JP7724753B2 - Information processing device, method, and program - Google Patents

Description

This disclosure relates to technology for assisting users in performing operations.

Conventionally, call center operators make phone calls to customers to urge them to pay their credit card bills, i.e., to collect debts (see Patent Document 1). Also known is a technology for supporting urging customers to change the tone of the sample phrases used by operators when responding to calls depending on the time of day the urging work is being carried out (see Patent Document 2).

Japanese Patent Application Laid-Open No. 2001-282994 JP 2010-224617 A

Conventionally, operations (such as calling a customer) have been performed to prompt users to take a specific action (such as transferring money or depositing funds into an account), and it is desirable to quantitatively evaluate the effectiveness of such operations. However, some users may lack the data necessary to evaluate the effectiveness of the operation, making it difficult to carry out an evaluation.

In consideration of the above-mentioned problems, the present disclosure aims to estimate the effectiveness of an operation that prompts a user to take a specific action, even for users who lack the data necessary to evaluate the effectiveness of the operation.

One example of the present disclosure is an information processing device that includes an effect estimation means that estimates the effect that a predetermined operation on a user to prompt the user to perform a predetermined action will have on whether the user will perform the action, and a reference user identification means that identifies at least one reference user who is related to a target user, and the effect estimation means estimates a target effect, which is the effect on the target user, based on data including a reference effect, which is the effect estimated for the reference user.

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

This disclosure makes it possible to estimate the effectiveness of an operation that prompts a user to take a specific action, even for users who lack the data necessary to evaluate the effectiveness of that operation.

1 is a schematic diagram illustrating a configuration of an information processing system according to an embodiment. FIG. 1 is a diagram illustrating an outline of a functional configuration of an information processing apparatus according to an embodiment. FIG. 10 is a diagram illustrating an example in which values of IP address data are common in the embodiment. FIG. 10 is a diagram illustrating an example of graph data according to the embodiment. FIG. 10 is a diagram illustrating an example in which address data values are common in the embodiment. FIG. 10 is a diagram illustrating an example of graph data according to the embodiment. FIG. 10 is a diagram illustrating an example in which the value of credit card number data is common in the embodiment. FIG. 10 is a diagram illustrating an example of graph data according to the embodiment. FIG. 10 is a diagram illustrating an example of graph data according to the embodiment. FIG. 2 is a diagram illustrating an example of a cluster according to an embodiment. FIG. 10 is a diagram illustrating an example of visualization of classification according to the embodiment. FIG. 10 illustrates an example of determining a relationship strength (closeness score) using a machine learning model according to an embodiment. FIG. 1 is a diagram illustrating a simplified concept of a decision tree of a machine learning model employed in an embodiment. FIG. 10 is a diagram showing the relationship between estimated effects and risks and operation conditions in an embodiment. 1 is a flowchart illustrating a flow of machine learning processing according to an embodiment. 10 is a flowchart showing the flow of an operation condition output process according to the embodiment. 10 is a flowchart showing a flow of variation (1) of an operation condition output process according to the embodiment. 10 is a flowchart showing a flow of variation (2) of the operation condition output process according to the embodiment. FIG. 10 is a diagram showing the relationship between the estimated target causal score, the reference causal score, and the risk index and the operation conditions according to the embodiment. FIG. 10 is a diagram showing the relationship between the estimated target causal score, the reference causal score, and the risk index and the operation conditions according to the embodiment. 10 is a flowchart showing the flow of an operation condition output process when the relationships between the target causal score, the reference causal score, and the risk index and the operation condition according to the embodiment are adopted. 10 is a flowchart showing the flow of an operation condition output process when the relationships between the target causal score, the reference causal score, and the risk index and the operation condition according to the embodiment are adopted.

Embodiments of an information processing device, method, and program according to the present disclosure will be described below with reference to the drawings. However, the embodiments described below are merely examples, and the information processing device, method, and program according to the present disclosure are not limited to the specific configurations described below. When implementing the present disclosure, specific configurations may be adopted as appropriate depending on the implementation mode, and various improvements and modifications may be made.

In this embodiment, we will explain an embodiment in which the technology disclosed herein is implemented in an operations center management system for collecting debts by urging payment of overdue credit card amounts. However, systems to which the technology disclosed herein can be applied are not limited to operations center management systems for urging payment of credit card amounts. The technology disclosed herein can be widely used in technologies for supporting operations for users, and the application of the present disclosure is not limited to the examples shown in the embodiments.

Normally, credit card payments are made by debiting the user's account on the monthly withdrawal date or by the user depositing the amount by a specified date. However, there are cases where payment of the credit card amount is not completed by the specified date due to reasons such as insufficient funds in the user's account or the user failing to deposit the amount by the specified date. For this reason, operations centers (call centers) have traditionally made phone calls or sent messages to customers to urge users to pay their credit card amounts and collect their debts.

In general, operations directed at users are effective in avoiding default (non-performance of debt), and the more operations there are, the higher the debt recovery rate. However, on the other hand, the more operations there are, the higher the costs for operations, such as labor costs, system usage fees, and system maintenance fees. Therefore, in consideration of the above-mentioned problems, the system disclosed herein employs technology to reduce operation costs without reducing debt recovery rates. Note that, while this embodiment mainly describes an example in which the specified operation is a phone call to the user, the content of the specified operation is not limited and may be a variety of operations to encourage the user to take a specified action.

<System configuration>
1 is a schematic diagram showing the configuration of an information processing system according to this embodiment. In the system according to this embodiment, an information processing device 1, an operations center management system 3, and a credit card management system 5 are connected to each other so that they can communicate with each other. An operation terminal (not shown) is installed in the operations center for performing operations in accordance with instructions from the operations center management system 3, and an operator operates the operation terminal to perform operations for a user. The user is a credit card user who deposits a credit card usage amount via a financial institution or the like, and payment history data for the credit card usage amount is notified to the operations center management system 3 via the credit card management system 5.

The information processing device 1 is an information processing device for outputting data for controlling operations by the operation center management system 3. The information processing device 1 is a computer equipped with a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage device 14 such as an EEPROM (Electrically Erasable and Programmable Read Only Memory) or an HDD (Hard Disk Drive), 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 omitted, replaced, or added as appropriate depending on the embodiment. Furthermore, the information processing device 1 is not limited to a device consisting of a single housing. The information processing device 1 may be realized by multiple devices using so-called cloud or distributed computing technology, etc.

The operation center management system 3, credit card management system 5, and operation terminal are all computers equipped with a CPU, ROM, RAM, storage device, communication unit, input device, output device, etc. (not shown). Furthermore, these systems and terminals are not limited to devices consisting of a single housing. These systems and terminals may be realized by multiple devices using so-called cloud or distributed computing technology, etc.

Figure 2 is a diagram showing an outline of the functional configuration of the information processing device 1 according to this embodiment. The information processing device 1 functions as an information processing device equipped with a graph data generation unit 21, a reference user identification unit 22, a relationship identification unit 23, a relationship strength determination unit 24, an effect estimation unit 25, a risk estimation unit 26, a machine learning unit 27, and a condition output unit 28, by reading a program recorded in the storage device 14 into the RAM 13 and executing it by the CPU 11, which controls each piece of hardware equipped in the information processing device 1. Note that in this embodiment and other embodiments described below, each function equipped in the information processing device 1 is executed by the CPU 11, which is a general-purpose processor, but some or all of these functions may be executed by one or more dedicated processors.

The graph data generation unit 21 generates graph data (social graph network) indicating the relationships between users by identifying pairs of users who are related to each other based on the attribute data groups of each of the users. More specifically, the graph data generation unit 21 generates graph data including, for example, node data 50 associated with each of multiple users, including the target user, and link data 52 associated with pairs of users who are related to each other (see Figures 4, 6, 8, and 9). The graph data generation unit 21 predicts and creates implicit links between users by learning (representation learning, relationship learning, embedding learning, knowledge graph embedding) a user relationship graph composed of nodes (users) connected by explicit links. In this case, the graph data generation unit 21 may perform this learning based on a known embedding model or its extension, as appropriate.

For example, as shown in Figure 3, suppose that a group of attribute data for user A is registered in electronic commerce system 40. Also, suppose that a group of attribute data for user B is registered in golf course reservation system 42. Also, suppose that attribute data for user C is registered in travel reservation system 44. Then, suppose that the value of the IP address data for user A registered in electronic commerce system 40, the value of the IP address data for user B registered in golf course reservation system 42, and the value of the IP address data for user C registered in travel reservation system 44 are the same.

In this case, as shown in FIG. 4, the graph data generation unit 21 generates graph data including node data 50a associated with user A, node data 50b associated with user B, node data 50c associated with user C, link data 52a indicating that user A is related to user B, link data 52b indicating that user A is related to user C, and link data 52c indicating that user B is related to user C. It is presumed that users with the same IP address use the same computer or share a global address at the same residence or workplace. For this reason, in this embodiment, such users are associated with each other.

Also, for example, as shown in Figure 5, suppose that attribute data groups for user D, user E, and user F are registered in the e-commerce system 40. Furthermore, suppose that the address data values for user D, user E, and user F registered in the e-commerce system 40 are the same.

In this case, as shown in FIG. 6, the graph data generation unit 21 generates graph data including node data 50d associated with user D, node data 50e associated with user E, node data 50f associated with user F, 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 52f indicating that user E is related to user F. Users with the same address are presumed to live together. For this reason, in this embodiment, such users are associated with each other.

For example, as shown in FIG. 7, suppose that a group of attribute data for user G is registered in electronic commerce system 40. Also, suppose that a group of attribute data for user H is registered in golf course reservation system 42. Also, suppose that a group of attribute data for user I is registered in travel reservation system 44. Then, suppose that the value of user G's credit card number data registered in electronic commerce system 40, the value of user H's credit card number data registered in golf course reservation system 42, and the value of user I's credit card number data registered in travel reservation system 44 are the same.

In this case, as shown in FIG. 8, the graph data generation unit 21 generates graph data including node data 50g associated with user G, node data 50h associated with user H, node data 50i associated with user I, 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 52i indicating that user H is related to user I. Users with the same credit card number are presumed to be family members, such as parent and child. For this reason, in this embodiment, such users are associated with each other.

Note that the criteria for determining whether a pair of users is related to each other are not limited to those described above. User pairs can be determined based on various criteria, such as location history and behavioral history.

Furthermore, the link indicated by the link data 52 associating users identified as having a relationship with each other, as described above, will be referred to as an explicit link. For example, suppose a user connected to a first user via an explicit link and a user connected to a second user via an explicit link have a predetermined number or more of them in common (e.g., three or more). In this case, in this embodiment, for example, the graph data generation unit 21 generates link data 52 indicating that the first user is related to the second user. The link indicated by the link data 52 generated in this way will be referred to as an implicit link.

For example, as shown in FIG. 9, assume that link data 52j indicating an explicit link connects node data 50j associated with user J and node data 50k associated with user K. Also, assume that link data 52k indicating an explicit link connects node data 50j associated with user J and node data 50l associated with user L. Also, assume that link data 52l indicating an explicit link connects node data 50j associated with user J and node data 50m associated with user M.

Also, it is assumed that link data 52m indicating an explicit link connects node data 50k associated with user K and node data 50n associated with user N. Also, it is assumed that link data 52n indicating an explicit link connects node data 50l associated with user L and node data 50n associated with user N. Also, it is assumed that link data 52o indicating an explicit link connects node data 50m associated with user M and node data 50n associated with user N.

In this case, the graph data generation unit 21 generates link data 52p (link data 52p indicating an implicit link) indicating that user J has a relationship with user N. In this way, user N is identified as a user who has a relationship with user J.

Also, for example, suppose that a user who is connected to a first user via an explicit or implicit link and a user who is connected to a second user via an explicit or implicit link share a predetermined number of common users (e.g., three or more). In this case, the graph data generation unit 21 may generate link data 52 indicating that the first user has a relationship with the second user (link data 52 indicating an implicit link).

The reference user identification unit 22 references the graph data generated by the graph data generation unit 21 and identifies at least one other user who is related to the target user among the users included in the graph data as a reference user for the target user. Here, the reference user identification unit 22 may identify at least one reference user who is related to the target user according to the relationship strength (closeness score between users) described below. Specifically, the reference user identification unit 22 can preferentially identify users who have a strong relationship strength (high closeness score) with the target user as reference users. For example, if the reference user identification unit 22 wishes to identify n reference users, it identifies the top n users who have the strongest relationship strength with the target user as reference users. Furthermore, the reference user identification unit 22 may identify users who are identified as users who are related to the target user and users who have a predetermined number or more of users identified as related users in common with the target user as reference users. Furthermore, the reference user identification unit 22 may identify users who are classified into the same cluster as the target user as reference users.

The reference user identification unit 22 may also identify a reference user from among a plurality of users based on the attributes of the target user and the attributes of the plurality of users. For example, other users who share an address or telephone number with the target user can be inferred to be cohabiting users who live with the target user, or users who share lifestyles or social activities with the target user even if they have different addresses. Therefore, the reference user identification unit 22 can identify other users who share attribute data such as addresses and telephone numbers with the target user as at least one reference user. The reference user identification unit 22 may also identify a user for whom a flag common to the target user is enabled (set to 1). The reference user identification unit 22 may also identify a user for whom various data estimated or determined based on the attribute data group exhibits a predetermined value or a predetermined combination, etc., as a reference user. The reference user identification unit 22 may identify a reference user based on at least one of the conditions for identifying a reference user.

The reference user identification unit 22 may, for example, identify a user associated with node data 50 that is connected to the target user by link data 52 indicating an explicit link or an implicit link, as a reference user for the target user.

The relationship identification unit 23 identifies relationships between users. Examples of relationships between users identified here include (1) parent-child or marital relationships between people living in the same household, (2) friendships, and (3) relationships between people working in the same workplace. However, the relationships identified are not limited to the examples in this disclosure. In this embodiment, the relationship identification unit 23 identifies relationships between users based on the results of clustering based on values associated with relationships between users. Here, the types of values that can be used as values associated with relationships between users are not limited, but may include, for example, at least one of the user's name, IP address, address, credit card number, age, gender, place of school, place of work, and place of stay.

The relationship identification unit 23 identifies the relationship between users. Here, the relationship identification unit 23 may identify 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. Furthermore, 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 relationship between the target user and the reference user may be identified based on the attribute data group of the target user registered in the e-commerce system 40 and the attribute data group of the reference user registered in the golf course reservation system 42.

The relationship identification unit 23 identifies, for example, pairs of node data 50 connected by link data 52. Then, the relationship identification unit 23 generates pair attribute data associated with the pair based on the user attribute data group of the two users associated with the pair. Here, the pair 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, pair gender data, a common school flag, a common workplace flag, a common place of stay flag, etc.

The IP common flag is, for example, a flag that indicates whether the value of the IP address data included in one attribute data of the pair is the same as the value of the IP address data included in the other attribute data. For example, if the IP address data values are the same, the value of the IP common flag may be set to 1, and if the IP address data values are different, the value of the IP common flag may be set to 0.

The address common flag, school common flag, workplace common flag, and place of stay common flag are flags that indicate, for example, whether the value of the address data, school data, workplace data, or place of stay data included in one attribute data group of the pair is the same as the value of the address data, school data, workplace data, or place of stay data included in the other attribute data group. For example, if the address data values are the same, the address common flag may be set to 1, and if the address data values are different, the address common flag may be set to 0.

The credit card number common flag is, for example, a flag that indicates whether the value of the credit card number data included in one attribute data group of the pair is the same as the value of the credit card number data included in the other attribute data group. For example, if the values of the credit card number data are the same, the value of the credit card number common flag may be set to 1, and if the values of the credit card number data are different, the value of the credit card number common flag may be set to 0.

The same last name flag is a flag that indicates, for example, whether the last name indicated by the name data included in one attribute data group of the pair is the same as the last name indicated by the name data included in the other attribute data group. For example, if the last names indicated by the name data are the same, the value of the same last name flag may be set to 1, and if the last names indicated by the name data are different, the value of the same last name flag may be set to 0.

Age difference data is, for example, data indicating the difference between the value of age data included in one attribute data group of the pair and the value of age data included in the other attribute data group.

Paired gender data is, for example, data that indicates a combination of the gender data value included in one attribute data group of the pair and the gender data value included in the other attribute data group.

Then, the relationship identification unit 23 performs clustering using a general clustering method based on the values of the pair attribute data group associated with each of the multiple pairs, thereby classifying the multiple pairs into multiple clusters 54 as shown in FIG. 10.

Figure 10 is a diagram schematically illustrating an example of how multiple pairs are classified into five clusters 54 (54a, 54b, 54c, 54d, and 54e). The crosses shown in Figure 10 correspond to pairs. Each of the multiple crosses is located at a position corresponding to the value of the pair attribute data of the pair corresponding to the cross. In the example of Figure 10, multiple pairs are classified into five clusters 54, but the number of clusters 54 into which multiple pairs are classified is not limited to five; for example, multiple pairs may be classified into four clusters 54.

Figure 11 is a diagram showing an example of visualization of classification when multiple pairs are classified into four clusters 54. As shown in Figure 11, pairs with the same address, the same gender, an age difference of more than X years, and the same last name may be classified into the first cluster. Pairs with the same address, the same gender, an age difference of X years or less, and the same last name may be classified into the second cluster. Pairs with the same address, different genders, an age difference of more than Y years, and the same last name may be classified into the third cluster. Pairs with the same address, different genders, an age difference of Y years or less, and the same last name may be classified into the fourth cluster.

In this case, the first cluster is presumed to be cluster 54 associated with, for example, parents and children of the same sex. The second cluster is presumed to be cluster 54 associated with siblings of the same sex. The third cluster is presumed to be cluster 54 associated with parents and children of the opposite sex. The fourth cluster is presumed to be cluster 54 associated with a married couple.

As described above, the relationship identification unit 23 may identify the relationship between the target user and the reference user based on the results of clustering based on values associated with relationships between users. Specific examples of creating clusters of friendships or working at the same workplace by clustering based on the common school flag, common workplace flag, and common place of stay flag are similar to the examples described above, and therefore will not be described here. The relationship identification unit 23 may also identify the relationship between the target user and the reference user based on the results of clustering based on at least one of last name, IP address, address, credit card number, age difference, gender, school, workplace, and place of stay.

The relationship strength determination unit 24 determines the relationship strength (hereinafter also referred to as the "closeness score") indicating the closeness between users (for example, between a target user and a reference user) based on an index indicating the strength of the relationship between the users, in accordance with criteria corresponding to the relationship between the users. In this embodiment, the relationship strength determination unit 24 determines the relationship strength (closeness score) indicating the closeness between the target user and the reference user based on the output when data representing the index is input into a trained machine learning model corresponding to the relationship between the target user and the reference user.

Here, the relationship strength determination unit 24 may include trained machine learning models each associated with the above-mentioned clusters 54. For example, if multiple pairs are classified into five clusters 54, the relationship strength determination unit 24 may include five machine learning models. The relationship strength determination unit 24 may then determine a closeness score indicating the closeness between the target user and the reference user based on the output when data representing an index indicating the strength of the relationship between the target user and the reference user is input into a trained machine learning model corresponding to the relationship between the target user and the reference user. In this case, the input/output relationship implemented in the trained machine learning model corresponds to the above-mentioned judgment criteria.

As shown in FIG. 12, the relationship strength determination unit 24 may input input data corresponding to a pair classified into a cluster 54 associated with the nth machine learning model to the nth machine learning model, which is the nth machine learning model. For example, if the relationship strength determination unit 24 includes five machine learning models, the above-mentioned value n may be any integer between 1 and 5. Then, the relationship strength determination unit 24 may determine the value of the output data output from the nth machine learning model in response to the input of the input data as the value of the closeness score for the pair.

The input data associated with a pair may include, for example, some or all of the pair attribute data associated with the pair. The input data may also include data not included in the pair attribute data. For example, the input data may include data indicating the usage history of the e-commerce system 40, or data obtained by the relationship strength determination unit 24 from other information sources such as SNS. More specifically, for example, the input data may include data indicating the number of calls or messages exchanged between the pair per unit period, the number of gifts sent by one party to the other, the number of mutual friends the pair has, etc.

Furthermore, the type of data included in the input data associated with a pair may be the same or different depending on the cluster 54 to which the pair 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 relationship strength determination unit 24 determines the closeness score, the nth machine learning model is trained in advance using a given plurality of training data items associated with the nth machine learning model. This training data is, for example, prepared in advance so that the determination of the closeness score for the cluster 54 associated with the nth machine learning model is valid. Here, the closeness score set in the training data may be a rule-based (annotated) closeness score. It may also be a closeness score that was previously output by the machine learning model and then revised by an administrator, etc.

Here, the nth machine learning model may be trained using weakly supervised learning. For example, the training data may include training input data that contains the same type of data as the input data input to the nth machine learning model, and training data that is compared with output data output from the nth machine learning model in response to the input of the training input data.

Here, for example, the closeness score described above takes on a value of either 0 or 1. For example, if a pair is closely related, the closeness score value for that pair is determined to be 1, and if not, the closeness score value for that pair is determined to be 0. In this case, the training data may include a valid closeness score value for the corresponding training input data and data indicating the probability that this value is valid. Then, for example, weakly supervised learning may be performed to update the parameter values of the nth machine learning model based on the value of the output data output from the nth machine learning model in response to input of the training input data included in the training data and the value of the training data included in the training data.

Note that the above-mentioned closeness score does not have to be binary data taking on a value of either 0 or 1. For example, the above-mentioned closeness score may be a real number (e.g., a real number between 0 and 10 inclusive) that increases the closer the relationship between the pair, or a multi-level integer value (e.g., an integer between 1 and 10 inclusive).

Furthermore, the learning method for machine learning models is not limited to weakly supervised learning. As a specific example, consider a pair of siblings. In this case, input data associated with the pair is input to a trained machine learning model that corresponds to the sibling relationship. For example, if the address data values for the pair are the same, one member of the pair has sent 50 gifts to the other, and the pair has made 1,200 phone calls to date, learning may be performed to output output data with a value of 1. Alternatively, for example, if the address data values for the pair are different, one member of the pair has sent 2 gifts to the other, and the pair has made 30 phone calls to date, learning may be performed to output output data with a value of 0. The criteria (e.g., threshold) for determining whether the value of the output data corresponding to the closeness score is 1 or 0 may differ depending on the machine learning model.

The effect estimation unit 25 estimates the effect that a predetermined operation on a user to prompt the user to perform a predetermined action will have on whether or not the user will perform the action. In this embodiment, the predetermined action is payment of a credit card amount that is overdue. Note that the specific payment method is not limited, and may be a transfer to a designated account, payment at a designated counter, or the like. Also, in this embodiment, the predetermined operation is a call to remind the user to pay the overdue credit card amount. Note that the call to the user may be an automated call using a recording or a machine-generated voice, or a call in which an operator (human) speaks with the user. In this embodiment, the effect estimation unit 25 estimates the effect of the operation using a machine learning model (hereinafter referred to as the "effect estimation model") that outputs a causality score that indicates the effect of the operation on the user in response to input of one or more user attributes related to the target user.

However, some users may lack user attribute data for estimating the effect of an operation, which poses a problem in that it is difficult to estimate the effect of an operation for such users based solely on their user attributes. For this reason, the effect estimation unit 25 of the present disclosure estimates the effect of an operation on whether or not a target user will perform an action for a user for whom there is insufficient user attribute data for estimating the effect of an operation, by using the effect of the operation on whether or not other users (reference users) related to the user (target user) will perform the action.

In the present disclosure, the effect (target effect) of an operation on whether a target user performs an action is referred to as the "target causal score," and the effect (reference effect) of an operation on whether a reference user performs an action is referred to as the "reference causal score." In this embodiment, the effect estimation unit 25 estimates the target causal score, which is the effect on the target user, based on data including the reference causal score, which is the effect estimated for the reference user. Note that in the present disclosure, the target causal score or the reference causal score may be simply referred to as the causal score, or may be simply referred to as the effect (the effect of the operation).

This is because, if there is a high probability that the reference user will receive an operation, such as a call, on behalf of the target user, the effect of the operation on the target user may be affected by the effect of the operation on the reference user. For this reason, the reference user may be, for example, another user who has a predetermined relationship with the target user and lives with the target user. Furthermore, for example, if the target user is an adult, it can be assumed that users corresponding to the target user's parents tend to be at home, so the target user's causal score (target causal score) may be further estimated or determined based on the causal score (reference causal score) of the target user's parent or co-resident user. However, the reference user may be any user whose effect on the target user may be affected by the effect of the operation on the reference user, and is not limited to co-resident users.

Specifically, for example, the effect estimation unit 25 can estimate the reference causal score as the target causal score. That is, the effect estimation unit 25 may treat the causal score of a reference user identified as a reference user having a relationship with the target user as the causal score of the target user. In addition, in this case, if the reference causal score is a causal score that shows a higher effect than the target causal score, the effect estimation unit 25 may estimate the reference causal score as the target causal score.

The effect estimation unit 25 may also estimate the target causal score based on data including multiple reference causal scores estimated for multiple reference users. For example, the effect estimation unit 25 can estimate the target causal score based on a statistical indicator (e.g., mean, weighted average, median, mode, etc.) based on multiple reference causal scores estimated for multiple reference users. In this case, the effect estimation unit 25 may also estimate, as the target causal score, the average (weighted average) of values obtained by weighting each of the multiple reference causal scores estimated for the multiple reference users according to the relationship strength (closeness score) indicating the closeness between the target user and the reference user. Note that the closeness score may be used as the weight when calculating the weighted average, or a weight converted to another scale based on the closeness score may be used.

The risk estimation unit 26 estimates risk based on the probability that a user will not perform the predetermined action described above. There are no limitations on how risk is expressed, and various indicators may be used. For example, risk can be expressed using the probability that a debt will not be collected and will result in default. In this embodiment, the risk estimation unit 26 estimates risk based on the probability that a user will not perform an action using a machine learning model (hereinafter referred to as the "risk estimation model") that outputs a risk indicator indicating the probability that a debt will not be collected and will result in default for the target user, in response to input of one or more user attributes related to the target user. However, risk may also be estimated according to, for example, predetermined rules without using a machine learning model. For example, risk may be obtained by storing a value corresponding to each user attribute or combination of user attributes in advance and reading this value. Furthermore, indicators other than the probability of default may be used as indicators of risk. For example, risk may be classified (ranked), and the class (rank) of the risk may be used as an indicator.

The machine learning unit 27 generates and/or updates an effect estimation model used for effect estimation by the effect estimation unit 25 and a risk estimation model used for risk estimation by the risk estimation unit 26. The effect estimation model is a machine learning model that, when one or more user attribute data related to a target user is input, outputs a causal score indicating the effectiveness of an operation on the user. Furthermore, the risk estimation model is a machine learning model that, when one or more user attribute data related to a target user is input, outputs a risk index indicating the degree of risk based on the likelihood that the user will not perform an action. The user attributes input into these machine learning models may include, for example, demographic attributes, behavioral attributes, or psychographic attributes. Here, demographic attributes include, for example, the user's gender, family structure, age, etc.; behavioral attributes include, for example, whether or not they have used cash advances, whether or not they have used revolving credit, deposit and withdrawal history for a specific account, and commercial transaction history for some products including gambling or lotteries (which may include online transaction history in online marketplaces, etc.); and psychographic attributes include, for example, preferences for gambling or lotteries. However, usable user attributes are not limited to the examples given in this embodiment. For example, "time required for an operation (such as a phone call)" and "amount used by credit card" may also be used as attributes.

When generating and/or updating the effect estimation model, the machine learning unit 27 creates the effect estimation model based on training data (machine learning data) in which, for each user attribute, a score based on a statistical quantity related to the action execution rate (debt collection rate) of users who received an operation among multiple users with a specified attribute and a statistical quantity related to the action execution rate of users who did not receive an operation is defined as a causal score indicating the effect of the operation on users with the specified attribute. In this embodiment, for example, if the operation involves making a call to a user, a causal score is calculated based on the difference between each statistical quantity using the formula "(debt collection rate when the user made a call) - (debt collection rate when the user did not make a call)." The calculated causal score is combined with the corresponding user attribute data and input to the machine learning unit 27 as training data. Note that this embodiment describes an example in which an average value is used as the statistical quantity. However, statistical indicators such as the mode or median may also be used as the statistical quantity. Here, the statistical quantity related to the action execution rate may be based on each user's past debt collection rate within a specified period (e.g., a specified month). Furthermore, in this embodiment, a causal score is calculated based on the difference between each statistical amount, but if no statistically significant difference is found in the difference, the causal score may be set to zero or approximately zero. Existing statistical methods may be used to determine whether or not a significant difference exists. For example, standard error and confidence intervals may be considered for a set of users' average response rates for each month, and the statistical significance of changes in response rates due to the presence or absence of calls may be considered. In this way, it is possible to calculate a causal score that takes into account variations in average response rates among users within the same group.

When creating training data, a user who has been called once cannot become a user who has not been called thereafter. Therefore, for a single user, it is possible to obtain either the recovery rate when that user is called or the recovery rate when that user is not called. Therefore, training data on the effectiveness of calls is created for each user group that shares a common attribute. That is, a causal score indicating the effectiveness of calls on a user group consisting of users with a common attribute is obtained, for example, by dividing multiple users with the common attribute into a first sub-user group that is called and a second sub-user group that is not called, calculating the average debt recovery rate from the first sub-user group that is called and the average debt recovery rate from the second sub-user group that is not called, and then calculating the difference between these average recovery rates using the formula described above. For example, if the average debt recovery rate from the first sub-user group that is called is 80% and the average debt recovery rate from the second sub-user group that is not called is 70%, the causal score indicating the effectiveness of calls on that user group is "10."

A machine learning model generation framework that can be used to implement the technology disclosed herein is based on an ensemble learning algorithm, for example. For example, a machine learning framework (e.g., LightGBM) based on a gradient boosting decision tree (GBDT) may be used. In other words, the framework may be a machine learning framework based on a decision tree model that inherits the error between the correct answer and the predicted value between previous and subsequent weak learners (weak classifiers). The predicted value here refers to, for example, a causal score or a predicted value of a risk index. In addition to LightGBM, the framework may also employ boosting methods such as XGBoost and CatBoost. A framework using a decision tree allows for the generation of a machine learning model with relatively high performance with less parameter adjustment effort than a framework using a neural network. However, the machine learning model generation framework that can be used to implement the technology disclosed herein is not limited to the example in this embodiment. For example, instead of a gradient boosting decision tree, a different learner such as a random forest may be used as the learner, or a learner that is not a so-called weak learner, such as a neural network, may be used. Furthermore, especially when a learner that is not a so-called weak learner, such as a neural network, is used, ensemble learning does not need to be used.

Figure 13 is a simplified diagram illustrating the concept of a decision tree in the machine learning model employed in this embodiment. When using a gradient boosting machine learning framework based on a decision tree algorithm, the branching conditions for each node in the decision tree are optimized. Specifically, in this gradient boosting machine learning framework based on a decision tree algorithm, a causal score indicating the effect of an operation is calculated for each user group having attributes indicated by two child nodes branching from a single parent node. The branching condition for the parent node is optimized so that the difference between these causal scores is large (e.g., so that the difference is maximized or exceeds a predetermined threshold), i.e., so that the two child nodes branch cleanly. For example, if the attribute indicated as the branching condition for a node is age, the age set as the branching threshold may be changed, or the branching condition may be changed to an attribute other than age. In this way, recursively optimizing the branching conditions for all nodes in the decision tree can improve the accuracy of estimating the effect of an operation.

When generating and/or updating a risk estimation model, the machine learning unit 27 creates a machine learning model based on training data in which a statistical quantity (in this embodiment, an average value; however, a statistical indicator such as the mode or median may also be used) related to the default incidence rate of multiple users having a specified attribute is defined as a risk index indicating the degree of risk for users having that attribute. The calculated risk index is combined with the corresponding user attribute data and input to the machine learning unit 27 as training data. Furthermore, when generating or updating a risk estimation model, there are no limitations on the machine learning model generation framework that can be used, but a gradient boosting machine learning framework based on a decision tree algorithm may be used, just as in the generation and/or updating of the effect estimation model described above.

The condition output unit 28 determines and outputs conditions (hereinafter referred to as "operation conditions") for operations to be performed on users at the operations center based on the estimated effects and risks (in this embodiment, the causal score and risk index). In this embodiment, the operation conditions include at least one of the following: whether the operation needs to be performed, the number of times the operation is to be performed, the order in which the operations are to be performed, the method of contacting the user during the operation, and the content of the contact with the user during the operation. Examples of contact methods include making a phone call or sending a message, and examples of the content of the contact include the content to be conveyed to the user during the call and the content of the message.

Figure 14 is a diagram showing the relationship between estimated effects and risks and operation conditions in this embodiment. Essentially, the condition output unit 28 outputs operation conditions such that higher priority is given to operations for users with higher estimated effects and at least some of the operations for users with higher estimated risks. Furthermore, the condition output unit 28 outputs operation conditions such that lower priority is given to operations for users with lower estimated effects and at least some of the operations for users with lower estimated risks.

Here, priority is a measure set for a user or an operation for that user, indicating the degree to which the operation is to be performed with priority. The higher the priority, the more likely or frequent the user will receive the operation, and the lower the priority, the less likely or frequent the user will receive the operation. Specifically, the condition output unit 28 outputs operation conditions, such as setting the operation as required to be performed, increasing the number of times the operation is performed, moving the order of execution of the operation earlier, or making the method or content of contact with the user for the operation more cost-effective or effective, as conditions for assigning a higher priority. Furthermore, the condition output unit 28 outputs operation conditions, such as setting the operation as not required to be performed, reducing the number of times the operation is performed, moving the order of execution of the operation later, or making the method or content of contact with the user for the operation less cost-effective or effective, as conditions for assigning a lower priority.

In this embodiment, the condition output unit 28 compares the estimated causal score and risk index with preset thresholds for the causal score and risk index, respectively, and determines and outputs operation conditions based on the comparison results. Specifically, in the example shown in FIG. 14, a threshold C1 and a threshold C2 greater than C1 are set for the causal score, and a threshold R1 and a threshold R2 greater than R1 are set for the risk index. Furthermore, for the risk index, a threshold R3 is set to determine whether or not a user is subject to operation condition setting based on the causal score, and a threshold R4 is set to determine users for whom high-priority operation conditions are set regardless of the causal score. Note that in this embodiment, an example is described in which the same value is used for thresholds R1 and R3, and the same value is used for thresholds R2 and R4 (see FIG. 14), but different values may be used for these thresholds.

Here, a case where an operation condition with a high priority is output will be described.
Case (1): The condition output unit 28 determines and outputs high-priority operation conditions for at least some users whose causal scores are equal to or greater than the threshold C2 because the operation is highly effective.
Case (2): Furthermore, the condition output unit 28 determines and outputs high-priority operation conditions for at least some users whose risk index is equal to or greater than the threshold R2, since the risk is high.
Case (3): In particular, the condition output unit 28 may determine and output the highest priority operation condition for a user whose causal score is equal to or greater than the threshold C2 and whose risk index is equal to or greater than the threshold R2 (see the area UR indicated by the dashed line in Figure 14).
Case (4): However, for users whose estimated risk index is lower than the threshold R3, the condition output unit 28 does not need to output operation conditions that are given high priority because the user is unlikely to become a default in the first place. To enhance the cost reduction effect, it is preferable to determine and output low-priority or medium-priority operation conditions for users whose risk index is lower than the threshold R3, regardless of the causal score (the region UL indicated by the dashed line in FIG. 14 has a causal score of C2 or higher but a risk index below R3, so does not become a high-priority operation condition). Therefore, the effect estimation unit 25 may estimate the effect of the operation for users whose risk index is estimated to be higher than the threshold R3, and may not estimate the effect of the operation for users whose risk index is estimated to be lower than the threshold R3 (omit the estimation process).

A case where an operation condition with a low priority is output will also be described.
Case (5): For at least some users whose causal scores are less than the threshold C1, the condition output unit 28 determines and outputs low-priority operation conditions because the operation effects are low.
Case (6): Furthermore, for at least some of the users whose risk index is less than the threshold R1, the condition output unit 28 determines and outputs low-priority operation conditions because the risk is low.
Case (7): In particular, the condition output unit 28 may determine and output the lowest priority operation condition for a user whose causal score is less than the threshold C1 and whose risk index is less than the threshold R1 (see area LL indicated by the dashed line in Figure 14).
Case (8): However, for a user whose estimated risk index is equal to or greater than the threshold R4, the condition output unit 28 does not need to output an operation condition that is given a low priority because the user is likely to be defaulted in the first place. Rather, for a user whose risk index is equal to or greater than the threshold R4, it is preferable to determine and output an operation condition with a high priority regardless of the causal score (the region LR indicated by the dashed line in FIG. 14 has a causal score less than C1, but the risk index is equal to or greater than R4, so it does not become an operation condition with a low priority).

As a result, in the example shown in Figure 14, the operation priority is increased for high-risk users and medium-risk and high-effectiveness users, and decreased for low-risk users and medium-risk and low-effectiveness users. Note that in the example described above, areas are divided based on thresholds, and common operation conditions are determined for users belonging to one area, but different operation conditions may be set for each user or for each user within one area. For example, even within the same area, operation conditions may be varied to create a gradation depending on the level of the causal score and/or risk index.

The amount of operations performed by the operations center management system 3 (total number of operations and number of users subject to operations) can be changed by adjusting the thresholds described above. For example, the amount of operations can be increased by lowering at least one of thresholds C1, C2, R1, and R2, and the amount of operations can be decreased by raising at least one of thresholds C1, C2, R1, and R2.

<Processing flow>
Next, a flow of processing executed by the information processing system according to this embodiment will be described. Note that the specific content and processing order of the processing described below are an example for implementing the present disclosure. The specific content and processing order may be selected as appropriate depending on the embodiment of the present disclosure.

Figure 15 is a flowchart showing the flow of machine learning processing according to this embodiment. The processing shown in this flowchart is executed at a timing specified by the administrator of the operation center management system 3.

In steps S101 and S102, an effect estimation model is generated and/or updated. The machine learning unit 27 calculates a causal score for each of multiple user attributes based on user attribute data, operation history data, and credit card payment history data previously accumulated in the operation center management system 3 or credit card management system 5, and creates training data including combinations of user attributes and causal scores (step S101). Here, the operation history data includes data that can be used to determine, for each user, whether an operation has been performed on that user, and the payment history data includes data that can be used to determine, for each user, whether that user has paid their credit card usage amount (whether or not there is default). The machine learning unit 27 then inputs the created training data into the machine learning model and generates or updates a machine learning model (effect estimation model) used by the effect estimation unit 25 to estimate effects (step S102). Processing then proceeds to step S103.

In steps S103 and S104, a risk estimation model is generated and/or updated. The machine learning unit 27 calculates a risk index for each of multiple user attributes based on user attribute data, operation history data, and credit card payment history data previously accumulated in the operation center management system 3 or credit card management system 5, and creates training data including combinations of user attributes and risk indexes (step S103). The machine learning unit 27 then inputs the created training data into the machine learning model, and generates or updates a machine learning model (risk estimation model) used for risk estimation by the risk estimation unit 26 (step S104). Thereafter, the processing shown in this flowchart ends.

Figure 16 is a flowchart showing the flow of the operation condition output process according to this embodiment. The process shown in this flowchart is executed at a preset time each month. More specifically, the execution timing of the process is set to be after the specified date for payment of the credit card usage amount and before the scheduled date for execution of the operation for the unpaid user. Note that here, it is assumed that graph data has already been generated for multiple users, including the target user, and that each machine learning model has already been trained.

In step S201, a risk is estimated based on the probability that a user will not perform an action. For each of multiple users, the risk estimation unit 26 inputs one or more user attribute data related to the target user into the risk estimation model generated and/or updated in step S104, and obtains a risk index corresponding to the target user as output. Processing then proceeds to step S202.

In step S202, it is determined whether a causal score can be obtained based on the attribute data. The information processing device 1 determines whether user attribute data for the target user can be obtained from the system to the extent that a causal score corresponding to the target user can be obtained from the effect estimation model. This determination may be made simply based on the number of attribute data, or based on the presence or absence of one or more predetermined types of attribute data that are important for obtaining a causal score. Furthermore, the one or more predetermined types of attribute data that are important for obtaining a causal score may be determined based on the accuracy of past effect estimation results using the effect estimation model. If it is determined that sufficient attribute data for estimating a causal score can be obtained from the system, processing proceeds to step S203.

In step S203, the effect of the operation on the target user is estimated. For each of multiple users, the effect estimation unit 25 inputs one or more pieces of user attribute data related to the target user into the effect estimation model generated and/or updated in step S102, and obtains, as output, a causal score corresponding to the target user. Processing then proceeds to step S207.

On the other hand, if it is determined in step S202 that sufficient attribute data cannot be obtained from the system to estimate the causal score, processing proceeds to step S204.

In steps S204 and S205, a reference user is identified, and the effect of the operation on the reference user is estimated. The reference user identification unit 22 references the graph data and identifies one or more other users corresponding to node data 50 that are connected to the node data 50 corresponding to the target user by an explicit or implicit link as reference users. Here, for example, a user who lives in the same household as the target user and has a parent-child, marital, or friendship relationship is identified as the reference user (step S204). Once the reference user is identified, the effect estimation unit 25 inputs one or more user attribute data related to the reference user into the effect estimation model generated and/or updated in step S102, and obtains, as output, a causal score corresponding to the reference user (reference causal score) (step S205). Processing then proceeds to step S206.

In step S206, the effect of the operation on the target user is estimated based on the effect of the operation on the reference user. The effect estimation unit 25 acquires a causal score corresponding to the target user based on data including the reference causal score acquired for the target user in step S205. As explained above, the effect estimation unit 25 may estimate the reference causal score as the target causal score, or may estimate the target causal score based on data including multiple reference causal scores estimated for multiple reference users. Processing then proceeds to step S207.

In step S207, operation conditions are determined and output. The condition output unit 28 determines operation conditions based on the causal score of the target user acquired in step S203 or step S206 and the risk index of the target user acquired in step S201, and outputs them to the operation center management system 3. In this embodiment, the condition output unit 28 identifies and outputs operation conditions that have been pre-mapped to the causal score and risk index. However, the method of determining operation conditions is not limited to the example given in this embodiment. For example, the operation conditions may include values calculated by inputting the causal score and risk index into a predetermined function. The processing shown in this flowchart then ends.

When the operation conditions are output, the operation center management system 3 manages the operation for the target user in accordance with the operation conditions, and the operation terminal executes the operation in accordance with the instructions output by the operation center management system 3.

<Effects>
According to this embodiment, by setting priorities of operation conditions according to the effectiveness and risk of operations for each user and suppressing operations for users with low effectiveness or low risk, it is possible to suppress operation costs without reducing the debt collection rate. In other words, according to the present disclosure, it is possible to suppress operation costs without reducing the effectiveness of operations that encourage users to take predetermined actions. Furthermore, according to this embodiment, by increasing operations for users with high effectiveness or high risk, it is expected that the debt collection rate will be increased while suppressing costs.

<Variation (1) of Operation Condition Output Processing>
In the embodiment described above, the flow of the operation condition output process has been described briefly with reference to FIG. 16, but more specifically, the operation condition output process may be performed as follows.

Figure 17 is a flowchart showing the flow of the operation condition output process when the determination methods for cases (1) to (4) described with reference to Figure 14 are adopted in this embodiment. According to the example shown in Figure 17, if the risk index calculated by the risk estimation unit 26 (step S301) for a user with a certain attribute is less than threshold R3 (NO in step S302), the calculation of the causal score by the effect estimation unit 25 is omitted, and low (or medium) priority operation conditions are determined and output (step S303).

If the risk index calculated for the user is equal to or greater than threshold R3 (YES in step S302), a causal score is calculated (step S304). If the causal score is equal to or greater than threshold C2 and the risk index is equal to or greater than threshold R2 (YES in step S305), the highest priority operation condition is determined and output (step S306). If the causal score is equal to or greater than threshold C2 or the risk index is equal to or greater than threshold R2 (YES in step S307), a high priority operation condition is determined and output (step S308). Note that if the causal score is less than threshold C2 and less than threshold R2, a medium priority operation condition is determined and output (step S309).

<Variation (2) of Operation Condition Output Processing>
Fig. 18 is a flowchart showing the flow of the operation condition output process when the determination methods of cases (5) to (8) described with reference to Fig. 14 are adopted in this embodiment. According to the example shown in Fig. 18, when the risk index calculated by the risk estimation unit 26 (step S401) for a user having a certain attribute is equal to or greater than the threshold value R4 (NO in step S402), the calculation of the causal score by the effect estimation unit 25 is omitted, and a high-priority (or medium-priority) operation condition is determined and output (step S403).

If the risk index calculated for the user is less than threshold R4 (YES in step S402), a causal score is calculated (step S404). If the causal score is less than threshold C1 and the risk index is less than threshold R1 (YES in step S405), the lowest priority operation condition is determined and output (step S406). If the causal score is less than threshold C1 or the risk index is less than threshold R1 (YES in step S407), a low priority operation condition is determined and output (step S408). Note that if the causal score is equal to or greater than threshold C1 and threshold R1 (NO in step S407), a medium priority operation condition is determined and output (step S409).

<Variation (3) of Operation Condition Output Processing>
In the above-described embodiment, an example was described in which two axes, a causal score indicating the effectiveness of the operation and a risk index indicating the risk, are used as evaluation axes for determining operation conditions, but in the technology disclosed herein, the evaluation axes for determining operation conditions only need to include at least the effectiveness of the operation, and other evaluation axes may be adopted, or three or more evaluation axes may be adopted. For example, (1) a third index other than the risk index may be adopted as an index to be combined with the causal score, (2) a third index may be adopted in addition to the causal score and risk index, or (3) three axes, a causal score for calling, a causal score for sending a message, and a risk index, may be adopted.

For example, the operation conditions may be set by referring to the target causal score and the reference causal score (without referring to the risk index). In this case, the condition output unit 28 outputs operation conditions that give higher priority to operations for users with higher target causal scores and at least some of the operations for users with higher reference causal scores. Also, the condition output unit 28 outputs operation conditions that give lower priority to operations for users with lower target causal scores and at least some of the operations for users with lower reference causal scores.

<Variation (4) of Operation Condition Output Processing>
As described above, in the technology disclosed herein, three or more evaluation axes may be used as evaluation axes for determining operation conditions. In this variation, the condition output unit 28 differs from the condition output unit 28 according to the embodiment described above in that, as the effect (causal score), the condition output unit 28 also refers to the effect estimated for the reference user (reference causal score) in addition to the effect estimated for the target user (target causal score) and risk index. That is, the condition output unit 28 according to this variation outputs operation conditions for the target user based on data including the target causal score and the reference causal score.

In this variation, as in the embodiment described above, the condition output unit 28 outputs operation conditions that, in principle, give higher priority to operations for users with higher estimated effects and at least some of the operations for users with higher estimated risks, and outputs operation conditions that give lower priority to operations for users with lower estimated effects and at least some of the operations for users with lower estimated risks.

In this variation, the condition output unit 28 basically outputs operation conditions such that higher priority is given to at least some of the operations for users with higher target causal scores, operations for users with higher reference causal scores, and operations for users with higher estimated risks. The condition output unit 28 also outputs operation conditions such that lower priority is given to at least some of the operations for users with lower target causal scores, operations for users with lower reference causal scores, and operations for users with lower estimated risks.

In addition, in this variation, the condition output unit 28 outputs a condition under which at least some of the operations on a target user for which the estimated effect (reference causal score) for the reference user is higher are given a higher priority than would be determined if the reference causal score were not referenced. In this case, the condition output unit 28 may compare the reference causal score with the target causal score to determine whether the reference causal score is higher, and if the reference causal score is higher than the target causal score, output a condition under which at least some of the operations on the target user are given a higher priority than would be determined if the reference causal score were not referenced.

Furthermore, in this variation, instead of or in addition to the above-described method of setting operation conditions, the condition output unit 28 may output a condition such that at least some of the operations for a target user for which the estimated effect (reference causal score) for the reference user is lower are given a lower priority than would be determined if the reference causal score were not referenced. In this case, the condition output unit 28 may compare the reference causal score with the target causal score to determine whether the reference causal score is lower, and if the reference causal score is lower than the target causal score, output a condition such that at least some of the operations for the target user are given a lower priority than would be determined if the reference causal score were not referenced.

<Variation (5) of Operation Condition Output Processing>
The setting of the operation condition that refers to the reference causal score in addition to the target causal score, as explained in variation (4) of the operation condition output process, may be implemented more specifically in the manner described below, for example. However, this variation does not limit the specific embodiment of variation (4).

Figure 19 is a diagram showing the relationship between the target causal score, reference causal score, and risk index estimated in this variation and the operation conditions. In this variation, the condition output unit 28 compares the estimated target causal score, reference causal score, and risk index with preset thresholds for the target causal score, reference causal score, and risk index, respectively, and determines and outputs operation conditions based on the comparison results. In the example shown in Figure 19, as in the embodiment described above, thresholds C1, C2, R1, R2, R3, and R4 are set, and the roles of these thresholds are the same as in the embodiment described above.

As a difference from the embodiment described above, in this variation, the condition output unit 28 outputs conditions that assign higher priority to at least some of the operations for a user with a higher reference causal score (e.g., a predetermined threshold RC2 or higher) when the target causal score is within a first range that is neither the high-effect range (above threshold C2) nor the low-effect range (below threshold C1) (i.e., within the medium-effect range that is equal to or greater than threshold C1 but less than C2) and the estimated risk index is within a second range that is neither the high-risk range (above threshold R2) nor the low-risk range (below threshold R1) (i.e., within the medium-risk range that is equal to or greater than threshold R1 but less than R2) (see range A1 enclosed by a dashed rectangle in the figure). In other words, in this variation, even if the target user's target causal score is estimated to be in the medium-effect range, the condition output unit 28 outputs operation conditions with higher priority for at least some of the operations for the target user if the reference user's causal score is higher and the target user's risk index is estimated to be medium or higher.

Furthermore, in this variation, if the target causal score is within a third range (below threshold C1) for determining that the target causal score is in the low-effectiveness range and the estimated risk index is within a second range that is neither the high-risk range (above threshold R2) nor the low-risk range (below threshold R1) (i.e., within the medium-risk range between thresholds R1 and R2) (see range A2 enclosed by the dashed rectangle in the figure), the condition output unit 28 outputs conditions that assign higher priority to at least some of the operations for a user with a higher reference causal score (for example, above a preset threshold RC2). In other words, in this variation, even if the target user's target causal score is estimated to be in the low-effectiveness range, the condition output unit 28 outputs operation conditions with higher priority for at least some of the operations for the target user if the reference user's causal score is higher and the target user's risk index is estimated to be medium or higher.

<Variation (6) of Operation Condition Output Processing>
Instead of or in addition to the method for setting operation conditions described in variation (5) of the operation condition output process, the following method for setting operation conditions may be used. However, this variation does not limit the specific embodiment of variation (4).

Figure 20 is a diagram showing the relationship between the target causal score, reference causal score, and risk index estimated in this variation and the operation conditions. In this variation, the condition output unit 28 compares the estimated target causal score, reference causal score, and risk index with preset thresholds for the target causal score, reference causal score, and risk index, respectively, and determines and outputs operation conditions based on the comparison results. In the example shown in Figure 20, as in the embodiment described above, thresholds C1, C2, R1, R2, R3, and R4 are set, and the roles of these thresholds are the same as in the embodiment described above.

In this variation, when the target causal score is within a first range that is neither the high-effect range (above threshold C2) nor the low-effect range (below threshold C1) (i.e., within the medium-effect range that is equal to or greater than threshold C1 but less than C2), and the estimated risk index is within a second range that is neither the high-risk range (above threshold R2) nor the low-risk range (below threshold R1) (i.e., within the medium-risk range that is equal to or greater than threshold R1 but less than R2) (see range A1 enclosed by a dashed rectangle in the figure), the condition output unit 28 outputs a condition that assigns a lower (or medium) priority to at least some of the operations for a user with a lower reference causal score (e.g., below a predetermined threshold RC1). In other words, in this variation, even if the target user's target causal score is estimated to be in the medium-effect range, the condition output unit 28 outputs operation conditions with a lower (or medium) priority for at least some of the operations for the target user if the reference user's causal score is lower and the target user's risk index is estimated to be medium or higher.

Furthermore, in this variation, if the target causal score is within a third range (below threshold C1) for determining that the target causal score is in the low-effectiveness range, and the estimated risk index is within a second range that is neither the high-risk range (above threshold R2) nor the low-risk range (below threshold R1) (i.e., within the medium-risk range between thresholds R1 and R2) (see range A2 enclosed by the dashed rectangle in the figure), the condition output unit 28 outputs a condition that assigns a lower (or medium) priority to at least some of the operations for a user with a lower reference causal score (for example, below a predetermined threshold RC1). In other words, in this variation, if the target user's target causal score is estimated to be in the low-effectiveness range, the reference user's causal score is lower, and the target user's risk index is estimated to be medium or higher, the condition output unit 28 outputs an operation condition with a lower (or medium) priority for at least some of the operations for the target user.

<Variation (7) of Operation Condition Output Processing>
21 and 22 are flowcharts showing the flow of the operation condition output process when the relationships between the target causal score, reference causal score, and risk index and the operation conditions described with reference to FIGS. 19 and 20 are adopted in this embodiment.

According to the example shown in this flowchart, if the risk index calculated by the risk estimation unit 26 (step S501) for a user with a certain attribute is less than threshold R3 (NO in step S501), the calculation of the causal score by the effect estimation unit 25 is omitted, and an operation condition with a low priority is determined and output (step S503). Also, if the risk index calculated for the user is equal to or greater than threshold R4 (NO in step S504), the calculation of the causal score by the effect estimation unit 25 is omitted, and an operation condition with a high priority is determined and output (step S505). Once the operation condition has been determined and output, the processing shown in this flowchart ends.

If the calculated risk index for the user is greater than or equal to threshold R3 and less than threshold R4 (i.e., moderate risk), a target causal score and a reference causal score are calculated (step S506). If the calculated reference causal score is determined to be in the low-effectiveness range (e.g., less than threshold RC1) (YES in step S507), low-priority operation conditions are determined and output (step S508). If the calculated reference causal score is determined to be in the high-effectiveness range (e.g., greater than or equal to threshold RC2) (YES in step S509), high-priority operation conditions are determined and output (step S510).

If it is determined that the reference causal score calculated for the user is neither in the low effect band nor the high effect band (in other words, it is in the medium effect band), operation conditions according to the target causal score are determined and output in the same manner as described with reference to FIG. 14 (step S511). That is, if the target causal score is less than threshold C1, low priority operation conditions are determined and output; if the target causal score is equal to or greater than threshold C1 and less than threshold C2, medium priority operation conditions are determined and output; and if the target causal score is equal to or greater than threshold C2, high priority operation conditions are determined and output.

In the example described above, an example was described in which operation conditions were determined by dividing the area based on a threshold value, but as explained above in the explanation of Figure 14, different operation conditions may be set for each user or for each user even within the same area. For example, even within the same area, operation conditions may be varied to create a gradation depending on the level of the causal score and/or risk index.

<Other variations>
In the above-described embodiment, an example was described in which the operation to the user was a call, but the type of operation to the user is not limited to a call. For example, a message may be sent as a type of operation to the user. Note that the means for sending the message is not limited here, and an email system, a short message service (SMS), a message sending and receiving service of a social networking service (SNS), or the like may be used.

In addition, while the above-described embodiment describes an example in which operation conditions are determined based on the estimated effect of one type of operation (a call), operation conditions may also be determined based on the estimated effect of each of multiple types of operations (e.g., a call and sending a message). In this case, the effect estimation unit 25 estimates a first effect that a first operation (e.g., a call) to a user to prompt the user to perform a predetermined action will have on whether the user will perform the action, and a second effect that a second operation (e.g., sending a message) to a user to prompt the user to perform a predetermined action will have on whether the user will perform the action, and the condition output unit 28 outputs operation conditions for the user based on the estimated first and second effects.

In this case, a machine learning model for estimating the effectiveness of an operation is also generated and updated for each type of operation. For example, if the first operation is making a call and the second operation is sending a message, a machine learning model for estimating the effectiveness of making a call and a machine learning model for estimating the effectiveness of sending a message may be generated and updated.

Furthermore, when operation conditions are determined based on the estimated effects of each of multiple types of operations, the type of operation that is most effective for the user may be selected from the multiple types of operations. In this case, the condition output unit 28 outputs operation conditions, including whether to perform the first operation or the second operation on the user, based on the estimated first and second effects. More specifically, by comparing the first effect (causal score related to the first operation) and the second effect (causal score related to the second operation) obtained for the target user, the type of operation with the higher causal score can be selected as the type of operation that is most effective for the user.

In addition, in the above-described embodiment, an example was described in which the difference between the action execution rate of a group of sub-users who received an operation and the action execution rate of a group of sub-users who did not receive an operation was used as the causal score. However, the causal score may be calculated using other methods. For example, when generating and/or updating an effect estimation model, the machine learning unit 27 may create a machine learning model based on training data in which, for each user attribute, a score based on statistics relating to the action execution rate of users who, among multiple users having a specified attribute, performed a specified reaction to an operation and statistics relating to the action execution rate of users who, among multiple users, did not perform the specified reaction is defined as the causal score for users having the specified attribute. For example, in this variation, the causal score is calculated using the formula "(the debt collection rate when the user performs the specified reaction) - (the debt collection rate when the user does not perform the specified reaction)." In this case, the condition output unit 28 may output a condition related to the operation based on the presence or absence of a user reaction, the content of that reaction, etc., and the causal score corresponding to the reaction. In this case, the priority of the operation condition output by the condition output unit 28 may be adjusted using the relationship between the causal score and the priority described with reference to FIG. 14.

Here, a predetermined reaction is, for example, a user's response to a call by pressing the dial, a conversation with an operator when the user calls back, a reply to a message, marking a message as read, etc. Furthermore, the content of the reaction may also be considered to be a positive response to payment, or whether or not there was a response regarding the due date of payment. If the user's reaction is a voice reaction, it is also possible to determine the user's emotions, etc. based on the user's voice and determine whether the reaction was positive. Furthermore, the priority of the next operation conditions may be adjusted based on the user's emotions, etc. determined based on the voice.

The priority of operation conditions may also be adjusted based on factors other than those mentioned above. For example, for users whose credit card payment settings are revolving payments or installment payments, the priority of operation conditions may be higher than for users who make lump-sum payments, and for users whose credit card usage includes cash advances, the priority of operation conditions may be higher than for users whose credit card usage does not include cash advances. The priority of operation conditions can also be adjusted based on the target user's transaction data other than credit card data (for example, balance data for the account from which credit card usage amounts are debited, transaction history data at affiliated banks, etc.). Here, the condition output unit 28 may, for example, adjust the various thresholds corresponding to the various scores shown in FIG. 14 based on the credit card usage conditions.

Furthermore, in the embodiment described above, an example of an information processing device including a graph data generation unit 21, a reference user identification unit 22, a relationship identification unit 23, a relationship strength determination unit 24, an effect estimation unit 25, a risk estimation unit 26, a machine learning unit 27, and a condition output unit 28 was described, but some of these functional units may be omitted to the extent that the invention disclosed herein can be implemented.

For example, in the embodiment described above, the relationship strength (closeness score) between users was generated and referenced when identifying the reference user or calculating the target causal score, but the generation and reference of the closeness score may be omitted when identifying the reference user or calculating the target causal score. In this case, the relationship strength determination unit 24 may be omitted from the functional units of the information processing device 1 described with reference to FIG. 2.

1. Information processing device

Claims (21)

an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a reference user specifying means for specifying at least one reference user who is related to the target user in accordance with a relationship strength indicating the closeness of the users, the relationship strength being determined based on an index indicating the strength of the relationship between the users in accordance with a judgment criterion corresponding to the relationship between the users;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user, as a target effect, which is the effect for the target user.
Information processing device. an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a reference user specifying means for specifying at least one reference user who is related to the target user, the reference user specifying means specifying another user who has attribute data in common with the target user as the at least one reference user, or specifying the reference user based on graph data showing the relationship between users;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user, as a target effect, which is the effect for the target user.
Information processing device. an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a reference user specifying means for specifying a plurality of reference users who are related to the target user, the reference user specifying means specifying other users who have attribute data in common with the target user as the plurality of reference users, or specifying the reference users based on graph data showing the relationships between users;
a relationship specifying means for specifying a relationship between users based on a result of clustering based on values associated with the relationship between users;
a relationship strength determination means for determining a relationship strength indicating the closeness of the users based on an index indicating the strength of the relationship between the users in accordance with a determination criterion corresponding to the relationship between the users,
The effect estimation means estimates a target effect, which is the effect for the target user, by averaging values obtained by weighting each of a plurality of reference effects, which are the plurality of effects estimated for the plurality of reference users, according to the relationship strength indicating the closeness between the target user and the reference user.
Information processing device. an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a relationship specifying means for specifying a relationship between users based on a result of clustering based on values associated with the relationship between users;
a relationship strength determination means for determining a relationship strength indicating the closeness of the users based on an index indicating the strength of the relationship between the users in accordance with a determination criterion corresponding to the relationship between the users;
a reference user specifying means for specifying at least one reference user who is related to the target user according to the relationship strength;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user , as a target effect, which is the effect for the target user.
Information processing device. The effect estimation means estimates the reference effect as the target effect when the reference effect shows a higher effect than the target effect.
The information processing device according to claim 1 . The relationship strength is determined based on an output when data representing the index is input into a trained machine learning model corresponding to the relationship between the target user and the reference user, and indicates the closeness between the target user and the reference user.
5. The information processing device according to claim 1, 3, or 4 . the relationship identification means identifies the relationship between the users based on a result of clustering based on at least one of the users' names, IP addresses, addresses, credit card numbers, ages, genders, schools they attend, places of employment, and places of stay;
5. The information processing device according to claim 3 or 4. The reference user identification means identifies another user having attribute data common to the target user as the at least one reference user.
The information processing device according to claim 1 or 4. The reference user identification means identifies the reference user based on graph data indicating relationships between users.
The information processing device according to claim 1 or 4. The system further includes a graph data generation means for generating the graph data by identifying pairs of users who are related to each other based on attribute data groups of the plurality of users.
The information processing device according to claim 9 . the effect estimation model is created based on training data in which a score based on a statistic relating to the rate of execution of the action by users who received the operation among a plurality of users having a predetermined attribute and a statistic relating to the rate of execution of the action by users who did not receive the operation among the plurality of users is defined as a score indicating the effect of the operation on users who have the attribute;
The information processing device according to claim 1 . a condition output means for outputting a condition regarding the operation for the target user based on at least the target effect, the condition output means outputting a condition such that a higher priority is given to the operation for a user whose estimated effect is higher;
The information processing device according to claim 1 . The method further comprises a risk estimation means for estimating a risk based on the probability that the user will not perform the action, using a machine learning model that outputs a risk index indicating the degree of risk based on the probability that the user will not perform the action, in response to input of one or more attributes related to the user;
the condition output means outputs, based on the estimated target effect and the risk, a condition under which a higher priority is given to at least a portion of the operations for users whose estimated effect is higher and users whose estimated risk is higher, and outputs a condition under which a lower priority is given to at least a portion of the operations for users whose estimated effect is lower and users whose estimated risk is lower.
The information processing device according to claim 12. The computer
an effect estimation step of estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and estimating the effect of the operation using an effect estimation model that outputs the causal score in response to input of one or more attributes related to the user;
a reference user specifying step of specifying at least one reference user who is related to the target user according to a relationship strength indicating the closeness of the users, the relationship strength being determined based on an index indicating the strength of the relationship between the specified users in accordance with a judgment criterion corresponding to the relationship between the users;
In the effect estimation step, a reference effect, which is the effect estimated for the reference user, is estimated as a target effect, which is the effect for the target user.
method. The computer
an effect estimation step of estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and estimating the effect of the operation using an effect estimation model that outputs the causal score in response to input of one or more attributes related to the user;
a reference user specifying step of specifying at least one reference user who is related to the target user, wherein another user who has attribute data common to the target user is specified as the at least one reference user, or the reference user is specified based on graph data showing relationships between users;
In the effect estimation step, a reference effect, which is the effect estimated for the reference user, is estimated as a target effect, which is the effect for the target user.
method. The computer
an effect estimation step of estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and estimating the effect of the operation using an effect estimation model that outputs the causal score in response to input of one or more attributes related to the user;
a reference user specifying step of specifying a plurality of reference users who are related to the target user, wherein other users who have attribute data in common with the target user are specified as the plurality of reference users, or the reference users are specified based on graph data showing relationships between users;
a relationship identification step of identifying relationships between users based on a result of clustering based on values associated with the relationships between users;
a relationship strength determination step of determining a relationship strength indicating a closeness between the users based on an index indicating the strength of the relationship between the users in accordance with a judgment criterion corresponding to the relationship between the users;
In the effect estimation step, a target effect, which is the effect on the target user, is estimated by estimating an average value of values obtained by weighting each of a plurality of reference effects, which are the effects estimated for the plurality of reference users, according to the relationship strength indicating the closeness between the target user and the reference user.
method. The computer
an effect estimation step of estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and estimating the effect of the operation using an effect estimation model that outputs the causal score in response to input of one or more attributes related to the user;
a relationship identification step of identifying relationships between users based on a result of clustering based on values associated with the relationships between users;
a relationship strength determination step of determining a relationship strength indicating a closeness between the users based on an index indicating the strength of the relationship between the users in accordance with a determination criterion corresponding to the relationship between the users;
a reference user identifying step of identifying at least one reference user who is related to the target user according to the relationship strength;
In the effect estimation step, a reference effect, which is the effect estimated for the reference user , is estimated as a target effect, which is the effect for the target user.
method. Computer,
an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
and a reference user specifying means for specifying at least one reference user who is related to the target user in accordance with a relationship strength indicating the closeness of the users, the relationship strength being determined based on an index indicating the strength of the relationship between the users in accordance with a judgment criterion corresponding to the relationship between the users;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user, as a target effect, which is the effect for the target user.
program. Computer,
an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a reference user specifying means for specifying at least one reference user who is related to the target user, the reference user specifying means being configured to specify another user who has attribute data in common with the target user as the at least one reference user, or to specify the reference user based on graph data showing the relationship between users;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user, as a target effect, which is the effect for the target user.
program. Computer,
an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a reference user specifying means for specifying a plurality of reference users who are related to the target user, the reference user specifying means specifying other users who have attribute data in common with the target user as the plurality of reference users, or specifying the reference users based on graph data showing the relationships between users;
a relationship specifying means for specifying a relationship between users based on a result of clustering based on values associated with the relationship between users;
a relationship strength determination means for determining a relationship strength indicating the closeness of the users based on an index indicating the strength of the relationship between the users in accordance with a determination criterion corresponding to the relationship between the users;
The effect estimation means estimates a target effect, which is the effect for the target user, by averaging values obtained by weighting each of a plurality of reference effects, which are the plurality of effects estimated for the plurality of reference users, according to the relationship strength indicating the closeness between the target user and the reference user.
program. Computer,
an effect estimation means for estimating the effect of a predetermined operation on a user to prompt the user to perform a predetermined action on whether or not the user will perform the action, using a causal score based on the difference between a statistical amount relating to the rate at which the action is performed by a user among a plurality of users who has received the operation and a statistical amount relating to the rate at which the action is performed by a user among the plurality of users who has not received the operation, and for estimating the effect of the operation using an effect estimation model that outputs the causal score in response to an input of one or more attributes related to the user;
a relationship specifying means for specifying a relationship between users based on a result of clustering based on values associated with the relationship between users;
a relationship strength determination means for determining a relationship strength indicating the closeness of the users based on an index indicating the strength of the relationship between the users in accordance with a determination criterion corresponding to the relationship between the users;
and a reference user specifying means for specifying at least one reference user who is related to the target user according to the relationship strength;
The effect estimation means estimates a reference effect, which is the effect estimated for the reference user , as a target effect, which is the effect for the target user.
program.