JP7215130B2 - Separation Prevention Support Program, Separation Prevention Support Method, and Information Processing Device - Google Patents

Description

The present invention relates to a separation prevention support program, a separation prevention support method, and an information processing apparatus.

Conventionally, there is a technology for extracting characteristic behavior of a consumer based on consumer data (for example, product purchase history) held by a company and tagging the consumer. The tagged information is used, for example, for marketing aimed at increasing sales at department stores, shopping centers, and the like.

As a related prior art, for example, when it is determined that an event will occur for the user based on the user's public information, based on the user's public information and the friend's product purchase history, a product to be given to the user is selected, There is one that notifies a friend of a recommendation to the effect that a selected product should be gifted to the user.

WO2016/125237

However, in the conventional technology, it is determined what kind of target and what kind of measures should be taken to prevent the user from leaving the store, such as not visiting the store or purchasing products or services. difficult.

In one aspect, an object of the present invention is to support formulation of measures to prevent user alienation.

In one embodiment, it is determined whether or not the user will depart after the predetermined period of time based on the information of a plurality of tags indicating characteristics of the user during the predetermined period of time, and a plurality of tags corresponding to the user determined to depart searching for a tag related to a measure for preventing the user from leaving the plurality of tags based on the determination result of whether or not the user will leave when the information of at least one of the tags is changed and outputting a search result or information identifying the search result in association with the user.

ADVANTAGE OF THE INVENTION According to one aspect of the present invention, there is an effect that it is possible to support formulation of a measure to prevent user separation.

FIG. 1 is an explanatory diagram showing an example of the separation prevention support method according to the first embodiment. FIG. 2 is an explanatory diagram showing a system configuration example of the information providing system 200. As shown in FIG. FIG. 3 is a block diagram showing a hardware configuration example of the information processing apparatus 101. As shown in FIG. FIG. 4 is an explanatory diagram showing an example of the contents stored in the learning data DB 220. As shown in FIG. FIG. 5 is an explanatory diagram showing an example of the contents stored in the prediction data DB 230. As shown in FIG. FIG. 6 is an explanatory diagram showing an example of the contents stored in the tag master 240. As shown in FIG. FIG. 7 is a block diagram showing a functional configuration example of the information processing apparatus 101. As shown in FIG. FIG. 8 is an explanatory diagram showing an example of updating prediction data. FIG. 9 is an explanatory diagram showing a specific example of the setting file 900. As shown in FIG. FIG. 10 is an explanatory diagram showing a specific example of the separation countermeasure file 1000. As shown in FIG. FIG. 11 is an explanatory diagram showing an example of searching for policy target tags. FIG. 12 is an explanatory diagram showing a screen example of the policy support screen 1200. As shown in FIG. FIG. 13 is a flowchart showing an example of the separation prevention support processing procedure of the information processing apparatus 101 according to the first embodiment. FIG. 14 is a flow chart showing an example of a specific processing procedure of defector prediction processing. FIG. 15 is a flowchart showing an example of a specific processing procedure of setting file creation processing. FIG. 16 is a flowchart illustrating an example of a specific processing procedure of simulation processing. FIG. 17 is an explanatory diagram of an example of the contents of the correlation coefficient table 1700. As shown in FIG. FIG. 18 is a flowchart illustrating an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the second embodiment. FIG. 19 is an explanatory diagram of an example of the contents of the total flag value table 1900. As shown in FIG. FIG. 20 is a flowchart showing an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the third embodiment. FIG. 21 is an explanatory diagram showing an example of the contents stored in the policy master 2100. As shown in FIG. FIG. 22 is an explanatory diagram showing an example of the contents of the reaction rate table 2200. As shown in FIG. FIG. 23 is a flowchart showing an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the fourth embodiment. FIG. 24 is an explanatory diagram showing an example of determination of simulation tags. FIG. 25 is a flowchart (part 1) showing an example of the separation prevention support processing procedure of the information processing apparatus 101 according to the fifth embodiment. FIG. 26 is a flowchart (part 2) showing an example of the separation prevention support processing procedure of the information processing apparatus 101 according to the fifth embodiment. FIG. 27 is a flow chart showing an example of a specific processing procedure of the second simulation processing.

Exemplary embodiments of a separation prevention support program, separation prevention support method, and information processing apparatus according to the present invention will be described below in detail with reference to the drawings.

(Embodiment 1)
FIG. 1 is an explanatory diagram showing an example of the separation prevention support method according to the first embodiment. In FIG. 1, an information processing apparatus 101 is a computer that assists in formulating measures to prevent user estrangement. The information processing apparatus 101 is, for example, a server or a PC (Personal Computer).

A user is a person who uses a product or service. Goods and services are goods and services that are sold and provided at stores, EC (Electronic Commerce) sites, and the like. A user's defection means that he or she stops using the product or service that has been used until now.

For example, a customer (user) may be disappointed or dissatisfied with a product or service provided at a certain store. In such cases, customers who stop visiting stores or using products or services are referred to as customer defection. However, what constitutes customer defection is determined according to the form of product or service provision.

Since customer defection leads to a decrease in sales of the company (operator), it is desirable to take some measures to prevent it. On the other hand, not all customers leave the company, and effective measures to prevent the defection may differ from customer to customer. For this reason, instead of blindly promoting sales, it is required to narrow down the target to customers who are likely to defect and take appropriate measures.

Therefore, in the present embodiment, separation is predicted from the information of a plurality of tags that indicate user characteristics, and by searching for tags related to measures to prevent user separation, measures to prevent separation are formulated. A separation prevention support method that supports A processing example of the information processing apparatus 101 will be described below.

(1) The information processing apparatus 101 determines whether or not the user will leave after the predetermined period T based on the information of the plurality of tags indicating the characteristics of the user during the predetermined period T. Here, the predetermined period T is a period in the past. What kind of period is set as the predetermined period T can be arbitrarily set, and is set according to, for example, the type of product, the content of the service, and the like.

A tag is for specifying a user's characteristics, and represents, for example, conditions related to actions or basic attributes. The basic attributes are sex, age, residential area, and the like. Action-related conditions include "purchase a specific product" and "use a specific service". Conditions related to basic attributes include “sex is female” and “age is 60 years old or older”.

The information of the multiple tags includes the content of each tag included in the multiple tags. The content of each tag indicates, for example, a value depending on whether the condition represented by the tag is satisfied. More specifically, for example, the content of each tag is information that becomes "1" when the condition represented by each tag is satisfied, and becomes "0" when the condition represented by each tag is not satisfied.

For example, the information processing apparatus 101 uses the machine learning model M to determine whether or not the user will leave. Here, the machine learning model M is a model that determines whether or not the user will leave from information on a plurality of tags that indicate the characteristics of the user. The machine learning model M is generated based on past data using an algorithm such as logistic regression.

In the example of FIG. 1, the user is "user 110" and the tags are "tags Ta, Tb, and Tc". User 110 is a customer of a department store. A tag Ta represents a condition of "purchase product A". A tag Tb represents a condition of "purchase product B". A tag Tc represents a condition of "purchase product C". The information of each tag Ta, Tb, Tc indicates a value (0 or 1) according to whether each condition is satisfied. Here, it is assumed that the machine learning model M is used to determine that the user 110 will leave after the predetermined period T from the information of the tags Ta, Tb, and Tc that indicate the characteristics of the user 110 during the predetermined period T.

(2) The information processing apparatus 101 selects a plurality of tags based on the determination result as to whether or not the user will depart when the information of at least one of the plurality of tags corresponding to the user determined to depart is changed. , search for tags associated with measures to prevent user defection. Here, changing tag information means, for example, changing a value according to whether or not a condition represented by a tag is satisfied.

A tag related to measures to prevent user separation is, for example, a tag that reverses the determination result of user separation. That is, the information processing apparatus 101 searches for a tag, among a plurality of tags, for which, when the information of the tag is changed, the prediction result of the user who is predicted to "leave" changes to the prediction result of "will not leave."

Specifically, for example, the information processing apparatus 101 changes the information of any one of the plurality of tags to "0⇒1" or "1⇒0", and then uses the machine learning model M to , to determine whether the user leaves. The information processing apparatus 101 repeats this process for each tag, thereby searching for a tag that changes the user's withdrawal determination result.

In the example of FIG. 1, it is assumed that when the information of the tag Ta among the tags Ta, Tb, and Tc is changed from "0" to "1", the determination result of the separation of the user 110 changes to "do not leave". . In this case, the tag Ta is searched as a tag related to measures to prevent the user 110 from leaving. When the value of the tag Ta is changed, the determination result of the separation of the user 110 is reversed.

(3) The information processing apparatus 101 outputs a search result or information specifying the search result in association with the user. Specifically, for example, the information processing apparatus 101 outputs information specifying the searched tag in association with the identification information of the user. Here, the identification information of the user is, for example, an ID or name for identifying the user.

Further, the information specifying the tag may be, for example, an ID identifying the tag, or information indicating a condition represented by the tag. Furthermore, the information specifying the tag may include a value (for example, "0" or "1") depending on whether the user satisfies the condition represented by the tag.

In the example of FIG. 1, it is assumed that the separation prevention tag information 120 is output. The separation prevention tag information 120 represents information specifying the searched tag Ta in association with the name of the user 110 . The information specifying the tag Ta includes the condition "purchase product A" represented by the tag Ta and a value "0" indicating that the user 110 does not satisfy the condition.

As described above, according to the information processing apparatus 101, when it is predicted that a user will leave the company, it is possible to provide information on measures for preventing the user from leaving. As a result, it becomes possible to determine what kind of measures should be taken for what kind of target to reduce the possibility of defection, and it is possible to support the formulation of measures to prevent defection in advance.

In the example of FIG. 1, a department store marketing staff or the like refers to the separation prevention tag information 120 to promote sales of product A to the user 110 (sets the value of the tag Ta to "1"). policy), it can be seen that the possibility of defection can be reduced.

(System configuration example of information providing system 200)
Next, the system configuration of the information providing system 200 including the information processing apparatus 101 shown in FIG. 1 will be described. The information providing system 200 is applied, for example, to a computer system that supports marketing in department stores, shopping centers, and the like. In the following description, "policy" for preventing user alienation may be referred to as "policy".

FIG. 2 is an explanatory diagram showing a system configuration example of the information providing system 200. As shown in FIG. In FIG. 2 , an information providing system 200 includes an information processing device 101 and a client device 201 . In an information providing system 200, an information processing device 101 and a client device 201 are connected via a wired or wireless network 210. FIG. The network 210 is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, or the like.

Here, the information processing apparatus 101 has a learning data DB (Database) 220 , a prediction data DB 230 and a tag master 240 . The information processing device 101 is, for example, a server. The storage contents of various DBs 220, 230, 240 will be described later with reference to FIGS. 4 to 6. FIG.

A client device 201 is a computer used by a user of the information providing system 200 . A user of the information providing system 200 is, for example, a person who conducts marketing operations for department stores, shopping centers, and the like. The client device 201 is, for example, a PC, a tablet PC, or the like.

In the above description, the information processing apparatus 101 and the client apparatus 201 are provided separately, but the present invention is not limited to this. For example, the information processing device 101 may be implemented by the client device 201 .

(Hardware Configuration Example of Information Processing Device 101)
FIG. 3 is a block diagram showing a hardware configuration example of the information processing apparatus 101. As shown in FIG. 3, an information processing apparatus 101 includes a CPU (Central Processing Unit) 301, a memory 302, a disk drive 303, a disk 304, a communication I/F (Interface) 305, and a portable recording medium I/F 306. , and a portable recording medium 307 . Also, each component is connected by a bus 300 .

Here, the CPU 301 controls the entire information processing apparatus 101 . The CPU 301 may have multiple cores. The memory 302 has, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), a flash ROM, and the like. Specifically, for example, a flash ROM stores an OS (Operating System) program, a ROM stores application programs, and a RAM is used as a work area for the CPU 301 . A program stored in the memory 302 is loaded into the CPU 301 to cause the CPU 301 to execute coded processing.

The disk drive 303 controls data read/write with respect to the disk 304 under the control of the CPU 301 . The disk 304 stores data written under the control of the disk drive 303 . Examples of the disk 304 include a magnetic disk and an optical disk.

The communication I/F 305 is connected to the network 210 through a communication line, and is connected to an external computer (for example, the client device 201 shown in FIG. 2) via the network 210 . A communication I/F 305 serves as an interface between the network 210 and the inside of the apparatus, and controls input/output of data from an external computer. For the communication I/F 305, for example, a modem or a LAN adapter can be adopted.

A portable recording medium I/F 306 controls reading/writing of data from/to a portable recording medium 307 under the control of the CPU 301 . The portable recording medium 307 stores data written under control of the portable recording medium I/F 306 . Examples of the portable recording medium 307 include CD (Compact Disc)-ROM, DVD (Digital Versatile Disk), USB (Universal Serial Bus) memory, and the like.

Note that the information processing apparatus 101 may include, for example, an SSD (Solid State Drive), an input device, a display, etc., in addition to the components described above. Further, the information processing apparatus 101 does not have to include, for example, the disk drive 303, the disk 304, the portable recording medium I/F 306, and the portable recording medium 307 among the components described above. Also, the client device 201 shown in FIG. 2 can be realized by a hardware configuration similar to that of the information processing device 101 . However, the client device 201 has an input device, a display, etc. in addition to the components described above.

(Stored contents of various DBs 220, 230, 240)
Next, the storage contents of various DBs 220, 230, and 240 of the information processing apparatus 101 will be described with reference to FIGS. 4 to 6. FIG. The various DBs 220, 230, 240 are realized by storage devices such as the memory 302 and disk 304 shown in FIG. 3, for example.

FIG. 4 is an explanatory diagram showing an example of the contents stored in the learning data DB 220. As shown in FIG. In FIG. 4, the learning data DB 220 has fields for year, customer ID, a plurality of tags, and a defection flag. 400-3) is stored as a record.

Here, the fiscal year indicates the period in which the customer's characteristics were extracted. A customer ID is an identifier that uniquely identifies a customer. A plurality of tags are for specifying customer characteristics. Each tag represents a condition regarding behavior or basic attributes. For example, the tag "buy a towel at the end of the year" represents a condition for the action "buy a towel at the end of the year".

Also, the tag "Valentine" represents a condition related to the action of "buying chocolate for Valentine's Day." A value of "0" or "1" is set in the field of each tag depending on whether or not the condition represented by each tag is satisfied. Here, a tag value of "0" indicates that the condition represented by the tag is not satisfied. A tag value of "1" indicates that the condition represented by the tag is satisfied.

The defection flag indicates whether or not the customer defected in the next fiscal year. A defected customer is, for example, a customer who has stopped visiting a store or using (purchasing) a product or service. Here, the defection flag "0" indicates that the customer did not defection in the next year. A defection flag "1" indicates that the customer defected in the next year.

For example, the learning data 400-1 contains multiple tag information (each tag value) indicating the characteristics of the customer with the customer ID "P1111" in 2017, and the fact that the customer left the company in the next fiscal year (2018). and a detachment flag "1" to indicate.

FIG. 5 is an explanatory diagram showing an example of the contents stored in the prediction data DB 230. As shown in FIG. In FIG. 5, the prediction data DB 230 has fields for year, customer ID, a plurality of tags, and a defection flag. 500-3) is stored as a record.

Here, the fiscal year indicates the period in which the customer's characteristics were extracted. A customer ID is an identifier that uniquely identifies a customer. A plurality of tags are for specifying customer characteristics. The defection flag indicates the result of predicting whether or not the customer will defection in the next fiscal year. In the initial state before prediction, the departure flag is "-(Null)".

For example, the prediction data 500-1 includes information of a plurality of tags (values of each tag) indicating the characteristics of the customer with the customer ID “P2100” in 2018 and whether or not the customer will defect in the next fiscal year (2019). and a deviation flag "-" indicating the prediction result.

In the following description, a plurality of tags for specifying user characteristics may be denoted as "tags T1 to Tm" (m: a natural number equal to or greater than 2). Also, an arbitrary tag among tags T1 to Tm may be denoted as "tag Tj" (j=1, 2, . . . , m).

FIG. 6 is an explanatory diagram showing an example of the contents stored in the tag master 240. As shown in FIG. In FIG. 6, the tag master 240 has fields of tag ID, tag name, and actionable flag, and by setting information in each field, tag management information (eg, tag management information 600-1 to 600-m) is stored as a record.

Here, the tag ID is an identifier that uniquely identifies the tag Tj. The tag name is the name of the tag Tj. Here, a character string indicating the condition represented by the tag Tj is used as the tag name. The actionable flag indicates whether or not the action is possible. "Possible to take action" indicates that the tag can be the target of measures to prevent customer estrangement.

The action possible flag "0" indicates that the action cannot be taken. The actionable flag "1" indicates that the action is possible. Here, "1" is set to the action possible flag of the tag Tj representing the action-related condition. On the other hand, "0" is set to the action possible flag of the tag Tj representing the condition regarding the basic attribute.

For example, the tag management information 600-1 indicates the tag name "Valentine" and the action possible flag "1" of the tag T1. Further, the tag management information 600-2 indicates the tag name "65 years old or older" and the action possible flag "0" of the tag T2.

(Example of functional configuration of information processing apparatus 101)
FIG. 7 is a block diagram showing a functional configuration example of the information processing apparatus 101. As shown in FIG. In FIG. 7 , the information processing apparatus 101 includes an acquisition unit 701 , a generation unit 702 , a determination unit 703 , a search unit 704 and an output unit 705 . Specifically, for example, the acquisition unit 701 to the output unit 705 cause the CPU 301 to execute a program stored in a storage device such as the memory 302 and the disk 304 shown in FIG. , to realize its function. The processing results of each functional unit are stored in a storage device such as the memory 302 or disk 304, for example.

Acquisition unit 701 acquires learning data. Here, the learning data is information used for generating the machine learning model M for determining whether or not the user will leave. The learning data includes information of tags T1 to Tm indicating characteristics of the user during a specific period and a withdrawal result indicating whether or not the user has withdrawn after the specific period.

A specific time period is a time period in the past. A user is, for example, a customer (consumer) in a retail store such as a department store or shopping center. For example, the learning data includes information (known) of tags T1 to Tm indicating characteristics of a customer in a certain year and a defection flag (known) indicating whether or not the customer defected in the next fiscal year in association with the customer ID. and

Specifically, for example, the acquisition unit 701 acquires the received learning data by receiving the learning data from another computer. Another computer is, for example, the client device 201 shown in FIG. Alternatively, the acquisition unit 701 may acquire the learning data through operation input using an input device (not shown).

The acquired learning data is stored, for example, in the learning data DB 220 shown in FIG. The learning data is generated by extracting the customer's basic attributes and actions based on POS (Point-Of-Sale) data held by companies (department stores, shopping centers, etc.).

The generator 702 generates a machine learning model M based on the acquired learning data. Specifically, for example, the generation unit 702 acquires learning data for the (X−1) year from the learning data DB 220 . (X-1) learning data is learning data for the year "X-1", and includes information on tags T1 to Tm that indicate customer characteristics in the (X-1) year, and the customer in year X and a separation flag that indicates whether or not they have separated.

At this time, the generation unit 702 acquires the same number of learning data with the estrangement flag "0" and learning data with the estrangement flag "1". Then, based on the acquired learning data for the year (X-1), the generation unit 702 sets the information of each tag T1 to Tm in the year (X-1) as an explanatory variable, and sets the value of the secession flag for the year X as an explanatory variable. A machine learning model M is generated as an objective variable.

That is, the generation unit 702 generates a machine learning model M (model formula) for determining whether or not the customer will defect in the next year from the information of each tag T1 to Tm in a certain year. For generating the machine learning model M, for example, a technique such as logistic regression can be used. However, the information processing apparatus 101 may acquire a machine learning model M generated by another computer.

The acquisition unit 701 also acquires prediction data. Here, the prediction data is information used to predict whether or not the user will leave. The prediction data includes information on tags T1 to Tm that indicate the characteristics of the target user in the predetermined period T. FIG. The target user is a target user for whom it is predicted (determined) whether or not to leave. The predetermined period T is a period in the past.

For example, the prediction data is associated with the customer ID, information of tags T1 to Tm indicating customer characteristics in a certain year (known), and a defection flag (unknown) indicating whether or not the customer will defect in the next year. and Specifically, for example, the acquisition unit 701 acquires the received prediction data by receiving the prediction data from another computer. Alternatively, the acquisition unit 701 may acquire the prediction data through operation input using an input device (not shown).

The acquired prediction data is stored, for example, in the prediction data DB 230 shown in FIG. The prediction data is generated by, for example, extracting the customer's basic attributes and behavior based on the POS data held by the company, similar to the learning data.

The determining unit 703 determines whether or not the target user will leave after the predetermined period T based on the information of the tags T1 to Tm that indicate the characteristics of the target user during the predetermined period T. Specifically, for example, the determination unit 703 uses the generated machine learning model M to determine whether or not the target user will leave.

More specifically, for example, if the predetermined period T is “X years”, the determination unit 703 selects prediction data for X years from the prediction data DB 230 . The prediction data for the year X is the prediction data for the year "X", and includes information on tags T1 to Tm indicating customer characteristics in the X year and defection indicating whether or not the customer will defect in the (X+1) year. Including flags. However, at this time, the estrangement flag is "-".

Then, using the generated machine learning model M, the determination unit 703 determines whether or not the target user will depart from the information of each of the tags T1 to Tm of the selected prediction data. As a result, it is possible to predict whether or not the customer will defect in the year (X+1) from the information of the tags T1 to Tm indicating the characteristics of the customer (target user) in the year X.

The determined determination result is stored in the prediction data DB 230, for example. That is, for the selected prediction data, the separation flag is set to "0" or "1" according to the determination result. Here, an example of updating the prediction data will be described.

FIG. 8 is an explanatory diagram showing an example of updating prediction data. In FIG. 8, prediction data 500-1 indicates information (values of "0" or "1") of tags T1 to Tm of a customer with a customer ID of "P2100" in 2018. Here, it is assumed that the machine learning model M is used to determine that the customer will leave in 2019. In this case, the determination unit 703 sets the separation flag of the prediction data 500-1 to "1".

The search unit 704 searches for the tag T1 based on the determination result as to whether or not the target user will depart when the information of at least one of the tags Tj of the tags T1 to Tm corresponding to the target user determined to depart is changed. Search for a policy target tag from ~Tm. Here, the policy target tag is a tag related to a policy to prevent the target user from leaving.

The policy target tag is, for example, a tag that changes the determination result (prediction result) of the defection of the target user. Measures to prevent alienation are, for example, sales promotion activities using direct mail, e-mail, coupons, and the like. Also, changing the information of the tag Tj means, for example, changing the value according to whether or not the condition represented by the tag Tj is satisfied.

For example, the search unit 704 determines whether or not the target user will depart when the value of at least one of the tags T1 to Tm Tj is changed from "0" to "1". search for a tag that inverts the decision result of separation from

In the following description, among the tags T1 to Tm, the tag for simulating whether or not the target user will leave by changing the value may be referred to as a "simulation tag".

More specifically, for example, the search unit 704 selects prediction data with a departure flag of “1” from the prediction data DB 230 . Also, the search unit 704 selects a simulation tag from the tags T1 to Tm. Next, the searching unit 704 changes the value of the selected simulation tag from "0" to "1" for the selected prediction data.

Then, using the machine learning model M, the search unit 704 determines whether or not the target user will depart from the values of the respective tags T1 to Tm of the prediction data after changing the value of the selected simulation tag. do. Here, when it is determined that the target user still leaves, the search unit 704 reselects the simulation tag and performs the same processing.

On the other hand, when it is predicted that the target user will not leave, the searching unit 704 identifies the selected simulation tag as the policy target tag. As a result, by changing the value from "0" to "1", it is possible to search for a tag that reverses the determination result of the target user's defection as a target of measures for preventing the target user's defection.

However, tags T1 to Tm include tags that cannot be subject to the measures. For example, the user's basic attributes such as gender and age do not change depending on the policy.

Therefore, the search unit 704 determines whether or not the target user will depart when the information of at least one of the tags Tj, excluding tags related to the basic attributes of the target user, is changed among the tags T1 to Tm. Alternatively, the policy target tag may be searched for. That is, among the tags T1 to Tm, tags related to basic attributes such as gender and age of the target user may be excluded from the search range.

Specifically, for example, the searching unit 704 refers to the tag master 240 shown in FIG. 6, and determines the tag with the action possible flag of "1" among the tags T1 to Tm as the simulation tag. As a result, the tag Tj representing the action-related condition can be determined as the simulation tag. The determined result is registered, for example, in a setting file 900 as shown in FIG.

FIG. 9 is an explanatory diagram showing a specific example of the setting file 900. As shown in FIG. In FIG. 9, a configuration file 900 contains tag IDs for simulation tags. The search unit 704 selects a simulation tag by referring to the setting file 900, for example. Then, the search unit 704 searches for the policy target tag based on the determination result of whether or not the target user will leave when the value of the selected simulation tag is changed.

The searched search result or information specifying the search result is associated with the target user, for example, and stored in the estrangement countermeasure file 1000 as shown in FIG.

FIG. 10 is an explanatory diagram showing a specific example of the separation countermeasure file 1000. As shown in FIG. In FIG. 10, the defection countermeasure file 1000 stores defection countermeasure information (for example, defection countermeasure information 1000-1 to 1000-3) including customer IDs and tag names. Here, the customer ID is an identifier that uniquely identifies the customer who is the target user. The tag name is the name of the policy target tag.

For example, the defection countermeasure information 1000-1 is information specifying the tag "Valentine" to be targeted for measures to prevent the defection of the customer (target user) with the customer ID "P2100".

Returning to the description of FIG. 7, the output unit 705 outputs the search result of the search or information specifying the search result in association with the target user. The output format of the output unit 705 includes, for example, storage in a storage device such as the memory 302 and the disk 304, transmission to another computer via the communication I/F 305, display on a display (not shown), and output to a printer (not shown). print output, etc.

Specifically, for example, the output unit 705 refers to the defection countermeasure file 1000 as shown in FIG. 10, and provides the client device 201 shown in FIG. The display of the screen 1200 may be controlled. A screen example of the measure support screen 1200 will be described later with reference to FIG. 12 .

(Example of searching for tags targeted for measures)
Next, an example of a search for a policy target tag will be described with reference to FIG. In the following description, the customer with the customer ID "P#" may be referred to as "customer P#".

FIG. 11 is an explanatory diagram showing an example of searching for policy target tags. In FIG. 11, prediction data 1101 to 1103 include the values of tags T1 to Tm indicating the characteristics of customers P1111 to P1113 in 2018 and the withdrawal flags indicating whether or not the customers P1111 to P1113 will withdraw in 2019. include. However, the year "2018" of each prediction data 1101 to 1103 is omitted from the illustration.

Here, it is determined that all of the customers P1111-P1113 will defect in 2019. In this case, the information processing apparatus 101 changes the value of the simulation tag for each of the customers P1111 to P1113, for example, and determines whether or not each of the customers P1111 to P1113 will leave. Here, it is assumed that the tag "Valentine" is selected as the simulation tag.

In this case, the information processing apparatus 101 changes the value of the tag "Valentine" of the prediction data 1101 to 1103 from "0" to "1". Next, using the machine learning model M, the information processing apparatus 101 calculates each It is determined whether or not the customers P1111 to P1113 are leaving.

Here, the customers P1112 and P1113 are still judged to be separated, but the customer P1111 is judged not to be separated. That is, by changing the value of the tag "Valentine" from "0" to "1", the customer P1111's defection determination result is reversed.

In this case, the information processing apparatus 101 identifies the tag "Valentine" as the policy target tag for the customer P1111. Then, the information processing apparatus 101 outputs information (for example, tag name) related to the tag "Valentine" in association with the customer P1111. This makes it possible to provide information on measures to prevent customer P1111 from leaving the company in fiscal 2019.

(Screen example of policy support screen 1200)
Next, a screen example of the measure support screen 1200 displayed on the client device 201 will be described with reference to FIG. 12 .

FIG. 12 is an explanatory diagram showing a screen example of the policy support screen 1200. As shown in FIG. In FIG. 12, a measure support screen 1200 is a screen for providing information on measures to prevent customer defection in the next year (2019), and includes defection prevention tag information 1201-1203. The separation prevention tag information 1201-1203 includes a customer ID and a tag name. A customer ID is an identifier that uniquely identifies a customer. The tag name is the name of the policy target tag.

According to the measure support screen 1200, it is possible to provide information on effective measures for preventing customer defection in the next year (2019). As a result, the company's marketing staff can determine what kind of measures should be taken to prevent the defection of customers in the next fiscal year.

For example, the marketing staff may change the value of the tag "Valentine" to "1" for customer P2100. I know it can be lowered. Similarly, the marketing staff can understand that if the customer P2101 is promoted to buy chocolates for Valentine's Day, the likelihood of defection in the next fiscal year can be reduced. In addition, the marketing manager has a plan to change the value of the tag "buy towels at the end of the year" to "1" for customer P2100. It can be seen that the possibility of separation can be reduced.

(Separation Prevention Support Processing Procedure of Information Processing Device 101)
Next, the separation prevention support processing procedure of the information processing apparatus 101 according to the first embodiment will be described with reference to FIG. The separation prevention support process of the information processing apparatus 101 is executed in response to a request from the client apparatus 201, for example.

FIG. 13 is a flowchart showing an example of the separation prevention support processing procedure of the information processing apparatus 101 according to the first embodiment. In the flowchart of FIG. 13, first, the information processing apparatus 101 acquires learning data for year (X-1) from the learning data DB 220 (step S1301). Next, the information processing apparatus 101 generates a machine learning model M for determining whether or not the customer will defect in the next fiscal year based on the acquired learning data (step S1302).

Then, the information processing apparatus 101 uses the generated machine learning model M to execute defector prediction processing for predicting whether or not the customer will defect in the next year (step S1303). A specific processing procedure of the defector prediction processing will be described later with reference to FIG. 14 .

Next, the information processing apparatus 101 executes setting file creation processing for creating a setting file for registering simulation tags (step S1304). A specific processing procedure of the setting file creation processing will be described later with reference to FIG. 15 .

Next, the information processing apparatus 101 selects, from the prediction data DB 230, the prediction data with the unselected separation flag "1" from the prediction data for year X (step S1305).

Then, the information processing apparatus 101 changes the value of the simulation tag for the selected prediction data, and executes simulation processing for simulating whether or not the customer will leave (step S1306). A specific processing procedure of the simulation processing will be described later with reference to FIG. 16 .

Next, the information processing apparatus 101 determines whether or not there is any prediction data with an unselected separation flag of "1" among the prediction data for year X (step S1307). If there is unselected prediction data (step S1307: Yes), the information processing apparatus 101 returns to step S1305.

On the other hand, if there is no unselected prediction data (step S1307: No), the information processing apparatus 101 outputs a separation countermeasure file (for example, see FIG. 10) created in the flowchart of FIG. 16 described later (step S1308), the series of processes according to this flowchart is terminated.

This makes it possible to provide information on measures to prevent the next year's customers (target users) from leaving.

• Specific Processing Procedures of the Separation Prediction Processing Next, the specific processing procedures of the defection prediction processing in step S1303 shown in FIG. 13 will be described with reference to FIG.

FIG. 14 is a flow chart showing an example of a specific processing procedure of defector prediction processing. In the flowchart of FIG. 14, first, the information processing apparatus 101 selects unselected prediction data for year X from the prediction data DB 230 (step S1401).

Next, the information processing apparatus 101 uses the machine learning model M generated in step S1302 to determine whether or not the customer will depart in the (X+1) year based on the selected prediction data (step S1402). . Then, the information processing apparatus 101 determines whether or not it is determined that the customer will leave (step S1403).

Here, if it is determined that the customer will not leave (step S1403: No), the information processing apparatus 101 sets "0" to the separation flag of the selected prediction data (step S1404), and proceeds to step S1406. . On the other hand, if it is determined that the customer will leave (step S1403: Yes), the information processing apparatus 101 sets the defection flag of the selected prediction data to "1" (step S1405).

Then, the information processing apparatus 101 determines whether or not there is unselected prediction data for year X that has not been selected from the prediction data DB 230 (step S1406). If there is unselected prediction data (step S1406: Yes), the information processing apparatus 101 returns to step S1401.

On the other hand, if there is no unselected prediction data (step S1406: No), the information processing apparatus 101 returns to the step of calling the defector prediction process. As a result, it is possible to predict whether or not the customer will depart in the (X+1) year from the prediction data of the customer in the X year.

Concrete Processing Procedure of Setting File Creation Processing Next, a specific processing procedure of the configuration file creation processing in step S1304 shown in FIG. 13 will be described with reference to FIG.

FIG. 15 is a flowchart showing an example of a specific processing procedure of setting file creation processing. In the flowchart of FIG. 15, first, the information processing apparatus 101 sets "j" of the tag Tj to "j=1" (step S1501), refers to the tag master 240, and selects the tag Tj (step S1502).

Next, the information processing apparatus 101 determines whether or not the action possible flag of the selected tag Tj is "1" (step S1503). Here, if the action possible flag is "0" (step S1503: No), the information processing apparatus 101 proceeds to step S1505.

On the other hand, if the action possible flag is "1" (step S1503: Yes), the information processing apparatus 101 registers the selected tag Tj as a simulation tag in the setting file (step S1504). The information processing apparatus 101 increments "j" of the tag Tj (step S1505).

The information processing apparatus 101 then determines whether "j" is greater than "m" (step S1506). Here, if "j" is equal to or less than "m" (step S1506: No), the information processing apparatus 101 returns to step S1502. On the other hand, if "j" is greater than "m" (step S1506: Yes), the information processing apparatus 101 returns to the step that called the setting file creation process.

Thus, among the tags T1 to Tm, tags related to basic attributes such as sex and age of the customer (target user) can be excluded from the search range.

Concrete Processing Procedures of Simulation Processing Next, concrete processing procedures of the simulation processing in step S1306 shown in FIG. 13 will be described with reference to FIG.

FIG. 16 is a flowchart illustrating an example of a specific processing procedure of simulation processing. In the flowchart of FIG. 16, first, the information processing apparatus 101 selects an unselected simulation tag by referring to the setting file created in the flowchart of FIG. 15 (step S1601).

Next, the information processing apparatus 101 refers to the prediction data selected in step S1305 shown in FIG. 13 and determines whether the value of the selected simulation tag is "0" (step S1602). Here, if the value of the simulation tag is "1" (step S1602: No), the information processing apparatus 101 proceeds to step S1608.

On the other hand, if the value of the simulation tag is "0" (step S1602: Yes), the information processing apparatus 101 changes the value of the simulation tag of the selected prediction data from "0" to "1" (step S1603). . Next, the information processing apparatus 101 uses the generated machine learning model M to determine whether or not the customer will depart in the (X+1) year based on the prediction data after the change in which the value of the simulation tag has been changed. (step S1604).

Then, the information processing apparatus 101 determines whether or not it is determined that the customer will leave (step S1605). Here, if it is determined that the customer will leave (step S1605: Yes), the information processing apparatus 101 proceeds to step S1607.

On the other hand, if it is determined that the customer will not leave the customer (step S1605: No), the information processing apparatus 101 associates the customer ID of the selected prediction data with the tag name of the selected simulation tag, and stores it in the defection countermeasure file. Register (step S1606).

Next, the information processing apparatus 101 returns the value of the simulation tag of the selected prediction data from "1" to "0" (step S1607). Then, the information processing apparatus 101 refers to the setting file and determines whether or not there is an unselected simulation tag (step S1608).

Here, if there is an unselected simulation tag (step S1608: Yes), the information processing apparatus 101 returns to step S1601. On the other hand, if there is no unselected simulation tag (step S1608: No), the information processing apparatus 101 returns to the step that called the simulation process.

As a result, it is possible to search for a simulation tag that reverses the customer's (target user's) estrangement determination result as a target of measures for preventing customer estrangement.

As described above, according to the information processing apparatus 101 according to the first embodiment, the target user departs after the predetermined period T based on the values of the tags T1 to Tm indicating the characteristics of the target user during the predetermined period T. It is possible to determine whether Further, according to the information processing apparatus 101, when the value of at least one of tags Tj of tags T1 to Tm corresponding to the target user determined to leave is changed, based on the determination result of whether or not the target user will leave , the target tag can be searched from the tags T1 to Tm. The policy target tag is a tag related to a policy to prevent the target user from leaving. Then, according to the information processing apparatus 101, it is possible to output the searched search result or the information specifying the search result in association with the target user.

This makes it possible to provide information on effective measures for preventing the defection of a target user (for example, a customer of a retail store) who is predicted to defect. For example, a company's marketing staff can decide what kind of measures to take for customers to prevent churn.

Further, according to the information processing apparatus 101, it is possible to search for a tag that changes the determination result of the defection of the target user as the policy target tag.

As a result, among the tags T1 to Tm, it is possible to search for a tag that changes the prediction result of the target user who is predicted to leave the target user to "do not leave" when the value is changed, which is an effective measure for preventing separation. can provide information about

Further, according to the information processing apparatus 101, when the value of at least one of the tags T1 to Tm excluding tags related to the basic attributes of the target user is changed, the determination result of whether or not the target user will leave Based on this, it is possible to search for policy target tags.

As a result, it is possible to prevent tags related to basic attributes such as gender and age, which cannot be changed by an action from the company side, from being searched as policy target tags.

Further, according to the information processing apparatus 101, by referring to the learning data DB 220, it is possible to generate a machine learning model M that determines whether or not the values of the tags T1 to Tm deviate. The learning data DB 220 stores, for each of a plurality of users, the values of tags T1 to Tm indicating the characteristics of each user during a specific period, and the separation result indicating whether or not each user has separated after the specific period. Stores training data containing Then, the information processing apparatus 101 can use the generated machine learning model M to determine whether or not each user will leave.

As a result, based on the past data, what kind of data (values of tags T1 to Tm) users tend to leave or do not tend to leave is learned, and each user's Separation can be predicted.

(Embodiment 2)
Next, the information processing apparatus 101 according to the second embodiment will be explained. In the second embodiment, a case will be described where the simulation tag is determined from the tags T1 to Tm in consideration of the correlation between the tag Tj and the separation flag. Illustrations and descriptions of the same portions as those described in the first embodiment are omitted.

Specific processing contents of each functional unit of the information processing apparatus 101 will be described below.

The determining unit 703 determines whether or not each user will leave after the predetermined period T, based on the information of the tags T1 to Tm indicating the characteristics of each user of the plurality of users during the predetermined period T. Here, a plurality of users are a plurality of users for whom it is predicted (determined) whether or not they will leave. For example, if the predetermined period T is “X years”, the multiple users are customers identified by the respective customer IDs of the multiple forecast data for the X year in the forecast data DB 230 .

Specifically, for example, the determination unit 703 uses the generated machine learning model M to determine whether or not each user will leave based on each prediction data for year X in the prediction data DB 230. do. As a result, it is possible to predict whether or not each customer will depart in (X+1) year from the information of tags T1 to Tm that indicate the characteristics of each customer (each user) in X year.

In the following description, the prediction data for which the separation flag is set as the judgment result (prediction result) as to whether or not to leave may be referred to as "prediction data (post-prediction)".

The search unit 704 determines a simulation tag (tag to be processed) from the tags T1 to Tm based on the determination result as to whether or not each user will leave. Then, the searching unit 704 searches for the policy target tag based on the determination result as to whether or not the target user will leave when the value of at least one of the determined simulation tags is changed.

Specifically, for example, the searching unit 704 calculates the degree of correlation between each tag Tj included in the tags T1 to Tm and the result of the separation judgment of each user. The searching unit 704 then determines a simulation tag based on the calculated degree of correlation of each tag Tj.

More specifically, for example, the search unit 704 uses the following formula (1) based on each prediction data (after prediction) in the prediction data DB 230 to determine the relationship between the tag Tj and the separation flag. A correlation coefficient Cj can be calculated. However, let a plurality of users be "customers P1 to Pn" (where n is a natural number equal to or greater than 2), and an arbitrary customer among the customers P1 to Pn be "customer Pi". x _i is the value (0 or 1) of tag Tj for customer Pi. y _i is the value (0 or 1) of the customer Pi's withdrawal flag. μ _x is the average value of the tag Tj values for the customers P1 to Pn. μ _y is the average value of the estrangement flags of the customers P1 to Pn.

When there is a positive correlation, the correlation coefficient Cj takes a positive value (0<Cj≤1). When there is no correlation, the correlation coefficient Cj is "0". When there is a negative correlation, the correlation coefficient Cj takes a negative value (-1≤Cj<0). The closer the correlation coefficient Cj is to "1", the stronger the positive correlation, and the closer the correlation coefficient Cj is to "-1", the stronger the negative correlation.

The calculated correlation coefficient Cj is stored, for example, in a correlation coefficient table 1700 as shown in FIG. The correlation coefficient table 1700 is realized by a storage device such as the memory 302 and disk 304 of the information processing apparatus 101 shown in FIG. 3, for example.

FIG. 17 is an explanatory diagram of an example of the contents of the correlation coefficient table 1700. As shown in FIG. In FIG. 17, correlation coefficient table 1700 has fields of tag ID and correlation coefficient, and by setting information in each field, correlation information 1700-1 to 1700-m is stored as records. The tag ID is an identifier that uniquely identifies the tag Tj. The correlation coefficient is an index value indicating the degree of correlation between the tag Tj and the separation flag.

Here, the stronger the negative correlation, the more the separation flag tends to become "0" when the value of the tag Tj becomes "1". Therefore, the search unit 704 may refer to the correlation coefficient table 1700 and determine N tags as simulation tags in descending order of the correlation coefficient C. FIG. N is a natural number of 1 or more and can be set arbitrarily.

However, it is assumed here that a measure is taken to change the value of the measure target tag from "0" to "1". Also, tags related to basic attributes may be excluded from the simulation tags.

(Separation Prevention Support Processing Procedure of Information Processing Device 101)
Next, a separation prevention support processing procedure of the information processing apparatus 101 according to the second embodiment will be described. However, except for the specific processing procedure of the setting file creation processing in step S1304 shown in FIG. 13, the separation prevention support processing of the information processing apparatus 101 according to the first embodiment is the same. Here, only the specific processing procedure of the setting file creation processing of the information processing apparatus 101 according to the second embodiment will be described.

FIG. 18 is a flowchart illustrating an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the second embodiment. In the flowchart of FIG. 18, first, the information processing apparatus 101 sets the number of simulations N (step S1801). Note that the number of simulations N can be arbitrarily set, and is set to a value of about 2 to 3, for example.

Next, the information processing apparatus 101 sets "j" of the tag Tj to "j=1" (step S1802), refers to the tag master 240, and selects the tag Tj (step S1803). Then, the information processing apparatus 101 determines whether or not the action possible flag of the selected tag Tj is "1" (step S1804).

Here, if the action possible flag is "0" (step S1804: No), the information processing apparatus 101 proceeds to step S1807.

On the other hand, if the action possible flag is "1" (step S1804: Yes), the information processing apparatus 101 calculates the correlation coefficient Cj between the selected tag Tj and the separation flag (step S1805). The correlation coefficient Cj is calculated from each prediction data (after prediction) in the prediction data DB 230 in the prediction data DB 230, for example, using the above equation (1).

The information processing apparatus 101 then records the calculated correlation coefficient Cj in the correlation coefficient table 1700 (step S1806). The information processing apparatus 101 increments "j" of the tag Tj (step S1807). The information processing apparatus 101 then determines whether "j" is greater than "m" (step S1808).

Here, if "j" is equal to or less than "m" (step S1808: No), the information processing apparatus 101 returns to step S1803. On the other hand, if "j" is greater than "m" (step S1808: Yes), the information processing apparatus 101 refers to the correlation coefficient table 1700 and extracts N tags in descending order of correlation coefficient C. (step S1809).

Then, the information processing apparatus 101 registers the extracted N tags as simulation tags in the setting file (step S1810), and returns to the step of calling the setting file creation process. As a result, for all customers in year X, a tag Tj that has a strong negative correlation with the estrangement flag, that is, a tag Tj that has a strong tendency for the estrangement flag to become "0" when the value becomes "1" is set as a simulation tag. can be determined to

As described above, according to the information processing apparatus 101 according to the second embodiment, based on the values of the tags T1 to Tm indicating the characteristics of each of the plurality of users in the predetermined period T, after the predetermined period T It is possible to determine whether or not the user will leave, and based on the determined result of whether or not each user will leave, the simulation tag (tag to be processed) can be determined from the tags T1 to Tm. Then, according to the information processing apparatus 101, it is possible to search for the policy target tag based on the determination result as to whether or not the target user will leave when the value of at least one of the determined simulation tags is changed. can.

As a result, it is possible to narrow down the search range for policy target tags in consideration of the prediction result of the separation of each user, and to improve the efficiency of the search process.

Further, according to the information processing apparatus 101, based on the determination result as to whether or not each user has departed, the correlation between each tag Tj included in the tags T1 to Tm and the determination result of each user's defection is calculated. A degree of correlation (for example, correlation coefficient Cj) can be calculated. Then, according to the information processing apparatus 101, a simulation tag can be determined based on the calculated degree of correlation of each tag Tj.

As a result, for example, among the tags T1 to Tm, if the value is "1", a tag that is highly likely to be predicted not to leave can be determined as a simulation tag. It is possible to efficiently search for tags that lead to specific measures.

(Embodiment 3)
Next, the information processing apparatus 101 according to the third embodiment will be described. In the third embodiment, a case will be described in which simulation tags (tags to be processed) are determined in consideration of values of tags of customers predicted not to leave. Illustrations and descriptions of the same portions as those described in the first and second embodiments are omitted.

Specific processing contents of each functional unit of the information processing apparatus 101 will be described below.

The determining unit 703 determines whether or not each user will leave after the predetermined period T, based on the information of the tags T1 to Tm indicating the characteristics of each user of the plurality of users during the predetermined period T. Specifically, for example, the determination unit 703 uses the generated machine learning model M to determine whether or not each user will leave based on each prediction data for year X in the prediction data DB 230. .

As a result, it is possible to predict whether or not each customer will depart in (X+1) year from the information of tags T1 to Tm that indicate the characteristics of each customer (each user) in X year.

The searching unit 704 identifies a user determined not to depart from among the plurality of users based on the determination result of whether or not each determined user will depart. Specifically, for example, the searching unit 704 extracts the prediction data (post-prediction) with the departure flag set to “0” from the prediction data (post-prediction) for year X in the prediction data DB 230 . Then, the search unit 704 identifies the customer identified by the customer ID of the extracted prediction data (after prediction) as a continuous predictor.

Thereby, it is possible to specify the user who was determined not to leave the company in the (X+1) year. In the following description, the customer (target user) determined not to leave may be referred to as "continuous predictor".

The search unit 704 determines a simulation tag based on the information of the tags T1 to Tm that indicate the characteristics of the identified continuous predictor. Specifically, for example, the searching unit 704 calculates the total flag value Sj by accumulating the values of the tags Tj of the prediction data (after prediction) with the extracted separation flag “0”.

The calculated flag total value Sj is stored, for example, in a flag total value table 1900 as shown in FIG. The total flag value table 1900 is implemented by a storage device such as the memory 302 and disk 304 of the information processing apparatus 101 shown in FIG. 3, for example.

FIG. 19 is an explanatory diagram of an example of the contents of the total flag value table 1900. As shown in FIG. In FIG. 19, the flag total value table 1900 has fields of tag ID and flag total value, and by setting information in each field, flag total value information 1900-1 to 1900-m is stored as records. The tag ID is an identifier that uniquely identifies the tag Tj. The total flag value is a total value obtained by accumulating the values of the tags Tj of all continuous predictors.

For example, the total flag value S1 is obtained by accumulating the value of the tag T1 of each continuous predictor. It can be said that the larger the total flag value Sj, the more likely the tag Tj is to the continuous predictors. Therefore, the searching unit 704 may refer to the total flag value table 1900 and determine N tags in descending order of the total flag value S as simulation tags.

However, it is assumed here that a measure is taken to change the value of the measure target tag from "0" to "1". Also, tags related to basic attributes may be excluded from the simulation tags.

(Separation Prevention Support Processing Procedure of Information Processing Device 101)
Next, a separation prevention support processing procedure of the information processing apparatus 101 according to the third embodiment will be described. However, except for the specific processing procedure of the setting file creation processing in step S1304 shown in FIG. 13, the separation prevention support processing of the information processing apparatus 101 according to the first embodiment is the same. Here, only the specific processing procedure of the setting file creation processing of the information processing apparatus 101 according to the third embodiment will be described.

FIG. 20 is a flowchart showing an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the third embodiment. In the flowchart of FIG. 20, first, the information processing apparatus 101 sets the number of simulations N (step S2001). Then, the information processing apparatus 101 extracts the prediction data (post-prediction) with the departure flag set to "0" from the prediction data (post-prediction) for year X in the prediction data DB 230 (step S2002).

Next, the information processing apparatus 101 sets "j" of the tag Tj to "j=1" (step S2003), refers to the tag master 240, and selects the tag Tj (step S2004). Then, the information processing apparatus 101 determines whether or not the action possible flag of the selected tag Tj is "1" (step S2005).

Here, if the action possible flag is "0" (step S2005: No), the information processing apparatus 101 proceeds to step S2008.

On the other hand, if the action possible flag is "1" (step S2005: Yes), the information processing apparatus 101 accumulates the values of the extracted prediction data (post-prediction) of the tag Tj to obtain the total flag value of the tag Tj. Sj is calculated (step S2006).

The information processing apparatus 101 then records the calculated total flag value Sj in the total flag value table 1900 (step S2007). Next, the information processing apparatus 101 increments "j" of the tag Tj (step S2008). The information processing apparatus 101 then determines whether "j" is greater than "m" (step S2009).

Here, if "j" is equal to or less than "m" (step S2009: No), the information processing apparatus 101 returns to step S2004. On the other hand, when "j" is greater than "m" (step S2009: Yes), the information processing apparatus 101 refers to the total flag value table 1900 and extracts N tags in descending order of the total flag value S. (step S2010).

Then, the information processing apparatus 101 registers the extracted N tags as simulation tags in the setting file (step S2011), and returns to the step of calling the setting file creation process. As a result, the tag Tj frequently used by the continuous predictors can be determined as the simulation tag.

As described above, according to the information processing apparatus 101 according to the third embodiment, based on the values of the tags T1 to Tm that indicate the characteristics of each user of a plurality of users during the predetermined period T, after the predetermined period T It can be determined whether or not the user leaves. Further, according to the information processing apparatus 101, it is possible to specify a continuation predictor determined not to depart from among a plurality of users based on the determination result of whether or not each determined user will depart. Then, according to the information processing apparatus 101, a simulation tag (tag to be processed) can be determined based on the values of the tags T1 to Tm indicating the characteristics of the identified continuous predictor.

As a result, it is possible to determine tags with a high rate of "1" among continuous predictors as simulation tags, and to efficiently search for tags that lead to effective measures to prevent customer defection. can be done.

(Embodiment 4)
Next, the information processing apparatus 101 according to the fourth embodiment will be described. In the fourth embodiment, a case will be described in which simulation tags (tags to be processed) are determined in consideration of the response rate to the policy. Illustrations and descriptions of the same parts as those described in the first to third embodiments are omitted.

Specific processing contents of each functional unit of the information processing apparatus 101 will be described below.

The search unit 704 determines a simulation tag for each tag Tj included in the tags T1 to Tm, based on the ratio of users who responded to the policy among the users for whom the policy related to each tag Tj was executed. Here, the measures related to the tag Tj are, for example, sales promotion activities using direct mail or the like regarding products and services related to the tag Tj.

A user who reacts to the policy is, for example, a user who takes actions such as visiting a store or purchasing a product or service in response to the policy. In the following description, the ratio of users who responded to the policy among the group of users for whom the policy related to the tag Tj was executed may be referred to as "response rate".

The response rate of each tag Tj is calculated, for example, based on the contents stored in the measure master 2100 as shown in FIG. Here, the storage contents of the measure master 2100 that the information processing apparatus 101 has will be described. The policy master 2100 is realized by a storage device such as the memory 302 and the disk 304, for example.

FIG. 21 is an explanatory diagram showing an example of the contents stored in the policy master 2100. As shown in FIG. In FIG. 21, the measure master 2100 has fields for tag ID, number of people responding to sales promotion, and number of people for sales promotion. Store as a record.

The tag ID is an identifier that uniquely identifies the tag Tj. The number of people who responded to the sales promotion is the number of customers who responded to the sales promotion (policy) related to the tag Tj. For example, the number of users who purchased a certain product within one to two weeks after the sales promotion was carried out is counted as the number of people who responded to the sales promotion. The sales promotion number is the number of customers for whom the sales promotion (policy) related to the tag Tj was performed.

For example, the policy information 2100-1 indicates the number of people who responded to the sales promotion of "2000" and the number of people who responded to the sales promotion of "10000" for the tag T1. Note that the information processing apparatus 101 receives policy information in the policy master 2100 by receiving it from another computer (for example, the client device 201 shown in FIG. 2) or by operating an input device (not shown). to get

The search unit 704, for example, refers to the policy master 2100 and calculates the reaction rate Rj of each tag Tj. More specifically, for example, the search unit 704 refers to the measure information 2100-1 and divides the number of people responding to the sales promotion “2000” by the number of people responding to the sales promotion “10000” to obtain the response rate “0.2” for the tag T1. Calculate

The calculated reaction rate Rj is stored, for example, in a reaction rate table 2200 as shown in FIG. The reaction rate table 2200 is implemented by a storage device such as the memory 302 and disk 304, for example.

FIG. 22 is an explanatory diagram showing an example of the contents of the reaction rate table 2200. As shown in FIG. In FIG. 22, the reaction rate table 2200 has fields of tag ID and reaction rate, and by setting information in each field, reaction rate information 2200-1 to 2200-m is stored as records.

The tag ID is an identifier that uniquely identifies the tag Tj. The response rate indicates the ratio of users who responded to the policy among the user group to whom the policy related to the tag Tj was implemented. Note that the response rate Rj of a tag Tj with no policy information in the policy master 2100, for example, a tag Tj with no track record of a policy, is set to "0".

Here, it can be said that the higher the response rate Rj of the tag Tj, the more likely it is that the effect of sales promotion can be obtained. Therefore, the search unit 704 may refer to the reaction rate table 2200 and determine the N tags in descending order of the reaction rate R as simulation tags. However, it is assumed here that a measure is taken to change the value of the measure target tag from "0" to "1". Also, tags related to basic attributes may be excluded from the simulation tags.

(Separation Prevention Support Processing Procedure of Information Processing Device 101)
Next, the separation prevention support processing procedure of the information processing apparatus 101 according to the fourth embodiment will be described. However, except for the specific processing procedure of the setting file creation processing in step S1304 shown in FIG. 13, the separation prevention support processing of the information processing apparatus 101 according to the first embodiment is the same. Here, only the specific processing procedure of the setting file creation processing of the information processing apparatus 101 according to the fourth embodiment will be described.

FIG. 23 is a flowchart showing an example of a specific processing procedure of setting file creation processing of the information processing apparatus 101 according to the fourth embodiment. In the flowchart of FIG. 23, first, the information processing apparatus 101 sets the number of simulations N (step S2301).

Next, the information processing apparatus 101 sets "j" of the tag Tj to "j=1" (step S2302), refers to the tag master 240, and selects the tag Tj (step S2303). Then, the information processing apparatus 101 determines whether or not the action possible flag of the selected tag Tj is "1" (step S2304).

Here, if the measure possible flag is "0" (step S2304: No), the information processing apparatus 101 proceeds to step S2307. On the other hand, if the measure possible flag is "1" (step S2304: Yes), the information processing apparatus 101 refers to the measure master 2100 and calculates the reaction rate Rj of the tag Tj (step S2305).

The information processing apparatus 101 then records the calculated reaction rate Rj in the reaction rate table 2200 (step S2306). The information processing apparatus 101 increments "j" of the tag Tj (step S2307). The information processing apparatus 101 then determines whether "j" is greater than "m" (step S2308).

Here, if "j" is equal to or less than "m" (step S2308: No), the information processing apparatus 101 returns to step S2303. On the other hand, when "j" is greater than "m" (step S2308: Yes), the information processing apparatus 101 refers to the reaction rate table 2200 and extracts N tags in descending order of reaction rate R ( step S2309).

Then, the information processing apparatus 101 registers the extracted N tags as simulation tags in the setting file (step S2310), and returns to the step of calling the setting file creation process. As a result, a tag Tj with a high response rate Rj to the measure (sales promotion) can be determined as a simulation tag.

As described above, according to the information processing apparatus 101 according to the fourth embodiment, for each tag Tj included in the tags T1 to Tm, out of the user group for whom the policy related to each tag Tj was executed, Simulation tags can be determined based on the percentage of users who responded (eg, response rate Rj).

As a result, it is possible to determine a tag that is likely to have the effect of a measure (sales promotion) as a simulation tag, and to efficiently search for a tag that leads to an effective measure to prevent customer defection.

(Embodiment 5)
Next, the information processing apparatus 101 according to the fifth embodiment will be described. In the fifth embodiment, a case will be described in which simulation tags (tags to be processed) are determined in consideration of the tag value of a continuation predictor who has a strong correlation with a defection predictor. Illustrations and descriptions of the same parts as those described in the first to fourth embodiments are omitted.

Specific processing contents of each functional unit of the information processing apparatus 101 will be described below. In the following description, a customer (target user) who is determined to leave the company may be referred to as a "predicted leaver".

The search unit 704 identifies a continuation predictor determined not to depart from among the plurality of users, based on the determination result of whether or not each determined user will depart. Specifically, for example, the searching unit 704 extracts the prediction data (post-prediction) with the departure flag set to “0” from the prediction data (post-prediction) for year X in the prediction data DB 230 .

Then, the search unit 704 identifies the customer identified by the customer ID of the extracted prediction data (after prediction) as a continuous predictor. Thereby, it is possible to identify the continuous predictors determined not to leave in the (X+1) year.

The searching unit 704, for each of the identified continuation predictors, based on the values of the tags T1 to Tm indicating the characteristics of each continuation predictor and the values of the tags T1 to Tm indicating the characteristics of the estrangement predictor, A degree of correlation indicating the degree of correlation with is calculated.

Specifically, for example, the search unit 704 uses the following formula (2 ) can be used to calculate the correlation coefficient C′ between the defection predictor and the continuation predictor. However, x _j is the value (0 or 1) of the tag Tj of the defection predictor. y _j is the value (0 or 1) of the tag Tj of the continuing predictor. μ _x is the average value of all tags T1 to Tm of the defection predictor. μ _y is the average value of the values of all tags T1 to Tm of the continuous predictor.

The search unit 704 extracts one or more continuous predictors from the identified continuous predictors based on the calculated degree of correlation of each continuous predictor. Specifically, for example, the search unit 704 extracts, from among the identified continuous predictors, continuous predictors whose calculated correlation coefficient C' is equal to or greater than a threshold. The threshold can be set arbitrarily, and is set to a value of about 0.4, for example. Further, for example, the search unit 704 may extract X consecutive predictors with the highest calculated correlation coefficient C' from among the identified consecutive predictors. X can be set arbitrarily, and is set to a value of about 3 to 5, for example.

Thus, by comparing the value patterns of the tags T1 to Tm of the estrangement predictor and each of the continuation predictors, it is possible to extract the continuation predictor having a strong correlation with the estrangement predictor. An example of extraction of continuous predictors will be described later with reference to FIG. 24 .

In the following description, a continuous predictor extracted from the identified continuous predictors may be referred to as a "continuous predictor (correlated)" in order to distinguish it from the identified continuous predictors.

The search unit 704 determines a simulation tag based on the extracted information of the tags T1 to Tm indicating the characteristics of one or more continuous predictors (correlated). Here, it can be said that the continuation predictor having a strong correlation with the estrangement predictor has a pattern of values of tags T1 to Tm similar to the estrangement predictor, and has a similar behavior pattern.

If the estrangement predictor takes an action that brings the behavior pattern closer to that of the continuation predictor (correlated), it is highly likely that estrangement can be prevented. For this reason, the searching unit 704 searches for tags that lead to measures that bring the behavior pattern of the estranged predictor closer to the behavior pattern of the continuation predictor (correlated).

Specifically, for example, for each tag Tj included in tags T1 to Tm, the search unit 704 accumulates the value of each tag Tj of the prediction data (after prediction) of each of the extracted continuous predictors (with correlation). By doing so, the total flag value α of each tag Tj is calculated. Then, based on the calculated total flag value α of each tag Tj, the searching unit 704 may determine K tags as simulation tags in descending order of the total flag value α among the tags T1 to Tm. . K can be set arbitrarily, and is set to a value of about 1 to 3, for example.

However, it is assumed here that a measure is taken to change the value of the measure target tag from "0" to "1". Also, tags related to basic attributes may be excluded from the simulation tags.

It should be noted that the flag total value α may be the same or about the same (for example, a difference of about ±1) between tags with a large flag total value α. is sometimes difficult to judge. Therefore, the searching unit 704 may calculate the total flag value β by accumulating the values of the tags Tj of the prediction data (after prediction) of each identified continuous predictor.

In this case, the searching unit 704 may determine K tags having a high total flag value α and a high total flag value β as simulation tags. More specifically, for example, when the search unit 704 selects K tags in descending order of flag total value α, if there are tags with the same or similar flag total value α, the flag total value Priority may be given to the one with the higher value β.

(Example of simulation tag determination)
Here, an example of determining a simulation tag will be described with reference to FIG. Here, the predictor of separation is "customer P1" and the predictors of continuation are "customers P3100 to P3103". Also, the tags T1 to Tm are referred to as "tags T1 to T6" (m=6).

FIG. 24 is an explanatory diagram showing an example of determination of simulation tags. In FIG. 24, the values of tags T1 to T6 and the defection flag "1" are shown for customer P1 who is the predictor of defection. Further, the values of tags T1 to T6 and the separation flag "0" are shown for each of the customers P3100 to P3103 who are the continuous predictors.

Further, the correlation coefficient C' between the customers P3100 to P3103 who are the continuation predictors and the customer P1 who is the defector predictor is shown. The values of the tags T1 to T6 and the values of the separation flags of the customers P1 and P3100 to P3103 are specified from the prediction data (after prediction) of the customers P1 and P3100 to P3103.

Here, the threshold value of the correlation coefficient C' is assumed to be "0.4". In this case, the searching unit 704 extracts customers whose correlation coefficient C' is 0.4 or more from customers P3100 to P3103 who are continuous predictors. Here, customers P3101 to P3103 are extracted as continuous predictors (with correlation).

Next, for each tag Tj included in tags T1 to T6, the search unit 704 accumulates the value of each tag Tj of each of the extracted continuous predictors (with correlation), thereby obtaining the total flag value α of each tag Tj. Calculate For example, the total flag value α of the tag T1 is "2".

Further, the searching unit 704 calculates the total flag value β of each tag Tj by accumulating the value of each tag Tj of each continuous predictor for each tag Tj included in the tags T1 to T6. For example, the total flag value β of tag T1 is "3".

Then, the searching unit 704 determines K tags having a high total flag value α and a high total flag value β among the tags T1 to T6 as simulation tags. However, it is assumed here that a measure is taken to change the value of the measure target tag from "0" to "1". In this case, the search unit 704 excludes the tag whose value of the defection predictor is "1" from the simulation tags.

Specifically, for example, the searching unit 704 determines that among the tags T1, T2, T4, T5, and T6 for which the value of the defection predictor (customer P1) is "0", the total flag value α is high and the flag N tags with a high total value β are determined as simulation tags. Here, the number K of simulation tags is assumed to be "1".

In this case, the searching unit 704 first identifies the tags T1 and T4 with the largest total flag value α among the tags T1, T2, T4, T5 and T6. Next, the search unit 704 determines the tag T1 having the higher total flag value β among the specified tags T1 and T4 as the simulation tag. As a result, the tag T1 having a high total flag value α and a high total flag value β can be determined as a simulation tag.

(Separation Prevention Support Processing Procedure of Information Processing Device 101)
Next, a separation prevention support processing procedure of the information processing apparatus 101 according to the fifth embodiment will be described.

25 and 26 are flowcharts showing an example of the separation prevention support processing procedure of the information processing apparatus 101 according to the fifth embodiment. In the flowchart of FIG. 25, first, the information processing apparatus 101 acquires learning data for year (X-1) from the learning data DB 220 (step S2501). Next, the information processing apparatus 101 generates a machine learning model M for determining whether or not the customer will defect in the next fiscal year based on the acquired learning data (step S2502).

Then, the information processing apparatus 101 uses the generated machine learning model M to execute defector prediction processing for predicting whether or not the customer will defect in the next year (step S2503). The specific processing procedure of the defector prediction process is the same as the procedure shown in FIG. 14, so illustration and description are omitted.

Next, the information processing apparatus 101 extracts the prediction data (after prediction) with the defection flag "0" for year X from the prediction data DB 230 (step S2504). Next, the information processing apparatus 101 selects, from the prediction data DB 230, the prediction data (post-prediction) of the prediction data (post-prediction) for the year X that has not been selected and has the unselected defection flag “1”. (Step S2505).

Then, the information processing apparatus 101, based on the selected separation flag "1" prediction data (post-prediction) and the extracted separation flag "0" prediction data (post-prediction), A correlation coefficient C′ with the continuous predictor is calculated (step S2506).

Next, the information processing apparatus 101 selects those whose correlation coefficient C′ is 0.4 or more among the continuous predictors specified from the prediction data (after prediction) with the separation flag “0” extracted in step S2504. are extracted (step S2507). Then, the information processing apparatus 101 calculates the total flag value α of each tag Tj by accumulating the value of each tag Tj in the prediction data (after prediction) of each of the extracted continuous predictors (correlated) ( step S2508).

Next, the information processing apparatus 101 calculates the total flag value β of each tag Tj by accumulating the value of each tag Tj in the prediction data (after prediction) with the separation flag “0” extracted in step S2504. (Step S2509).

Then, the information processing apparatus 101 rearranges the tags T1 to Tm based on the calculated total flag values α and β of each tag Tj and stores them in the list L (step S2510). Transition. At this time, the information processing apparatus 101 ranks flag total values α higher, and if flag total values α are the same, flag total values β higher.

In the flowchart of FIG. 26, the information processing apparatus 101 first determines whether or not the calculated maximum value α _max of the total flag value α of each tag Tj is "0" (step S2601). Here, if the maximum value α _max is "0" (step S2601: Yes), the information processing apparatus 101 proceeds to step S2603.

That is, if the value of all continuous predictors (with correlation) is "0" for all tags T1 to Tm, the information processing apparatus 101 determines that there is no tag to be a simulation tag, and step S2603. transition to

On the other hand, if the maximum value α _max is not "0" (step S2601: No), the information processing apparatus 101 executes a second simulation process (step S2602). The second simulation process changes the value of the simulation tag for the prediction data (post-prediction) with the defection flag "1" selected in step S2505, and simulates whether or not the customer will defect. be. A specific processing procedure of the second simulation processing will be described later with reference to FIG. 27 .

Next, the information processing apparatus 101 determines whether or not there is prediction data (post-prediction) with an unselected defection flag "1" that has not been selected among the prediction data (post-prediction) for year X ( step S2603). If there is unselected prediction data (post-prediction) (step S2603: Yes), the information processing apparatus 101 returns to step S2505.

On the other hand, if there is no unselected prediction data (step S2603: No), the information processing apparatus 101 outputs a separation countermeasure file created in the flowchart of FIG. end the processing of

This makes it possible to provide information on measures to prevent the next year's customers (target users) from leaving.

Concrete Processing Procedure of Second Simulation Processing Next, a concrete processing procedure of the second simulation processing in step S2602 shown in FIG. 26 will be described with reference to FIG.

FIG. 27 is a flow chart showing an example of a specific processing procedure of the second simulation processing. In the flowchart of FIG. 27, first, the information processing apparatus 101 sets the number of times of simulation K (step S2701). Note that the number of times of simulation K can be set arbitrarily, and is set to a value of about 1 to 3, for example.

Next, the information processing apparatus 101 sets the variable k to "k=1" (step S2702), and selects the highest tag from the unselected tags not selected from the list L (step S2703). Next, the information processing apparatus 101 refers to the prediction data (after prediction) selected in step S2505 shown in FIG. 25, and determines whether or not the value of the selected tag is "0" (step S2704).

Here, if the value of the tag is "1" (step S2704: No), the information processing apparatus 101 returns to step S2703 and selects the highest tag among the unselected tags that have not been selected from the list L. select. On the other hand, if the value of the tag is "0" (step S2704: YES), the information processing apparatus 101 refers to the tag master 240 and determines whether the action possible flag of the selected tag is "1". (Step S2705).

Here, if the action possible flag is "0" (step S2705: No), the information processing apparatus 101 returns to step S2703 and selects the highest tag from among the unselected tags that have not been selected from the list L. select. On the other hand, if the action possible flag is "1" (step S2705: Yes), the selected tag is determined as a simulation tag (step S2706).

Then, the information processing apparatus 101 changes the value of the simulation tag of the selected prediction data (after prediction) from "0" to "1" (step S2707). Next, the information processing apparatus 101 uses the generated machine learning model M to determine whether the customer will depart in the (X+1) year based on the changed prediction data (after prediction) in which the value of the simulation tag is changed. It is determined whether or not (step S2708).

Then, the information processing apparatus 101 determines whether or not it is determined that the customer will leave (step S2709). Here, if it is determined that the customer will leave (step S2709: Yes), the information processing apparatus 101 increments "k" (step S2710) and determines whether "k" is greater than "K". (step S2711).

Here, if "k" is equal to or less than "K" (step S2711: No), the information processing apparatus 101 returns to step S2703 and selects the highest tag among the unselected tags that have not been selected from the list L. to select.

On the other hand, if "k" is greater than "K" (step S2711: Yes), the information processing apparatus 101 returns to the step that called the second simulation process. In this case, the defection countermeasure file will be information in which no tag name is registered for the customer of the selected prediction data (post-prediction).

If it is determined in step S2709 that the customer will not leave (step S2709: No), the information processing apparatus 101 associates the customer ID of the selected prediction data (after prediction) with the tag name of the determined simulation tag. and registers it in the separation countermeasure file (step S2712), and returns to the step that called the second simulation process.

As a result, a simulation tag that reverses the determination result of customer defection can be searched for as a target of measures for preventing customer defection. A maximum of K tag name combinations are registered in the estrangement countermeasure file in association with customer IDs.

As described above, according to the information processing apparatus 101 according to the fifth embodiment, based on the determination result of whether each user will leave can be specified. In addition, according to the information processing apparatus 101, for each of the specified continuous predictors, the values of tags T1 to Tm indicating the characteristics of each continuous predictor and the values of tags T1 to Tm indicating the characteristics of the defection predictor (target user) A degree of correlation (for example, correlation coefficient C′) that indicates the degree of correlation with the defector predictor can be calculated based on the value. Then, according to the information processing apparatus 101, one or more continuous predictors (correlated) are extracted from the identified continuous predictors based on the calculated degree of correlation of each continuous predictor, and the extracted one Alternatively, a simulation tag can be determined based on the values of tags T1 to Tm that indicate characteristics of multiple continuous predictors (correlated).

As a result, tags with a high rate of "1" among continuous predictors who have a strong correlation with defection predictors can be selected as simulation tags, leading to effective measures to prevent customer defection. Tags can be searched efficiently.

From these, according to the information processing apparatus 101 according to the present embodiment, by utilizing the information tagged to the user, it is possible to narrow down the target to the customers who are likely to defect, and to take appropriate measures, Efficient and effective marketing can be realized.

The separation prevention support method described in this embodiment can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. This separation prevention support program is recorded in a computer-readable recording medium such as a hard disk, flexible disk, CD-ROM, DVD, USB memory, etc., and executed by being read from the recording medium by a computer. Also, the separation prevention support program may be distributed via a network such as the Internet.

Further, the information processing apparatus 101 described in the present embodiment can also be realized by application specific ICs such as standard cells and structured ASICs (Application Specific Integrated Circuits), and PLDs (Programmable Logic Devices) such as FPGAs.

Further, the following additional remarks are disclosed with respect to the above-described embodiment.

(Appendix 1) determining whether or not the user will leave after the predetermined period based on the information of a plurality of tags indicating the characteristics of the user during the predetermined period;
Departure of the user from the plurality of tags based on a determination result as to whether or not the user will depart when the information of at least one of the tags corresponding to the user determined to depart is changed. search for tags associated with measures to prevent
outputting the searched search result or information specifying the search result in association with the user;
A separation prevention support program characterized by causing a computer to execute processing.

(Appendix 2) The searching process is
The alienation prevention support program according to appendix 1, wherein a tag that changes a judgment result of the alienation of the user is searched as the tag related to the policy.

(Appendix 3) The searching process is
tags related to the measures based on the determination result of whether or not the user will depart when the information of at least one of the plurality of tags excluding the tag relating to the basic attributes of the user is changed; The separation prevention support program according to appendix 1 or 2, characterized by searching.

(Appendix 4) Based on the information of the plurality of tags indicating the characteristics of each user of the plurality of users in a predetermined period, it is determined whether or not each user will leave after the predetermined period,
causing the computer to execute a process of determining a tag to be processed from the plurality of tags based on the determination result as to whether or not each of the users will depart,
The searching process includes
Searching for a tag related to the policy based on a determination result as to whether or not the user will leave when the information of at least one of the determined tags to be processed is changed. Separation prevention support program according to any one of Appendices 1 to 3.

(Appendix 5) A degree of correlation indicating the degree of correlation between each tag included in the plurality of tags and the determination result of defection of each user, based on the determination result of whether or not each user defected. calculating, causing the computer to execute a process;
The process of determining
The separation prevention support program according to appendix 4, wherein the tag to be processed is determined based on the calculated degree of correlation of each tag.

(Appendix 6) Cause the computer to execute a process of identifying a continuation predictor determined not to depart from among the plurality of users based on the determination result of whether or not each of the users will depart,
The process of determining
6. The separation prevention support program according to appendix 4 or 5, wherein the tag to be processed is determined based on the information of the plurality of tags indicating the characteristics of the identified continuous predictor.

(Appendix 7) The process of determining
For each tag included in the plurality of tags, the tag to be processed is determined based on the ratio of users who responded to the measure among a group of users for whom the measure related to each tag was executed. A separation prevention support program according to any one of appendices 4 to 6.

(Additional remark 8) Based on the determination result of whether or not each of the determined users will leave, identify the continuous predictor determined not to leave among the plurality of users,
For each of the identified continuous predictors, the degree of correlation with the user based on the information of the plurality of tags indicating characteristics of each continuous predictor and the information of the plurality of tags indicating characteristics of the user. Calculate the degree of correlation indicating
causing the computer to execute a process of extracting one or more continuous predictors from among the identified continuous predictors based on the calculated degree of correlation of each of the continuous predictors;
The process of determining
8. The tag to be processed is determined based on the extracted information of the plurality of tags indicating the characteristics of the one or more continuous predictors. Separation prevention support program.

(Appendix 9) For each of a plurality of users, the information of the plurality of tags indicating the characteristics of each user in a specific period, and the defection result indicating whether or not each user has defected after the specific period causing the computer to execute a process of generating a machine learning model for determining whether or not to deviate from the information of the plurality of tags by referring to a storage unit that stores and
The estrangement prevention support program according to any one of appendices 1 to 8, wherein the determination of whether or not the user will estrange is performed using the machine learning model.

(Appendix 10) Any one of Appendices 1 to 9, wherein the information of the plurality of tags indicates a value according to whether or not each tag included in the plurality of tags satisfies a condition expressed by the tag. the alienation prevention assistance program described in .

(Appendix 11) The withdrawal prevention support program according to any one of appendices 1 to 10, characterized in that the withdrawal of the user is withdrawal from a product or service used by the user.

(Appendix 12) Based on the information of a plurality of tags indicating the characteristics of the user in a predetermined period, it is determined whether the user will leave after the predetermined period,
Departure of the user from the plurality of tags based on a determination result as to whether or not the user will depart when the information of at least one of the tags corresponding to the user determined to depart is changed. search for tags associated with measures to prevent
outputting the searched search result or information specifying the search result in association with the user;
A separation prevention support method, characterized in that the processing is executed by a computer.

(Appendix 13) A determination unit that determines whether or not the user will depart after the predetermined period of time based on information of a plurality of tags indicating characteristics of the user during the predetermined period of time;
from the plurality of tags based on the determination result as to whether or not the user will depart when the information of at least one of the plurality of tags corresponding to the user determined to depart by the determination unit is changed; , a search unit that searches for tags related to measures to prevent the user from leaving;
an output unit that outputs a search result searched by the search unit or information specifying the search result in association with the user;
An information processing device comprising:

101 information processing device 110 user 120, 1201, 1202, 1203 separation prevention tag information 200 information providing system 201 client device 220 learning data DB
230 prediction data DB
240 Tag master 300 Bus 301 CPU
302 memory 303 disk drive 304 disk 305 communication I/F
306 portable recording medium I/F
307 portable recording medium 701 acquisition unit 702 generation unit 703 determination unit 704 search unit 705 output unit 900 setting file 1000 defection countermeasure file 1200 policy support screen 1700 correlation coefficient table 1900 flag total value table 2100 policy master 2200 reaction rate table

Claims (10)

Determining whether or not the user will leave after the predetermined period based on information of a plurality of tags indicating characteristics of the user during the predetermined period,
Based on the determination result of whether or not the user will leave when the information of at least one of the plurality of tags corresponding to the user determined to leave , excluding tags related to basic attributes of the user, is changed , from the plurality of tags, search for tags related to measures to prevent the user from leaving,
outputting the searched search result or information specifying the search result in association with the user;
A separation prevention support program characterized by causing a computer to execute processing. The searching process includes:
2. The alienation prevention support program according to claim 1, wherein a tag that changes a judgment result of said user's alienation is searched as a tag related to said measure. Based on the information of the plurality of tags indicating the characteristics of each user of the plurality of users in a predetermined period of time, determining whether each user will leave after the predetermined period of time,
determining a tag to be processed from among the plurality of tags, excluding tags related to basic attributes of each user, based on the determined result of whether or not each user will leave the computer; and run
The searching process includes:
Searching for a tag related to the policy based on a determination result as to whether or not the user will leave when the information of at least one of the determined tags to be processed is changed. The separation prevention support program according to claim 1 or 2. Correlation of each tag among the plurality of tags, excluding tags related to the basic attributes of each user, with the determination result of the separation of each user, based on the determination result of whether or not each user will depart causing the computer to execute a process of calculating the degree of correlation indicating the degree of
The process of determining
4. The separation prevention support program according to claim 3, wherein the tag to be processed is determined based on the calculated degree of correlation of each tag. cause the computer to execute a process of identifying a continuation predictor determined not to depart from among the plurality of users based on the determination result of whether or not each determined user will depart,
The process of determining
The tag to be processed is determined based on the information of the plurality of tags indicating the characteristics of the identified continuous predictor, excluding the tag related to the basic attribute of each user. The alienation prevention support program according to 3 or 4. The process of determining
Among the plurality of tags, for each tag excluding the tag related to the basic attribute of each user, among the user group for whom the measure related to each tag was executed, based on the ratio of users who responded to the measure, 6. The separation prevention support program according to any one of claims 3 to 5, wherein a tag to be processed is determined. Based on the determination result of whether or not each of the determined users will depart, identifying the continuous predictor determined not to depart from among the plurality of users,
For each of the identified continuous predictors, the degree of correlation with the user based on the information of the plurality of tags indicating characteristics of each continuous predictor and the information of the plurality of tags indicating characteristics of the user. Calculate the degree of correlation indicating
causing the computer to execute a process of extracting one or more continuous predictors from among the identified continuous predictors based on the calculated degree of correlation of each of the continuous predictors;
The process of determining
The tag to be processed is determined based on the information of the plurality of extracted tags indicating the characteristics of the one or more continuous predictors, excluding tags related to the basic attributes of each user. The separation prevention support program according to any one of claims 3 to 6. For each of a plurality of users, information of the plurality of tags indicating characteristics of each user during a specific period and a defection result indicating whether or not each user defected after the specific period is stored. causing the computer to execute a process of referring to the storage unit and generating a machine learning model for determining whether or not the information of the plurality of tags is deviated;
Determination of whether the user leaves is performed using the machine learning model,
8. The separation prevention support program according to any one of claims 1 to 7, wherein said plurality of tags include a tag relating to each user's behavior and a tag relating to each user's basic attributes. Determining whether or not the user will leave after the predetermined period based on information of a plurality of tags indicating characteristics of the user during the predetermined period,
Based on the determination result of whether or not the user will leave when the information of at least one of the plurality of tags corresponding to the user determined to leave, excluding tags related to basic attributes of the user, is changed , from the plurality of tags, search for tags related to measures to prevent the user from leaving,
outputting the searched search result or information specifying the search result in association with the user;
A separation prevention support method, characterized in that the processing is executed by a computer. a determination unit that determines whether or not the user will depart after the predetermined period based on information of a plurality of tags indicating characteristics of the user during the predetermined period;
Determining whether or not the user will depart when the information of at least one of the plurality of tags corresponding to the user determined to depart by the determination unit, excluding tags related to basic attributes of the user, is changed a search unit that searches for a tag related to a measure for preventing the user from leaving the plurality of tags based on the determination result;
an output unit that outputs a search result searched by the search unit or information specifying the search result in association with the user;
An information processing device comprising:

Images