users is a parent-child relationship or a husband-wife relationship living in the same household, it can be mainly assumed that the variables related to money and the variables representing the actions of the entire household will be the same. Therefore, when the appropriate relationship between users is specified, the attribute selection unit 25 can select, for example, household income, household annual income, as the type of attribute data generated by the attribute generation unit 26 , place of residence, whether there is insurance (for the entire household), amount of savings, financial assets, whether there is a newspaper subscription, etc.

(2)Friendship When the relationship between users is a friend relationship, it can be assumed that groups of the same gender/age/interest are more likely to become friends. Therefore, when the appropriate relationship between users is specified, the attribute selection unit 25 can select, for example, interests and frequently visited places/regions as the type of attribute data generated by the attribute generation unit 26 , age, gender, etc.

(3) Relationship between working in the same workplace When the relationship between users is that they work in the same workplace, it can be assumed that groups with the same educational level and professional fields will tend to work in the same workplace. Therefore, when the appropriate relationship between users is specified, the attribute selection unit 25 can select, for example, the type of purchased professional books, education, etc. as the type of attribute data generated by the attribute generation unit 26. Level etc.

In this embodiment, it has been described that the attribute selection unit 25 selects the type of attribute data of the completion target (generation target) based on rules, but the method of selecting the type of attribute data of the completion target is not limited to Example in this embodiment. For example, it is also possible to select a method of completing the type of attribute data of the object using a machine learning model that has learned the presence or absence of correlation or the degree of correlation between the type of relationship between users and the type of approximate attribute data. .

The attribute generation unit 26 generates attribute data necessary to complete missing attribute data or low-reliability attribute data among the attribute data group of the target user, based on at least 1 specific attribute data for the target user. are generated by referring to information related to the user. Here, as information related to the reference user, the attribute generation unit 26 refers to the attribute data of the type selected by the attribute selection unit 25 among the attribute data groups of the reference user, and generates the referenced attribute data. The attribute data of the corresponding object user is generated.

Specifically, when the relationship between the target user and the reference user is "(1) Parent-child relationship or husband-wife relationship living in the same household", the attribute generation unit 26 generates a system based on household income, The attribute data of the user is referenced based on the attribute data of the user, such as household annual income, place of residence, insurance coverage (for the entire household), amount of savings, financial assets, newspaper subscriptions, etc., and a corresponding map of the target user is generated based on it. Attribute data. In addition, when the relationship between the target user and the reference user is "(2) friend relationship", the attribute generation unit 26 generates attribute data for interests, frequented places/regions, age, gender, etc. The attribute data of the reference user is referred to, and corresponding attribute data of the target user is generated based on it. In addition, when the relationship between the target user and the reference user is "(3) the relationship of working in the same workplace", the attribute generation unit 26 generates the attribute based on the type of purchased professional books, education level, etc. The attribute data refers to the attribute data of the reference user, and based on it, corresponding attribute data of the target user is generated.

The attribute generation unit 26 may also generate the attribute data of the target user by directly copying the parameters of the reference user's attribute data into the corresponding attribute data of the target user. However, the attribute generation unit 26 may also generate corresponding attribute data of the target user by applying certain processing to the parameters of the reference user's attribute data. For example, when generating the attribute data of the target user, the attribute generation unit 26 may also refer to the proximity score determined for the reference user, and generate based on the parameters and proximity score of the reference user's attribute data. Attribute data of the object user.

For example, the attribute generating unit 26 may generate the attribute data of the target user by weighting the parameters of the reference user's attribute data based on the proximity score. In this case, the attribute generation unit 26 sets a larger weighting coefficient as the strength of the relationship between the users represented by the proximity score between the target user and the reference user is higher. Then, by performing processing using weighting coefficients on the parameters of the reference user's attribute data (for example, simply accumulating the weighting coefficients for the parameters, etc.), the parameters of the attribute data that are completed for the target user can be made. Parameters that will be close to the attribute data of the referenced user.

Furthermore, here, when a plurality of reference users are specified, the attribute data of the target user can also be generated based on the plurality of reference users. For example, the attribute generation unit 26 obtains the proximity score and the parameters of the completion target attribute data for each of the plurality of reference users, and performs the operation based on the proximity score based on the parameters obtained from each reference user. Weighting will be based on the average of the plurality of weighted parameters obtained for each reference user (not limited to the average, other statistics such as the median can also be used), as the parameters of the corresponding attribute data of the target user.

Furthermore, for example, the attribute generation unit 26 may use: a parameter that completes at least a part of the attribute data group of the target user before completion, a parameter of at least a part of the attribute data group of the reference user, and the object usage. The proximity score between the user and the reference user is used as the input value, and the attribute data of the target user to be completed is used as the output value of the attribute generation model to generate the attribute data of the target user. As in the case of using weighting, in the case of using an attribute generation model, the attribute generation model is such that the higher the proximity score between the target user and the reference user, the higher the proximity score between the target user and the reference user. The parameters of the complete attribute data will be generated and/or updated in a manner that is closer to the parameters of the referenced user's attribute data. In addition, if the proximity scores and attribute data of multiple reference users are input to the attribute generation model, the parameters of the target user's complementary object attribute data will be output. This is the same as the weighted case mentioned above.

The attribute completion part 27 is to complete the attribute data group about the user based on at least part of the generated attribute data. The attribute data group related to the user includes attribute data including account information and usage history data obtained from the service providing system 5. However, at this time, the attribute complement part 27 will have been modified by the attribute generation part 26. At least part of the generated attribute data is determined to be at least part of the attribute data group related to the target user, and the attribute data group related to the user is completed.

Here, the attribute data completed by the attribute completion part 27 may include: demographic attributes, behavioral attributes, or psychographic attributes. Demographic attributes include, for example, the user's gender, family composition, age, etc.; behavioral attributes include, for example: whether electronic cash is used, whether fixed limit payment is used, and the history of deposits and withdrawals involved in the specified account, including Business transaction history related to certain products of gambling or lottery (which may include online transaction history in online markets, etc.), etc. Psychological variable attributes are, for example, interest in gambling or lottery, etc. However, the user attributes that can be used are not limited to the examples in this embodiment. For example, "time required for customer service (outbound calls, etc.)" and "credit card usage/postpaid billing usage" from call center services can also be used as attribute data.

The user score estimating unit 28 estimates the user score to be set to the user based on the completed attribute data group. In this embodiment, the user score estimation unit 28 inputs the user's attribute data group into the user score estimation model to estimate the user score to be set to the corresponding user. Here, the output value of the user score estimation model is the user score normalized/normalized with 0 as the minimum value and 1 as the maximum value. Here, the attribute data group of the target user input to the user score estimation model includes attribute data generated by the attribute generation unit 26 . As mentioned above, the attribute data generated by the attribute generation unit 26 may include, for example: household income, household annual income, place of residence, insurance coverage (for the entire household), savings amount, financial assets, newspapers Subscription status, interests, places/regions frequently visited, age, gender, types of professional books purchased, education level, etc.

The machine learning unit 29 generates and/or updates the user score estimation model used in the user score estimation by the user score estimation unit 28 . The user score estimation model may be such that when one or a plurality of attribute data (attribute data groups) related to the target user is input, it may represent some measure (such as credit, etc.) related to the user. A machine learning model that outputs user scores can also be some function or statistical model that can output user scores.

When generating and/or updating the user score estimation model, the machine learning unit 29 creates, for each user, a model containing the demographic attributes of the user based on the data obtained from the service providing system 5. The attribute data group is defined as the input value and the user score related to the user is defined as the training data of the output value. Then, the machine learning unit 29 creates a user score estimation model based on the training data. As described above, the attribute data group input to the user score estimation model includes attribute data generated by the attribute generation unit 26, which is combined with the user score of the corresponding user to become training data. Input to machine learning unit 29. The user scores set to the training data may also be rule-based (annotated) user scores. Alternatively, the user score may be a user score that has been corrected by a manager or the like after being outputted in the past by the user score estimation model.

The framework for machine learning model generation that can be used in the implementation of the technology involved in this disclosure is, as an example, based on an ensemble learning algorithm. For example, a machine learning framework (such as LightGBM) based on Gradient Boosting Decision Tree (GBDT) can be used in this framework. In other words, this framework can adopt a machine learning framework based on this kind of decision tree model that inherits the error between the correct answer and the predicted value between the preceding and following weak learners (weak classifiers). The predicted value here refers to, for example, the predicted value of the user's score. In addition, in addition to LightGBM, this framework can also use boosting techniques such as XGBoost or CatBoost. If based on a framework using decision trees, compared to a framework using neural networks, a machine learning model with relatively high performance can be generated with less parameter adjustment procedures. However, the framework for generating machine learning models that can be used when implementing the technology involved in the present disclosure is not limited to the examples in this embodiment. For example, as a learner, the gradient boosting decision tree can be replaced by other learners such as random forest, or a learner such as a neural network that is not called a weak learner can be used. Furthermore, especially when a learner such as a neural network, which is not called a so-called weak learner, is used, ensemble learning does not need to be used.

FIG. 13 is a simplified illustration of the concept of a decision tree of the machine learning model used in this embodiment. In the case of using a machine learning framework based on gradient boosting of the decision tree algorithm, the optimization of the divergence conditions of each node of the decision tree will be performed. Specifically, in the machine learning framework of gradient boosting based on the decision tree algorithm, user scores are calculated for user groups with attributes represented by each of the two child nodes branched from a parent node. The divergence condition of the parent node is optimized so that the difference in user scores becomes larger (for example, the difference becomes the maximum, or exceeds a predetermined threshold), that is, the two child nodes can clearly diverge. For example, if the attribute represented as the divergence condition of a node is age, the age set as the threshold for divergence may be changed, or the divergence condition may be changed to an attribute other than age. In this way, by recursively optimizing the divergence conditions of all nodes of the decision tree, the estimation accuracy of the user score based on the attribute data group can be improved.

In addition, when the attribute generation unit 26 generates the attribute data of the completion target using the attribute generation model, the machine learning unit 29 also generates the completion target attribute data of the target user caused by the attribute generation unit 26. The attribute generation model used is generated and/or updated. The attribute generation model is machine learning that outputs complete object attribute data about the target user when one or more attribute data and proximity scores about one or more reference users are input. Model.

When generating and/or updating the attribute generation model, the machine learning unit 29 defines the attribute data and proximity scores of one or more reference users among the data obtained from the service providing system 5 as input values. And define one attribute data (complete object attribute data about the object user) as the training data of the output value. Here, the output values set in the training data used when generating and/or updating the attribute generation model (parameters for completing the object attribute data of the target user) may also be based on rules (such as the above The output value is set (annotated) based on the calculation method due to weighting. Alternatively, the output value may be an output value that has been corrected by a manager or the like after being outputted in the past using an attribute generation model.

Then, the machine learning unit 29 generates or updates the attribute generation model based on the training data. One or more attribute data and proximity scores are combined with corresponding attribute data to become training data and are input to the machine learning unit 29 . In addition, in the generation or update of attribute generation models, the framework for generating machine learning models that can be used is not limited, but the machine learning framework of gradient boosting based on the decision tree algorithm can be used. This is explained in the above description. The user score inference model is the same.

＜Processing flow＞ Next, the flow of processing executed by the information processing system according to this embodiment will be described. In addition, the specific content and processing sequence of the processing described below are only examples necessary for implementing the present disclosure. The specific processing content and processing sequence can be appropriately selected according to the implementation form of the present disclosure.

FIG. 14 is a flowchart showing the flow of machine learning processing according to this embodiment. The processing shown in this flowchart is executed at the timing specified by the administrator.

In this embodiment, during the machine learning process, the user score estimation model is generated and/or updated. The machine learning unit 29 creates training data including a combination of the attribute data group of each user accumulated in the past in the service providing system 5 and the user score predetermined for the corresponding user (step S101). Then, the machine learning unit 29 inputs the created training data into the user score estimation model, and generates the user score estimation model used in the user score estimation by the user score estimation unit 28 or Update (step S102). Thereafter, the processing shown in this flowchart ends. In addition, when the attribute generation unit 26 uses an attribute generation model for attribute completion, the generation and/or update of the attribute generation model may also be performed in the same process flow.

FIG. 15 is a flowchart showing the flow of user score estimation processing according to this embodiment. The processing shown in this flowchart is executed for each user of the object at the timing specified by the administrator. Here, the target user is a user whose attribute data is defective or whose attribute data is less reliable. Examples of attribute data with low reliability include attribute data generated based on an insufficient amount of accumulated historical data, attribute data that clearly contradicts the contents of other attribute data, and the like. In addition, it is assumed here that graphic data on a plurality of users including the target user has been generated, and that each machine learning model has been learned.

In steps S201 and S203, the reference user is specified, and the relationship between the target user and the reference user is specified. The reference user specifying part 22 refers to one or a plurality of other users corresponding to the node data 50 that are linked explicitly or implicitly to the node data 50 corresponding to the target user with reference to the graph data, The reference user is specified (step S201). Then, for each pairing of the target user and the one or plurality of reference users specified in step S201, the relationship specifying unit 23 determines the type of relationship between the users (specifically, residence). Parent-child relationship/couple relationship/friend relationship/relationship working in the same workplace, etc.) in the same family are specified (step S202). Thereafter, the process proceeds to step S203.

In steps S203 and S204, the type of attribute data that is the completion object is selected, and the proximity score between users is determined. The attribute selection unit 25 selects the type of attribute data to be completed for the target user in accordance with the type of relationship specified in step S202 (step S203). Furthermore, the relationship strength determination unit 24 determines, for each pairing between the target user and each reference user, the value of the proximity score to be associated with the pairing (S204). Thereafter, the process proceeds to step S205.

In step S205, attribute data completed for the target user is generated. The attribute generation unit 26 generates an attribute for the target user based on the parameters of the attribute data of the reference user corresponding to the attribute data of the completion target and the proximity score determined in step S204 for the reference user. The completed attribute data. Thereafter, the process proceeds to step S206.

In steps S206 and S207, the user score system is estimated and output. The attribute complementing unit 27 adds the completed attribute data generated in step S205 as the corresponding user to a group of attribute data obtained from the service providing system 5 and held in advance for the target user. attribute data group (step S206). Then, the user score estimation unit 28 inputs the attribute data group including the attribute data completed for the target user in step S206 into the user score estimation model, and treats the output value as the value to be evaluated. The user score set to the user is obtained (step S207). However, the method of estimating the user score is not limited to the example in this embodiment. For example, the user score system may include a value calculated by inputting the attribute data group to a predetermined function of a non-machine learning model. Thereafter, the processing shown in this flowchart ends.

The user score set for each user is provided to other systems such as the service providing system 5 and so on, and is used to provide services to the target users through other systems such as the service providing system 5 etc. Customization, etc.

This embodiment can also be used to estimate the user score of a newly added target user whose corresponding node data 50 is not included in the graph node. For example, the node data 50 corresponding to the target user and at least one link data 52 connected to the node data 50 may also be generated based on the user attribute data of the newly added target user. Then, the user connected to the node data 50 corresponding to the target user through the link data 52 can also be specified as a reference user of the target user.

＜Effect＞ According to this embodiment, the user's missing attributes can be completed based on the social graph network that includes the relationship between users, and the user's score can be estimated/determined using the completed attribute group. In cases where the target user's information is missing or the information is less reliable, the user score can still be calculated, or the accuracy of the calculated user score can be improved. In addition, by using various user attribute data, even when a certain range of attribute data (such as that of the credit card department) cannot be used due to regulations or laws, etc., or when only a part of the target user is used. When attribute data exists, high-precision user scores can still be calculated.

＜Modification＞ The above-described embodiment includes a graphic data generating unit 21, a reference user specifying unit 22, a relationship specifying unit 23, a relationship strength determining unit 24, an attribute selecting unit 25, an attribute generating unit 26, and an attribute complementing unit 27. An example of the information processing device of the user score estimation unit 28 and the machine learning unit 29 will be described. However, part of these functional units may be omitted within the scope of implementing the invention of the present disclosure.

For example, in the embodiment described above, the relationship strength (proximity score) between the target user and the reference user is generated or referred to when the attribute data of the completion object is generated. However, when the attribute data of the completion object is generated, When generating, the generation and reference of the proximity score can also be omitted. In this case, among the functional units of the information processing device 1 described with reference to FIG. 2 , the relationship strength determining unit 24 may be omitted. In addition, when generating attribute data, the attribute generation unit 26 may generate the complementary target attribute data of the target user based on the attribute data of the reference user without performing weighting with reference to the proximity score.

For example, the attribute generation unit 26 may also use at least a part of the parameters of the attribute data group of the reference user and the proximity score between the target user and the reference user as input values, and use the parameters to be supplemented as input values. The complete attribute data of the object user is used as the attribute generation model of the output value to generate the attribute data of the object user. At this time, the attribute generation model is appropriately learned in advance according to the form of the input value and the output value.

Furthermore, for example, the attribute generation unit 26 may also use: parameters of at least a part of the attribute data group of the target user and/or parameters of at least a part of the attribute data group of the reference user as input values, and generate the required parameters. The completed attribute data of the object user is used as the attribute generation model of the output value to generate the attribute data of the object user. At this time, the attribute generation model is appropriately learned in advance according to the form of the input value and the output value. In addition, at this time, the attribute generation unit 26 is based on a plurality of attribute generation models that differ according to the relationship and/or proximity scores between each target user and the reference user, and processes accordingly. The type of relationship and/or the proximity score between the object user of the object and the reference user determines a predetermined attribute generation model to generate attribute data of the object user to be completed. Here, each of the plurality of attribute generation models, for example, may be pre-learned based on training data in which the type of relationship and/or the proximity score are common or similar (falling within a predetermined range). handle.

Furthermore, for example, the attribute generation unit 26 may use the parameter that is at least a part of the attribute data group of the user (target user, reference user) to be the embedded representation (vector representation, feature) of the user on the graphic data. Expression) is used as the input value and the attribute data of the object user to be completed is used as the output value of the attribute generation model to generate the attribute data of the object user. In addition, the attribute generation model can include the distance or inner product between the target user and the reference user on the graphic data (the distance or inner product on the vector space based on the graphic data) in the input value. At this time, the attribute generation model is appropriately learned in advance according to the form of the input value and the output value.

Also, for example, the attribute completion part 27 is when the attribute data output by the attribute generation model is a missing value (missive attribute data) in the attribute data group of the target user before completion is performed, or In the case of inappropriate values (attribute data with low reliability), the attribute data that has been output can be determined to become part of the attribute data group of the target user.

Furthermore, for example, the attribute selecting unit 25 or the attribute complementing unit 27 may regard attribute data with a higher weight in an ensemble learning model such as a gradient boosting decision tree used as a user score estimation model or the like as a complement target. attribute information. Here, the attribute data with a higher weight may be, for example, attribute data corresponding to a tree whose weight exceeds a predetermined weight in the user score estimation model, or may be represented in the user score estimation model. The attribute data corresponding to the tree with the higher weight (above the determined order).

1:Information processing device 5:Service provision system 11:CPU 12:ROM 13:RAM 14:Memory device 15: Communication unit 21: Graphic data generation department 22: Refer to user-specific parts 23: Relational specific department 24:Relationship Strength Determining Department 25:Attribute selection department 26:Attribute generation department 27: Complement all attributes 28: User score estimation part 29:Machine Learning Department 40: E-commerce transaction system 42: Golf course reservation system 44:Travel reservation system 46:Card management system 50(50a,50b,50c,50d,50e,50f,50g,50h,50i,50j,50k,50l,50m,50n): Node information 52(52a,52b,52c,52d,52e,52f,52g,52h,52i,52j,52k,52l,52m,52n,52o,52p): Link data 54(54a,54b,54c,54d,54e):Gathering

[Fig. 1] A schematic diagram of the structure of the information processing system according to the embodiment. [Fig. 2] A schematic diagram showing the functional configuration of the information processing device according to the embodiment. [Fig. 3] A schematic diagram showing an example of a common value of IP address data in the embodiment. [Fig. 4] An illustration of an example of graphic data according to the embodiment. [Fig. 5] A schematic diagram showing an example of a common value of address data in the embodiment. [Fig. 6] An illustration of an example of graphic data according to the embodiment. [Fig. 7] A schematic diagram showing an example of a common value of credit card number data in the embodiment. [Fig. 8] An illustration of an example of graphic data according to the embodiment. [Fig. 9] An illustration of an example of graphic data according to the embodiment. [Fig. 10] An illustration of an example of clustering according to the embodiment. [Fig. 11] An illustration of an example of visualization of classification according to the embodiment. [Fig. 12] A diagram illustrating an example of determination of relationship strength (proximity score) using a machine learning model according to the embodiment. [Fig. 13] A simplified illustration of the concept of a decision tree of the machine learning model used in the embodiment. [Fig. 14] A flowchart of the flow of machine learning processing described in the embodiment. [Fig. 15] A flowchart illustrating the flow of user score estimation processing according to the embodiment.

1:Information processing device

21: Graphic data generation department

22: Refer to user-specific parts

23: Relational specific department

24:Relationship Strength Determining Department

25:Attribute selection department

26:Attribute generation department

27: Complement all attributes

28: User score estimation part

29:Machine Learning Department

Claims (15)

An information processing system having: Reference user specifying means is used to identify the reference user who is related to the target user; and Attribute generation means is used to generate corresponding attribute data of the object user based on the attribute data of a specific prefix reference user for the user of the object; and Attribute completion means is used to complete the corresponding attribute data group of the user of the preceding object based on at least part of the corresponding attribute data of the user of the preceding object that has been generated; and The user score estimation means is used to estimate the user score to be set to the target user based on the corresponding previous attribute data group that has been completed for the previous target user. An information processing system as described in claim 1, wherein, The means of specifying a reference user in the preamble refers to specifying the reference user in the preamble based on graphic data showing the relationship between users. An information processing system as described in claim 2, wherein, It also has: graphic data generating means for specifying a pair of users who are related to each other based on the attribute data group of each of the plurality of users to generate the aforementioned graphic data. An information processing system as described in claim 1, wherein, It also has: relationship specific means, which is used to specify the relationship between users. An information processing system as described in request item 4, wherein, The above-mentioned relationship specifying means specifies the relationship between the users mentioned above based on the result of clustering based on the values corresponding to the relationships between the users. An information processing system as described in request item 5, wherein, The aforementioned relationship specific method is the result of clustering based on at least one of the aforementioned user's name, IP address, address, credit card number, age, gender, place of study, place of work, and place of stay, and Specify the relationship between the users mentioned above. An information processing system as described in request item 4, wherein, It also includes: a relationship strength determination means that is based on an index indicating the strength of the relationship between the target user and the reference user based on the judgment criteria corresponding to the relationship between the target user mentioned above and the reference user mentioned above. Determine the relationship strength indicating the distance between the object user and the reference user; The means for generating the prescript attribute is to generate attributes corresponding to the prescript target user based on information about at least one prescript reference user and the determined relationship strength of the prescript with respect to the reference user. material. An information processing system as described in claim 7, wherein, The means for determining the strength of the aforementioned relationship determines the user who represents the aforementioned object based on the output of a machine learning model that has been learned corresponding to the relationship between the aforementioned user and the aforementioned reference user and inputs data representing the aforementioned index. The strength of the relationship between the user and the reference user. An information processing system as described in claim 1, wherein, It also has: attribute selection means for selecting the type of prescript attribute data generated by the prescript attribute generating means in accordance with the type of relationship between the prescript object user and the prescript reference user; The attribute generating means mentioned above is to generate corresponding attribute data of the object user based on the attribute data of the type selected by the attribute selecting means mentioned above, among the attribute data group of the reference user mentioned above. An information processing system as described in claim 1, wherein, The user score estimation method mentioned above is to estimate the user score to be set to the target user by inputting the attribute data group of the target user mentioned above into the machine learning model. An information processing system as described in claim 10, wherein, The aforementioned user score estimation method uses: a machine learning model generated based on the machine learning framework of the gradient boosting decision tree to estimate the aforementioned user score. An information processing system as described in claim 10, wherein, The aforementioned user score estimation method is generated by using training data that uses an attribute data group containing the user's demographic attributes as an input value and uses the previous user score related to the user as an output value. The machine learning model of the prefix is used to estimate the user score to be set to the prefix target user. An information processing system as described in claim 1, wherein, The attribute completion method mentioned above is the attribute data required to complete the attribute data that is missing or has low reliability in the attribute data group of the target user mentioned above. It is based on the attribute data of the user mentioned above. be generated. An information processing method performed by a computer: The reference user specific step is used to identify the reference user who has a relationship with the target user; and The attribute generation step is used to generate corresponding attribute data of the object user based on the attribute data of the specified previous reference user for the previous object user; and The attribute completion step is used to complete the corresponding attribute data group of the user of the previous object based on at least part of the corresponding attribute data of the user of the previous object that has been generated; and The user score estimation step is used to estimate the user score to be set to the target user based on the corresponding previous attribute data group that has been completed for the previous target user. A program product that enables a computer to function as: Reference user specifying means is used to identify the reference user who is related to the target user; and Attribute generation means is used to generate corresponding attribute data of the object user based on the attribute data of a specific prefix reference user for the user of the object; and Attribute completion means is used to complete the corresponding attribute data group of the user of the preceding object based on at least part of the corresponding attribute data of the user of the preceding object that has been generated; and The user score estimation means is used to estimate the user score to be set to the target user based on the corresponding previous attribute data group that has been completed for the previous target user.

Images