JP2015014828A - Withdrawal estimation device, cancellation estimation device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately estimate a withdrawal time of a primary phenomenon.SOLUTION: A distribution parameter estimation section 24 updates a distribution parameter of a primary phenomenon on the basis of each observation duration and information for a plurality of individuals i, and updates a distribution parameter of each secondary phenomenon op on the basis of each observation duration, information and lapsed time for the plurality of individuals i. A first main service cancellation time estimation section 26 estimates a withdrawal time of the primary phenomenon in the individual i of an estimation object according to a distribution of a withdrawal probability based on the distribution parameter and the observation duration of the primary phenomenon, and a distribution of a withdrawal probability based on the distribution parameter, the observation duration and the lapsed time of each secondary phenomenon op withdrawing the individual of the estimation object, on the basis of the distribution parameter of the primary phenomenon and the distribution parameter of each secondary phenomenon op, the observation duration for the individual i of the estimation object of which the primary phenomenon is observed, the observation duration of each secondary phenomenon op, and the information of each secondary phenomenon op and the lapsed time of each secondary phenomenon op.

Description

  The present invention relates to a departure estimation device that estimates the timing of leaving a main phenomenon, a cancellation estimation device, a method, and a program that estimate the timing to cancel a main service.

  There are two types of services, a main service and an optional service subordinate to the main service (one main service, one or more optional service types, hereinafter referred to as “composite service”). An example of a composite service is an insurance service. In the case of insurance, a basic contract of insurance is a main service and an accompanying special contract is an optional service.

  In such a complex service, the main service contract period τ (the elapsed time from the main service contract start time to the cancellation time of each user) is estimated based on the option service cancellation time, and the change in the number of main service contractors Considering simulation (hereinafter, the main service contract time of a user i is called τ (i, basic)) as problem 1.

  In addition, if a model that can simulate the transition of the main service contract period τ (i, basic) and the number of main service contractors based on the cancellation timing of the optional service is established, when some optional services are abolished, the main service contract The problem 2 is to simulate how the period τ (i, basic) and the number of main service contractors change.

Previously, for the above problem 1 “Estimation of contract period τ (i, basic) and simulation of changes in the number of main service subscribers based on the option service cancellation period”, the contract between main service and option service The data of the situation and the past contract period is input to some model (for example, logit / Cox (Non-patent Document 1), HMM (Non-patent Document 2), etc.), and the contract period τ (i, basic) is output as the model. And the estimation result of the number of main service subscribers.

In addition, regarding the above problem 2 “Estimation of elapsed time τ (i, basic) and simulation of the number of main service subscribers when the service lineup is abolished based on the cancellation timing of the optional service”, the main service and optional service When the optional service is abolished through some model as in Problem 1 above, after defining the relationship between the main service and optional service through a separate questionnaire survey or market survey in addition to the contract status and past contract period data The impact on the number of primary service subscribers is estimated.

"Practice 3 Survival Analysis", [online] [searched on June 24, 2013], Internet <URL: http://www.med.nagoya-u.ac.jp/spss/files/jissen3.pdf> David A. Schweidel, Eric T. Bradlow, Peter S. Fader. Portfolio Dynamics for Customers of a Multiservice Provider, MANAGEMENT SCIENCE, Vol. 57, No. 3, pp. 471-486, March 2011,

  However, with regard to the above problem 1, in the conventional technology, the influence on the main service cancellation due to the option service cancellation is always constant (logit / Cox) after the option service cancellation, or appears only in a short time immediately after (HMM). In the conventional method, for example, there is a problem that it is not possible to deal with a case where the influence period of the option service cancellation on the main service cancellation is long. For example, when considering maintenance of equipment, events that cause equipment failures often occur at relatively long intervals, but conventional methods cannot cope with this problem.

  In addition, regarding the above problem 2, in the conventional technology, it is necessary to define the relationship between the main service and the optional service through a separate questionnaire survey or market survey, so additional costs are incurred, and the analysis results vary greatly depending on the investigator. There is a problem.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide a departure estimation apparatus, method, and program capable of accurately estimating the departure time of the main phenomenon.

  It is another object of the present invention to provide a cancellation estimation apparatus, method, and program capable of accurately estimating the cancellation timing of the main service.

In order to achieve the above object, a departure estimation apparatus according to a first aspect of the present invention is a departure estimation apparatus that estimates the distribution of departure probabilities of a main phenomenon that occurs in an individual and has a sub-phenomenon having a main-sub relationship. And the observation continuation period dr (i, basic) collected for each of a plurality of individuals i until the observation end time, and the sub phenomenon until the observation end time for each sub phenomenon op. an observation continuation period dr (i, op) in which op is continued, information c (i, basic) indicating whether or not the main phenomenon is observed at the end of the observation, and at the end of the observation for each sub phenomenon op. Information c (i, op) indicating whether or not the sub phenomenon is observed, and an elapsed time st () from the start time of the main phenomenon to the start time of the sub phenomenon for each sub phenomenon op , Op), an initial value of a distribution parameter representing the distribution of the probability of withdrawal of the main phenomenon at each time, and the main at each time from the withdrawal of the sub phenomenon for each sub phenomenon op An initial value setting means for setting an initial value of a distribution parameter representing the distribution of the probability of withdrawal of the phenomenon, the observation duration dr (i, basic) of each of the plurality of individuals i acquired by the data acquisition means, and the Based on the information c (i, basic), the distribution parameter of the main phenomenon is updated, and the observation duration dr (i, op) of each of the plurality of individuals i acquired by the data acquisition unit, Distribution parameter updating means for updating the distribution parameter of each sub phenomenon op based on the information c (i, op) and the elapsed time st (i, op); Until satisfied because a defined iteration termination condition, and repeats determination means for repeating the updating by the distribution parameter updating means,
The distribution parameters of the main phenomenon updated by the distribution parameter updating means, the distribution parameters of each sub-op, and the observation duration dr (i, i, for the estimation target individual i where the main phenomenon is observed. basic), the observation duration dr (i, op) of each sub-op, the information c (i, op) of each sub-op, and the elapsed time st (i, op) of each sub-op. The distribution probability distribution based on the distribution parameter of the main phenomenon and the observation continuation period dr (i, basic), the distribution parameter of each sub-op that has departed in the individual to be estimated, the observation continuation According to the period dr (i, op) and the distribution of the probability of withdrawal based on the elapsed time st (i, op), the withdrawal timing of the main phenomenon in the individual i to be estimated Is configured to include a withdrawal period estimating means for estimating, the.

  The withdrawal estimation method of the second invention includes a data acquisition means, an initial value setting means, a distribution parameter update means, an iterative determination means, and a withdrawal timing estimation means, and is a main phenomenon that occurs in an individual, A departure estimation method in a departure estimation apparatus for estimating a distribution of a departure probability of a main phenomenon having a slave phenomenon having a master-slave relationship, wherein the data acquisition means collects each of a plurality of individuals i until an observation end point. The observation continuation period dr (i, basic) in which the main phenomenon has continued, the observation continuation period dr (i, op) in which the sub phenomenon op has continued until the observation end time for each sub phenomenon op, and the observation end point Information c (i, basic) indicating whether or not the main phenomenon is observed, and whether or not the sub phenomenon is observed at the end of the observation for each sub phenomenon op. The information c (i, op) and the elapsed time st (i, op) from the start time of the main phenomenon to the start time of the sub phenomenon for each sub phenomenon op are acquired, and the initial value setting means Is an initial value of a distribution parameter that represents the distribution of the probability of departure of the main phenomenon at each time, and a distribution parameter that represents the distribution of the probability of departure of the main phenomenon at each time from the departure of the sub phenomenon for each sub phenomenon op The distribution parameter updating unit sets the observation duration period dr (i, basic) and the information c (i, basic) of each of the plurality of individuals i acquired by the data acquiring unit. On the basis of the distribution parameter of the main phenomenon, the observation duration dr (i, op) of each of the plurality of individuals i acquired by the data acquisition means, The distribution parameter of each sub-op is updated based on the information c (i, op) and the elapsed time st (i, op), and the iterative determination means satisfies a predetermined iteration end condition. Until the distribution parameter update unit repeats the update, the departure time estimation unit updates the distribution parameter of the main phenomenon updated by the distribution parameter update unit and the distribution parameter of each sub-op, and the main phenomenon , The observation duration dr (i, basic) for the estimation target individual i, the observation duration dr (i, op) of each sub-op, and the information c (i, op) of each sub-op op) and the elapsed time st (i, op) of each sub-op, based on the distribution parameter of the main phenomenon and the observation duration dr (i, basic). And the distribution parameter of each sub-op that has departed in the individual to be estimated, the separation based on the observation duration dr (i, op), and the elapsed time st (i, op) According to the probability distribution, the departure time of the main phenomenon in the individual i to be estimated is estimated.

  According to the first invention and the second invention, it is possible to accurately predict the departure time of the main phenomenon.

  A churn estimation apparatus according to a third aspect of the present invention is a churn estimation apparatus for estimating a churn probability distribution of a main service having an optional service, the main service collected for each of a plurality of users i up to the end of observation. The observation continuation period dr (i, op), the observation service continuation period dr (i, op) that the option service op continues until the observation end time for each option service op, and the main service at the observation end point C (i, basic) indicating whether or not the option service is contracted, information c (i, op) indicating whether or not the option service is contracted at the end of the observation for each option service op, and each From the start of the contract for the main service for the optional service op, Data acquisition means for acquiring the elapsed time st (i, op) up to about the start time, initial values of distribution parameters representing the distribution of the churn probability of the main service at each time, and the options for each optional service op Initial value setting means for setting an initial value of a distribution parameter representing the distribution of the probability of cancellation of the main service at each time from service cancellation; and the observation of each of the plurality of users i acquired by the data acquisition means Based on the duration dr (i, basic) and the information c (i, basic), the distribution parameter of the main service is updated, and the observation of each of the plurality of users i acquired by the data acquisition unit Based on the duration period dr (i, op), the information c (i, op), and the elapsed time st (i, op) A distribution parameter update unit that updates the distribution parameter of each option service op, an iterative determination unit that repeats the update by the distribution parameter update unit until a predetermined iteration end condition is satisfied, and the distribution parameter update unit The updated distribution parameter of the main service and the distribution parameter of each option service op, the observation duration dr (i, basic) for the estimation target user i who subscribes to the main service, and each option service Based on the observation duration dr (i, op) of op, the information c (i, op) of each option service op, and the elapsed time st (i, op) of each option service op The distribution parameter and the observation duration dr (i, basic) And the distribution parameters of each option service op canceled by the user i to be estimated, the observation duration dr (i, op), and the elapsed time st (i, op) And a cancellation timing estimating means for estimating the cancellation timing of the main service in the estimation target user i according to the distribution of the cancellation probability based on.

The churn estimation method of the fourth invention includes a data acquisition means, an initial value setting means, a distribution parameter update means, an iterative determination means, and a churn time estimation means, and the churn probability of the main service having an optional service. A churn estimation method in a churn estimation device for estimating a distribution, comprising:
The data acquisition means collects each of a plurality of users i, the observation continuation period dr (i, basic) that the main service has continued until the observation end point, and the observation end point for each optional service op. The observation continuation period dr (i, op) in which the optional service op has continued, the information c (i, basic) indicating whether or not the main service is contracted at the end of the observation, and the observation for each optional service op Information c (i, op) indicating whether or not the option service is contracted at the end time, and the elapsed time st from the contract start time of the main service to the start time of the option service contract for each option service op (I, op) and the initial value setting means at each time The initial value of the distribution parameter representing the distribution of the churn probability of the main service and the initial value of the distribution parameter representing the distribution of the churn probability of the main service at each time from the cancellation of the option service for each option service op are set. Then, the distribution parameter updating unit is configured to determine the main parameter based on the observation duration dr (i, basic) and the information c (i, basic) of each of the plurality of users i acquired by the data acquisition unit. The distribution parameter of the service is updated, and the observation duration dr (i, op), the information c (i, op), and the elapsed time st of each of the plurality of users i acquired by the data acquisition unit Based on (i, op), the distribution parameter of each option service op is updated, and the iterative determination means The update by the distribution parameter update unit is repeated until a predetermined iteration end condition is satisfied, and the cancellation timing estimation unit is configured to update the distribution parameter of the main service updated by the distribution parameter update unit and each optional service. the distribution parameters of op, the observation duration dr (i, basic) for the estimation target user i contracted with the main service, the observation duration dr (i, op) of each optional service op, Based on the information c (i, op) of the option service op and the elapsed time st (i, op) of each option service op, the distribution parameters of the main service and the observation duration dr (i, basic) Churn probability distribution based on the above and each of the churns in the estimation target user i Canceling the main service for the user i to be estimated according to the distribution parameter of the optional service op, the observation duration dr (i, op), and the distribution of the cancellation probability based on the elapsed time st (i, op) Estimate the time.

  According to the third and fourth inventions, it is possible to accurately estimate the cancellation time of the main service.

  In the third invention, the churn time estimation means churns the distribution probability distribution based on the distribution parameter of the main service and the observation duration dr (i, basic) and the estimation target user i. Distribution of churn probability based on the distribution parameter of each option service op, the observation duration dr (i, op), and the elapsed time st (i, op), and the option service op to be abolished , According to the distribution parameter of each option service op continuing in the estimation target user i, the observation duration period dr (i, op), and the distribution of the churn probability based on the elapsed time st (i, op) The departure time of the main service for the user i to be estimated may be estimated.

  Moreover, the program of this invention is a program for functioning a computer as each means which comprises said departure estimation apparatus or cancellation estimation apparatus.

  As described above, according to the departure estimation apparatus, method, and program, it is possible to accurately predict the departure timing of the main phenomenon.

  Further, according to the cancellation estimation apparatus, method, and program, it is possible to accurately estimate the cancellation timing of the main service.

It is a block diagram which shows the functional structure of the cancellation estimation apparatus in embodiment of this invention. It is a figure which shows the example of contract information. It is a figure which shows the example of an output result. It is a flowchart which shows the cancellation estimation process routine in the cancellation estimation apparatus which concerns on embodiment of this invention. It is a flowchart which shows the distribution parameter (lambda) (basic) estimation processing routine in the cancellation estimation apparatus which concerns on embodiment of this invention. It is a flowchart which shows the distribution parameter (lambda) (x) estimation processing routine in the cancellation estimation apparatus which concerns on embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<Principle of the present invention>
The principle of the present invention will be described. In the embodiment of the present invention, for each user, the main service information (contract continuation period and final contract state) and optional service contract information (time when contracted after the start of the main service contract) at a certain observation end time , Contract duration, and final contract status), and using the collected data, the temporal correlation between the main service and the optional service is estimated as a distribution model. The contract for the main service is an example of the main phenomenon, and the contract for the option service is an example of the sub phenomenon.

  Based on the estimation result, the correlation between the cancellation timing from the start of the optional service contract and the cancellation timing of the main service contract, and the cancellation timing from the start of the main service contract can be estimated. Realize transition simulation of the number of people.

  Similarly, when the optional service is abolished, similarly, using the distribution model of the main service and the optional service corresponding to the collected data, the cancellation time τ (i, basic) from the contract start of the main service contractor, and Realize the transition simulation of the number of main service subscribers.

<Configuration of cancellation estimation device>
Next, the configuration of the churn estimation apparatus according to the embodiment of the present invention will be described. As shown in FIG. 1, a churn estimation device 100 according to an embodiment of the present invention is a computer that includes a CPU, a RAM, and a ROM that stores a program for executing a churn estimation processing routine (to be described later) and various data. Can be configured. The churn estimation apparatus 100 functionally includes an input unit 10, a calculation unit 20, and an output unit 50 as shown in FIG.

  The input unit 10 receives a combination of an estimation target user group, an estimation target composite service, and an optional service for performing abolition simulation from an input device such as a keyboard. Note that the input unit 10 may accept input from the outside via a network or the like.

  The computing unit 20 includes a storage unit 22, a distribution parameter estimation unit 24, a first main service cancellation timing estimation unit 26, a first main service contractor number transition prediction unit 28, and a second main service cancellation timing estimation unit 30. And a second main service contractor number transition prediction unit 32 and a result generation unit 34.

  The storage unit 22 stores information related to the composite service for each user, that is, the contract continuation period (dr (i, basic)) of the main service for each user, and the contract continuation period of each optional service to the present. (Dr (i, op1), dr (i, op2), ..., dr (i, opX)), the elapsed time st (st (i, op1), st ( i, op2), ..., st (i, opX)), current contract status (contracted / cancelled) (main service contract status c (i, basic), option service contract status c (i, op1) ), c (i, op2),..., c (i, opX)), and other information related to the user (such as the area where they live) are recorded.

  Here, the contract continuation period dr up to the present time indicates the period from the contract start date to the contract end date if the current contract status has been cancelled, and if the current contract status is in contract, the contract continuation period dr The elapsed time up to the date of. i represents a user's ID number and is a natural number.

  Also, the contract continuation period dr (dr (i, basic), dr (i, op1), dr (i, op2), ..., dr (i, opX)) and the elapsed time st (st (i, opX)) op1), st (i, op2), ..., st (i, opX)) take a scalar value, and the current contract status c (c (i, basic), c (i, op1), c (i, op2 ),..., C (i, opX)) take a binary value of 0 and 1. The contract status c indicates “contracted” when 0, and “cancelled” when 1. An example is shown in FIG. Note that X is the maximum value of the number of option services, and it is assumed that the value of the contract status c of the option service that is not contracted or canceled does not have any value of 0 or 1.

  The distribution parameter estimation unit 24 determines the contract duration period dr (dr (i, basic), dr for the current time for each of the users to be estimated among the user groups of the complex service to be estimated received by the input unit 10. (i, op1), dr (i, op2), ..., dr (i, opX)), the elapsed period from the start of the main service contract to the start of each optional service contract st (st (i, op1), st (i , op2), ..., st (i, opX)), current contract status c (c (i, basic), c (i, op1), c (i, op2), ..., c (i, opX)) Are respectively read from the storage unit 22, and distribution parameters representing the distribution of the churn probability at each time of the main service and distribution parameters representing the distribution of the main service churn probability at each time from the cancellation of each optional service are estimated. The cancellation period τ (i, basic) based on the contract start of the main service follows the distribution of the cancellation probability at each time of the main service. In addition, the main service cancellation timing end (end (i, op1), end (i, op2), ..., end (i, opX)) based on the cancellation of each optional service is the amount from the cancellation of each optional service. It follows the distribution of the main service churn probability at each time. In this embodiment, the termination time τ (i, basic) based on the contract start of the main service, and the termination timing end (end (i, op1), end of the main service based on each optional service cancellation Each of (i, op2), ..., end (i, opX)) is assumed to follow a distribution having different distribution parameters. In the distribution of churn probability, the horizontal axis represents time, and the ordinate represents the main service churn probability.

  Here, the distribution of the main service cancellation period end (end (i, op1), end (i, op2), ..., end (i, opX)) based on the cancellation of each optional service is changed. Thus, the problem in Problem 1 can be solved. For example, in the case of “the case where the influence period of the optional service cancellation on the main service cancellation is long” mentioned in the above problem 1, it is assumed that the attenuation of the influence degree is constant in all periods, and the influence depends on the magnitude of the index distribution parameter. It is possible to cope with this by setting a model that can handle both the length and shortness of the model in the distribution.

  Specifically, in this embodiment, the termination time τ (i, basic) based on the contract start of the main service, and the termination timing end (end (i, op1) of the main service based on the termination of each optional service , End (i, op2), ..., end (i, opX)) follow the distributions represented by the exponential distribution parameters λ (basic), λ (1), λ (2), ..., λ (X), respectively. And each exponential distribution parameter is estimated according to a Markov chain Monte Carlo method (MCMC). The distribution parameter λ (basic) is an example of the distribution parameter of the main phenomenon, and the distribution parameters λ (1), λ (2)..., Λ (X) are examples of the distribution parameter of the sub phenomenon.

  The first main service churn time estimation unit 26 is a distribution parameter (λ (basic)) representing the distribution of the churn probability of the main service estimated by the distribution parameter estimator 24, and the churn probability of the main service from each optional service churn. Each of the distribution parameters (λ (1), λ (2),... Λ (X)) representing the distribution of, and the target user (c (i, basic)) whose main service contract status is “Contracting” = 0) is estimated based on each contract information (dr, st, c), and the cancellation time of the main service of the user to be estimated. For the user whose main service contract status is “cancelled” (c (i, basic) = 1), the cancellation timing is not estimated because the cancellation timing is known.

  Specifically, for each user to be estimated, the churn probability of the main service obtained from the distribution parameter (λ (basic)) estimated by the distribution parameter estimation unit 24 and the contract continuation period dr (i, basic). Distribution (exponential distribution), estimated distribution parameters (λ (1), λ (2),…, λ (X)), contract duration (dr (i, op1), dr (i, op2),… , Dr (i, opX)) and elapsed time (st (i, op1), st (i, op2),..., St (i, opX)), each optional service (c ( i, p) = 1; p = op1, op2,..., opX) are overlaid in consideration of the churn probability distribution (exponential distribution) of the main service in consideration of the churn time of the optional service. Then, for each time t, a new distribution of churn probability is created in which the churn probability of the main service at time t is the maximum value among the churn probabilities at the time t of each original distribution.

  For each estimation target user, the first main service cancellation timing estimation unit 26 samples and determines whether to cancel at each time t based on the new distribution of the cancellation probability of the user. Estimate the cancellation time of the user's main service. For the user whose main service contract status is “cancelled” (c (i, basic) = 1), since the cancellation timing is known, the cancellation timing is not estimated.

  The first main service contractor number transition prediction unit 28 predicts the transition of the number of contractors of the main service based on the cancellation timing of each estimation target user estimated by the first main service cancellation time estimation unit 26.

  The second main service churn time estimation unit 30 receives a distribution parameter (λ () representing a distribution of the churn probability of the main service estimated by the distribution parameter estimation unit 24 and a combination of optional services that are received by the input unit 10 for performing abolition simulation. basic)), distribution parameters (λ (1), λ (2),..., λ (X)) of the cancellation probability of the main service from each optional service cancellation, and the contract status of the main service are “under contract” ”Is estimated based on the contract information (dr, st, c) of each estimation target user. For the user whose main service contract status is “cancelled” (c (i, basic) = 1), the cancellation timing is not estimated because the cancellation timing is known.

  Specifically, for each user to be estimated, it is determined whether or not at least one of the optional services under contract is included in the combination of optional services for which the abolition simulation received by the input unit 10 is included. If all of the optional services under contract are not included in the combination of optional services for which the abolition simulation is performed, the cancellation time of the main service of the user is set as the first main service cancellation time estimation unit 26 described above. Estimate in the same way.

  On the other hand, when at least one of the optional services under contract is included in the combination of optional services for which abolition simulation is performed, a distribution representing the distribution of the churn probability of the main service estimated by the distribution parameter estimation unit 24 Each option service (c (i, p) = c) is canceled by the user in the distribution (exponential distribution) of the cancellation probability of the main service obtained from the parameter (λ (basic)) and the contract duration period dr (i, basic). 1; each of the distributions (exponential distributions) of the churn probability of the main service from the churn of p = op1, op2,..., OpX) is superposed in consideration of the churn time of the optional service. Then, for each time t, a new churn probability distribution D_con is created in which the churn probability of the main service at the time t is the maximum value among the churn probabilities at the time t of each original distribution.

  Next, cancel the optional services (c (i, p) = 0; p = op1, op2,…, opX) for which the user is contracting, with the current time being the cancellation date. Each of the distributions (exponential distributions) representing the distribution of the probability of churn of the main service from is overlaid in consideration of the contract continuation period (dr (i, p)) and the elapsed period (st (i, p)). Then, for each time t, a new churn probability distribution D_churn is created in which the churn probability of the main service at the time t is the maximum value among the churn probabilities at the time t of each original distribution.

  Then, the newly created churn probability distributions D_con and D_churn are overlapped, and for each time t, the churn probability of the main service at time t is set to the maximum value among the churn probabilities at time t of the original distributions. A new churn probability distribution D is created.

  The second main service churn time estimation unit 30 creates, for each user included in the combination of option services for which at least one of the optional services under contract performs an abolition simulation among the users to be estimated. Based on the new churn probability distribution D of the user, it is estimated by sampling whether to churn at each time t, thereby estimating the churn time of the user's main service.

  The second main service contractor number transition prediction unit 32 is a contractor of the main service in the case of having the abolition option service based on each cancellation time of the estimation target user estimated by the second main service cancellation time estimation unit 30. Predict changes in numbers.

  The result generation unit 34 displays the transition of the number of subscribers of the main service predicted by the first main service contractor number transition prediction unit 28 and the abolition option service predicted by the second main service contractor number transition prediction unit 32. A result as shown in FIG. 3 is generated according to the transition of the number of subscribers of the main service in the case of having. In FIG. 3, the horizontal axis represents the elapsed time from the present (at the end of observation), and the vertical axis represents the number of subscribers to the main service.

<Operation of cancellation estimation device>
Next, the operation of the churn estimation device 100 according to the embodiment of the present invention will be described. Before executing the cancellation estimation processing routine shown in FIG. 4, the contract continuation period (dr (i, basic)) of the main service for each of the users, which is information related to each of the composite services, by the input unit 10. , The contract continuation period (dr (i, op1), dr (i, op2), ..., dr (i, opX)) of each option service up to the present time, the elapsed time from the start of the contract of the main service to the start of each option service contract st (st (i, op1), st (i, op2), ..., st (i, opX)), current contract status (contracted / cancelled) (main service contract status c (i, basic), Accepting and storing each option service contract status c (i, op1), c (i, op2), ..., c (i, opX)) and other information related to the user (such as the area where they live) 22 to store. When the input unit 10 receives a combination of a user group to be estimated, a composite service to be estimated, and an optional service for performing abolition simulation, the churn estimation apparatus 100 executes a churn estimation processing routine shown in FIG.

  First, in step S100, the estimation target user group received by the input unit 10 is read.

  Next, in step S102, the estimation target composite service received by the input unit 10 is read.

  Next, in step S104, the combination of optional services to be abolished simulation received in the input unit 10 is read.

  Next, in step S106, the contract continuation period (dr (i, basic)) of the main service for each of the estimation target users acquired in step S100 in the target composite service acquired in step S102, The contract continuation period (dr (i, op1), dr (i, op2), ..., dr (i, opX)) of the optional service up to the present time, the elapsed time st ( st (i, op1), st (i, op2), ..., st (i, opX)), current contract status (contracted / cancelled) (main service contract status c (i, basic), and each The contract status c (i, op1), c (i, op2),..., C (i, opX)) of the option service is read from the storage unit 22.

  Next, in step S108, the contract continuation period (dr (i, basic)) of each main service of the user acquired in step S106 and the current contract status (contracted / cancelled) (main service Based on the contract status c (i, basic), a distribution parameter (λ (basic)) representing the distribution of the churn probability of the main service is estimated.

  Next, in step S109, the contract continuation period (dr (i, basic)) up to the present time of each main service of the user acquired in step S106 and the contract continuation period (dr (i, op1) up to the present time of each option service are obtained. ), dr (i, op2), ..., dr (i, opX)), the elapsed time st (st (i, op1), st (i, op2), ... , St (i, opX)), current contract status (contracted / cancelled) (main service contract status c (i, basic), and optional service contract status c (i, op1), c (i , op2),..., c (i, opX)), distribution parameters (λ (1), λ (2),..., λ (X) representing the distribution of the churn probability of the main service from each optional service churn )) Is estimated.

  Next, in step S110, it is determined whether the main service of the estimation target user is under contract. If the estimation target user's main service is under contract, the process proceeds to step S112. If the estimation target user's main service has been canceled, the process proceeds to step S114.

  Next, in step S111, for each of the distribution parameter (λ (basic)) representing the distribution of the cancellation probability of the main service acquired in step S108 and the optional service that has been canceled for the estimation target user, the optional service cancellation is performed. Therefore, a new distribution of churn probability is created based on the distribution parameters representing the distribution of churn probability of the main service.

  Next, in step S112, based on the new churn probability distribution created in step S111, the churn time of the main service of the estimation target user is estimated.

  Next, in step S114, it is determined whether or not processing has been completed for all estimation target users. If the process has been completed for all estimation target users, the process proceeds to step S116. If the process has not been completed for all estimation target users, the estimation target users are switched, and steps S110 to S110 are performed. The process of S114 is repeated.

  Next, in step S116, the transition of the contractor of the main service is predicted based on the cancellation time of the first main service of each estimation target user estimated in step S112.

  Next, in step S118, it is determined whether the main service of the estimation target user is under contract. When the main service of the estimation target user is under contract, the process proceeds to step S120, and when the main service of the estimation target user has been canceled, the process proceeds to step S124.

  Next, in step S120, it is determined whether or not at least one of the optional services that are in the contract of the estimation target user is included in the combination of optional services that perform the abolition simulation acquired in step S104. When at least one of the optional services under contract is included in the combination of optional services for which the abolition simulation is performed, the process proceeds to step S122, and all the optional services that are under contract of the estimation target user are abolished. If it is not included in the combination of optional services for simulation, the process proceeds to step S124.

  Next, in step S122, the distribution parameter (λ (basic)) representing the distribution of the churn probability of the main service of the estimation target user acquired in step S108, and each of the churn of the estimation target user acquired in step S109. A new churn probability distribution D_con is created based on the distribution parameter representing the distribution of the churn probability of the main service from the churn of the option service for each option service.

  Next, in step S124, the distribution representing the distribution of the probability of cancellation of the main service from the cancellation of the optional service for each of the optional services to be abolished, for which the estimation target user acquired in step S109 is under contract. Based on the parameters, a new churn probability distribution D_churn is created.

  Next, in step S126, a new churn probability distribution D is created based on the new churn probability distribution D_con acquired in step S122 and the new churn probability distribution D_churn acquired in step S124.

  Next, in step S128, based on the new churn probability distribution D acquired in step S126, the churn time of the main service of the estimation target user is estimated.

  Next, in step S130, it is determined whether or not processing has been completed for all estimation target users. If the process has been completed for all estimation target users, the process proceeds to step S132. If the process has not been completed for all estimation target users, the process proceeds to step S118, and the estimation target user is changed. The process of step S124 is repeated.

  Next, in step S132, based on the cancellation time of the main service of the estimation target user acquired in step S128, the transition of the contractor of the main service when the abolition option service is provided is predicted.

  Next, in step S128, a result is generated based on the transition of the contractor of the main service acquired in step S116 and the transition of the contractor of the main service when having the abolition option service acquired in step S132. The process ends.

  Step S108 is realized by a distribution parameter λ (basic) estimation processing routine shown in FIG.

  First, in step S200, the value of variable k is set to zero.

  Next, in step S201, the value of k_max is set to 15000.

  Next, in step S202, the initial value of each distribution parameter λ (basic) representing the distribution of the churn probability of the main service is set to an appropriate value.

  Next, in step S204, the value of the variable k is set to k + 1.

  Next, in step S206, each distribution parameter λ (basic) representing the distribution of the probability of cancellation of the main service is added with a uniform random number that takes a value from −1 to 1, and a distribution of the cancellation probability of the new main service is obtained. A distribution parameter λnew (basic) representing is generated.

  In step S208, the distribution parameter λnew (basic) representing the distribution of the new main service churn probability acquired in step S206 or the distribution parameter λ (basic) representing the distribution of the churn probability of the main service is more likely. As a criterion for determining whether or not, the acceptance rate α (α (basic)) representing the result of comparing the likelihood of the distribution parameter λnew (basic) and the likelihood of the distribution parameter λnew (basic) is calculated. Specifically, α (basic) is calculated by the following equation (1).

  In step S210, a uniform random number R (basic) of 0 to 1 is generated.

  In step S212, the acceptance rate α (basic) acquired in step S208 is compared with the uniform random number R (basic) acquired in step S210, and the acceptance rate α (basic) is obtained from the uniform random number R (basic). It is determined whether or not is too large. If the acceptance rate α (basic) is larger than the uniform random number R (basic), the process proceeds to step S214. If the acceptance rate α (basic) is equal to or less than the uniform random number R (basic), the process proceeds to step S216. Transition.

  In step S214, the distribution parameter λnew (basic) representing the distribution of the new main service churn probability generated in step S206 is set as the value of the distribution parameter λ representing the churn probability distribution of the main service.

  In step S216, it is determined whether or not the variable k is larger than k_max. If the variable k is greater than k_max, the process ends. If the variable k is equal to or less than k_max, the process proceeds to step S204 and the process is repeated.

  The above step S109 is realized by a distribution parameter λ (x) estimation processing routine representing the distribution of the cancellation probability of the main service from each optional service cancellation shown in FIG.

  First, in step S300, the value of variable k is set to zero.

  Next, in step S301, the value of k_max is set to 15000.

  Next, in step S302, the value of the variable x is set to 1.

  Next, in step S303, the initial value of the distribution parameter λ (x) representing the distribution of the cancellation probability of the main service from the cancellation of the optional service x is set to an appropriate value.

  Next, in step S304, it is determined whether or not the value of the variable x matches X. When the value of the variable x matches X, the process proceeds to step S306, and when the value of the variable x does not match X, the process proceeds to step S305.

  In step S305, the value of variable x is set to x + 1, and the process proceeds to step S303.

  In step S306, the value of variable k is set to k + 1.

  Next, in step S307, the value of the variable x is set to zero.

  Next, in step S308, the value of the variable x is set to x + 1.

  Next, in step S309, a uniform random number is added to the value of the distribution parameter λ (x) of the option service x to generate a new distribution parameter λnew (x) of the option service x. The uniform random number takes a value of −1 to 1, for example.

  Next, in step S310, based on the distribution parameter λnew (x) of the new option service x acquired in step S309 and the distribution parameter λ (x) of the option service x, distribution parameters λnew (x) and λ ( As a criterion for determining which is more likely to be x), an acceptance rate α (α (x)) representing the result of comparing the likelihoods of the two is calculated. Specifically, α (x) is calculated by the following equation (2).

  In step S311, uniform random numbers R (x) from 0 to 1 are generated.

  In step S312, the acceptance rate α (x) acquired in step S310 is compared with the uniform random number R (x) acquired in step S311, and the acceptance rate α (x) is obtained from the uniform random number R (x). It is determined whether or not is too large. If the acceptance rate α (x) is larger than the uniform random number R (x), the process proceeds to step S314. If the acceptance rate α (x) is equal to or less than the uniform random number R (x), the process proceeds to step S315. Transition.

  In step S314, the value of the distribution parameter λnew (s) of the new option service x generated in step S309 is set as the value of the distribution parameter λ (x) of the option service x.

  In step S315, it is determined whether or not the value of the variable x matches X. If the value of the variable x matches X, the process proceeds to step S316. If the value of the variable x does not match X, the process proceeds to step S308.

  In step S316, it is determined whether or not the variable k is larger than k_max. If the variable k is greater than k_max, the process ends. If the variable k is equal to or less than k_max, the process proceeds to step S306 and the process is repeated.

  As described above, according to the churn estimation apparatus according to the embodiment of the present invention, the churn time of the main service can be accurately estimated.

  In addition, it is possible to model the temporal relationship between the time when the optional service is canceled and the time when the main service is canceled.

  In addition, it is possible to deal with problems such as the effect of starting the main service only immediately after canceling a certain optional service, which can be handled by the conventional method, by changing the distribution form.

  In addition, as a result of estimating the transition of the main service user withdrawal, it is possible to estimate the period until the main service is canceled after the optional service is canceled for a certain service, and for users who cancel the optional service, Measures such as making phone calls to prevent the cancellation of the main service within the expected cancellation period through the call center can be taken.

  In addition, it is possible to estimate the period from the cancellation of the optional service to the cancellation of the main service from the multiple optional services, but there are multiple options for determining the estimation result based on the multiple results. The algorithm is applicable.

  In addition, when conducting the abolition simulation of optional services, the understanding of the relationship between the main service and the optional service through questionnaire surveys, etc., which was necessary in the past, is canceled when voluntarily canceled or when the service is canceled forcibly Assuming that the degree of influence is the same, the distribution representing the temporal correlation between the two is replaced by estimating only from the user's contract information, avoiding the cost of the survey and the fluctuation of the analysis results by the researcher be able to.

  Also, assuming that the degree of influence is the same when voluntarily canceling and when the service is forcibly canceled due to the abolition of service provision, the relationship between the main service and option service through questionnaire surveys and market surveys, which has been a conventional issue It solves the problem of additional cost that defines sex and the problem that analysis results vary greatly depending on the investigator.

  In addition, since the parameter estimation for this event is performed on the assumption that there is a temporal correlation according to a certain distribution between the cancellation of the optional service and the cancellation of the main service, the above problem 1 is solved. Can respond.

  Further, assuming that the degree of influence is the same when voluntarily canceling and when the service is forcibly canceled due to the abolition of service provision, it is possible to reduce the input to the device necessary for estimation.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  In the present embodiment, the case has been described where the input unit 10 receives a composite service to be estimated, but the present invention is not limited to this. For example, when there is only one composite service type stored in the storage unit 22, the composite service to be estimated need not be input information.

  In the present embodiment, the case of modeling the distribution of the cancellation probability of the contract service has been described, but the present invention is not limited to this. For example, the distribution of withdrawal probabilities of a plurality of phenomena that can be expressed in two states and have a master-slave relationship may be modeled. For example, the present invention may be applied to a departure estimation apparatus that estimates the distribution of a shopping mall departure probability, with the occurrence of a shopping mall as a main phenomenon and the occurrence of a shopping mall store child as a sub phenomenon.

  In the present embodiment, the case of using information on the past contract period dr, elapsed time st, and current contract status c has been described, but the present invention is not limited to this. For example, when simulating the transition of the number of main service subscribers in a specific area, it is necessary to narrow down the estimation target to users who use the service in the corresponding area, so the area where each user lives You may use the information showing.

  In this embodiment, τ (i, basic), end (i, op1), end (i, op2), ... are exponential distribution parameters λ (basic), λ (1), λ (2), respectively. However, the present invention is not limited to this. For example, τ (i, basic), end (i, op1), end (i, op2),... May follow a distribution represented by a Weibull distribution parameter or a lognormal distribution parameter, respectively.

DESCRIPTION OF SYMBOLS 10 Input part 20 Calculation part 22 Storage part 24 Distribution parameter estimation part 26 1st main service cancellation time estimation part 28 1st main service contractor number transition prediction part 30 2nd main service cancellation time estimation part 32 2nd main service contractor Number transition prediction unit 34 Result generation unit 50 Output unit 100 Departure estimation device

Claims (6)

A departure estimation device that estimates a distribution of probability of departure of a main phenomenon that occurs in an individual and that has a subordinate phenomenon with a master-subordinate relationship,
The observation continuation period dr (i, basic) collected for each of a plurality of individuals i until the end of the observation continues, and the sub-op until the end of the observation for each sub-op. The observation duration period dr (i, op), information c (i, basic) indicating whether or not the main phenomenon is observed at the end of the observation, and the sub phenomenon at the end of the observation for each sub phenomenon op. Information c (i, op) indicating whether or not is observed, and an elapsed time st (i, op) from the start time of the main phenomenon to the start time of the sub phenomenon for each sub phenomenon op Data acquisition means for acquiring
The initial value of the distribution parameter representing the distribution of the probability of departure of the main phenomenon at each time, and the initial value of the distribution parameter representing the distribution of the probability of departure of the main phenomenon at each time from the departure of the sub phenomenon for each sub phenomenon op An initial value setting means for setting a value;
Updating the distribution parameter of the main phenomenon based on the observation duration dr (i, basic) and the information c (i, basic) of each of the plurality of individuals i acquired by the data acquisition means; Based on the observation duration dr (i, op), the information c (i, op), and the elapsed time st (i, op) of each of the plurality of individuals i acquired by the data acquisition unit, Distribution parameter updating means for updating the distribution parameter of each sub phenomenon op;
Iterative determination means that repeats the update by the distribution parameter update means until a predetermined iteration end condition is satisfied,
The distribution parameters of the main phenomenon updated by the distribution parameter updating means, the distribution parameters of each sub-op, and the observation duration dr (i, i, for the estimation target individual i where the main phenomenon is observed. basic), the observation duration dr (i, op) of each sub-op, the information c (i, op) of each sub-op, and the elapsed time st (i, op) of each sub-op. The distribution probability distribution based on the distribution parameter of the main phenomenon and the observation continuation period dr (i, basic), the distribution parameter of each sub-op that has departed in the individual to be estimated, the observation continuation According to the period dr (i, op) and the distribution of the probability of withdrawal based on the elapsed time st (i, op), the withdrawal timing of the main phenomenon in the individual i to be estimated And the withdrawal time estimating means for estimating,
A departure estimation apparatus including: A churn estimation device for estimating a churn probability distribution of a main service having an optional service,
An observation continuation period dr (i, basic) that is collected for each of a plurality of users i until the end of the observation, and the optional service op is continued until the end of the observation for each option service op. The observation continuation period dr (i, op), information c (i, basic) indicating whether or not the main service is contracted at the end of the observation, and the optional service at the end of the observation for each optional service op C (i, op) indicating whether or not the service is contracted, and the elapsed time st (i, op) from the contract start time of the main service to the start time of the contract of the option service for each option service op Data acquisition means for acquiring
An initial value of a distribution parameter representing a distribution of churn probability of the main service at each time, and an initial value of a distribution parameter representing a distribution of churn probability of the main service at each time from the cancellation of the option service for each option service op An initial value setting means for setting a value;
Updating the distribution parameter of the main service based on the observation duration dr (i, basic) and the information c (i, basic) of each of the plurality of users i acquired by the data acquisition means; Based on the observation duration dr (i, op), the information c (i, op), and the elapsed time st (i, op) of each of the plurality of users i acquired by the data acquisition unit, Distribution parameter updating means for updating the distribution parameter of each option service op;
Iterative determination means that repeats the update by the distribution parameter update means until a predetermined iteration end condition is satisfied,
The distribution parameters of the main service updated by the distribution parameter updating means and the distribution parameters of each optional service op, and the observation duration dr (i, basic), the observation duration dr (i, op) of each option service op, the information c (i, op) of each option service op, and the elapsed time st (i, op) of each option service op. The distribution probability distribution based on the distribution parameter of the main service and the observation duration dr (i, basic), the distribution parameter of each optional service op canceled by the user i to be estimated, and the observation Based on the duration dr (i, op) and the elapsed time st (i, op) Accordance with the distribution of the churn probability, and churn timing estimation means for estimating the termination timing of the main service in the user i of the estimation target,
A churn estimation device. The cancellation time estimation means is:
Distribution of churn probability based on the distribution parameter of the main service and the observation duration dr (i, basic), the distribution parameter of each optional service op crushed by the user i to be estimated, the observation duration The distribution of churn probability based on dr (i, op) and the elapsed time st (i, op) and the optional service op to be abolished, and each optional service continuing for the user i to be estimated According to the distribution parameter of op, the observation duration dr (i, op), and the distribution of the churn probability based on the elapsed time st (i, op), the departure time of the main service for the user i to be estimated is The churn estimation apparatus according to claim 2 for estimation. A main phenomenon that includes a data acquisition means, an initial value setting means, a distribution parameter update means, an iterative determination means, and a withdrawal timing estimation means, and is a main phenomenon that occurs in an individual and has a subordinate phenomenon with a master-slave relationship A departure estimation method in a departure estimation device that estimates the distribution of departure probability of
The data acquisition means collects each of a plurality of individuals i, an observation duration dr (i, basic) in which the main phenomenon continues until the observation end time, and until the observation end time for each sub phenomenon op. The observation duration period dr (i, op) in which the sub phenomenon op continues, information c (i, basic) indicating whether or not the main phenomenon is observed at the end of the observation, and the observation about each sub phenomenon op Information c (i, op) indicating whether or not the sub phenomenon is observed at the end time, and the elapsed time st from the start time of the main phenomenon to the start time of the sub phenomenon for each sub phenomenon op (I, op) and
The initial value setting means includes an initial value of a distribution parameter representing a distribution of the probability of departure of the main phenomenon at each time, and a probability of the main phenomenon leaving at each time from the departure of the sub phenomenon for each sub phenomenon op. Set the initial value of the distribution parameter representing the distribution,
The distribution parameter updating unit is configured to determine the main phenomenon based on the observation duration dr (i, basic) and the information c (i, basic) of each of the plurality of individuals i acquired by the data acquisition unit. The distribution parameter is updated, and the observation duration dr (i, op), the information c (i, op), and the elapsed time st (i) of each of the plurality of individuals i acquired by the data acquisition unit. , Op) to update the distribution parameters for each sub phenomenon op,
The iterative determination means repeats the update by the distribution parameter update means until a predetermined iteration end condition is satisfied,
The departure time estimation means includes the distribution parameter of the main phenomenon updated by the distribution parameter update means, the distribution parameter of each sub phenomenon op, and the estimation target individual i where the main phenomenon is observed. The observation duration dr (i, basic), the observation duration dr (i, op) of each sub-op, the information c (i, op) of each sub-op, and the elapsed time st ( i, op) based on the distribution parameter of the main phenomenon and the distribution of the probability of withdrawal based on the observation duration dr (i, basic), and each sub-op of each sub-op According to the distribution parameter, the observation duration dr (i, op), and the distribution of the withdrawal probability based on the elapsed time st (i, op), the individual i to be estimated is Withdrawal estimation method to estimate the withdrawal period of the main phenomenon that. A churn estimation in a churn estimation device that estimates a churn probability distribution of a main service having an optional service, including data acquisition means, initial value setting means, distribution parameter update means, iterative determination means, and churn time estimation means. A method,
The data acquisition means collects each of a plurality of users i, the observation continuation period dr (i, basic) that the main service has continued until the observation end point, and the observation end point for each optional service op. The observation continuation period dr (i, op) in which the optional service op has continued, the information c (i, basic) indicating whether or not the main service is contracted at the end of the observation, and the observation for each optional service op Information c (i, op) indicating whether or not the option service is contracted at the end time, and the elapsed time st from the contract start time of the main service to the start time of the option service contract for each option service op (I, op) and
The initial value setting means includes an initial value of a distribution parameter representing a distribution of the churn probability of the main service at each time, and a churn probability of the main service at each time from the churn of the option service for each option service op. Set the initial value of the distribution parameter representing the distribution,
The distribution parameter updating unit is configured to determine the main service based on the observation duration dr (i, basic) and the information c (i, basic) of each of the plurality of users i acquired by the data acquisition unit. The distribution parameter is updated, and the observation duration dr (i, op), the information c (i, op), and the elapsed time st (i) of each of the plurality of users i acquired by the data acquisition unit. , Op) to update the distribution parameters of each optional service op,
The iterative determination means repeats the update by the distribution parameter update means until a predetermined iteration end condition is satisfied,
The cancellation timing estimation means includes the distribution parameter of the main service updated by the distribution parameter update means and the distribution parameter of each optional service op, and the estimation target user i who contracts the main service. Observation duration dr (i, basic), the observation duration dr (i, op) of each option service op, the information c (i, op) of each option service op, and the elapsed time st ( i, op) based on the distribution parameter of the main service and the distribution of the churn probability based on the observation duration dr (i, basic), and each optional service op churn by the user i to be estimated The distribution parameter, the observation duration dr (i, op), and the elapsed time st (i, op) according the distribution of cancellation probabilities based on surrender estimation method for estimating the termination timing of the main service in the user i of the estimated target. The program for functioning a computer as each means which comprises the withdrawal estimation apparatus of Claim 1, or the cancellation estimation apparatus of Claim 2 or 3.