JP5722276B2 - First purchase estimation apparatus, method, and program - Google Patents

Description

  The present invention relates to an initial purchase estimation apparatus, method, and program, and more particularly to an initial purchase estimation apparatus, method, and program for estimating an initial purchase of an unpurchased user.

  At EC sites, there are more potential customers who have not made purchases than existing customers who have already made purchases. Therefore, there is a need for a way to predict potential customer purchases and discover approaches that lead potential customers to purchase.

  In addition, there is a known method to estimate the effect of various marketing variables on the survival rate by assuming the assumption that purchases and withdrawals occur randomly and calculating the survival rate taking into account the visit cycle of each individual. Patent Document 1).

M. Abe, ““ Counting Your Customers ”One by One: A Hierarchical BayesExtension to the Pareto / NBD Model”, Marketing Science, 28 (3), pp. 541-553, 2009.

  However, the technique described in Non-Patent Document 1 has a problem that it cannot be applied to unpurchased users because it is calculated based on purchase data.

  The present invention has been made to solve the above problem, and an object of the present invention is to provide an initial purchase estimation apparatus, method, and program capable of estimating the initial purchase of an unpurchased user.

In order to achieve the above object, an initial purchase estimation apparatus according to the present invention is an initial purchase estimation apparatus that estimates an initial purchase of an unpurchased user, and the predetermined purchase of the user collected for each of a plurality of stores. History acquisition means for acquiring a history relating to pre-purchase behavior and a history relating to the purchase of the user, and based on the behavior history, whether or not the user has made a purchase and initial purchase for each of the plurality of stores. Based on the action history, pre-processing means for obtaining a first marketing period from the previous action to the first purchase, obtaining a predetermined marketing variable relating to the user, initial setting means for setting an initial value of the user's determination parameter λ, For the store where the user has taken the pre-purchase action and has not yet purchased, the initial value or the determination parameter λ Is determined using the determination power parameter λ determined last time, according to an exponential distribution representing the probability distribution of the initial period, an initial period determination means for determining a value of the initial period, and the predetermined determination parameter λ Candidate determination means for determining a candidate λ _tmp for the value of the determination parameter λ, the probability distribution of the determination parameter λ, the value of the initial period determined by the initial period determination means, and the previous The determination power based on the likelihood of the determination parameter λ calculated based on the initial period determined by the processing means and the prior distribution of the determination parameter λ based on the predetermined marketing variable for the user The initial value of the parameter λ or the determination power parameter λ determined last time and the determination power parameter determined by the candidate determination means. Parameter determination means that employs one of the parameter λ value candidates λ _tmp , and repetition of repeating the determination by the initial period determination means, the determination by the candidate determination means, and the adoption by the parameter determination means until a predetermined number of repetitions A determination means; a parameter calculation means for calculating the value of the determination power parameter λ based on each of the adoption results of the determination power parameter λ by the parameter determination means; and the determination calculated by the parameter calculation means. Based on the value of the force parameter λ, for a store where the user has taken the pre-purchase behavior and has not yet purchased, the probability that the user will make the first purchase within the specified period and the expected date of the first purchase And initial purchase estimation means for estimating a value.

The initial purchase estimation method of the present invention includes a history acquisition unit, a preprocessing unit, an initial setting unit, an initial period determination unit, a candidate determination unit, a parameter determination unit, an iterative determination unit, a parameter calculation unit, and an initial purchase estimation unit. An initial purchase estimation method in an initial purchase estimation apparatus for estimating an initial purchase of an unpurchased user, which is collected for each of a plurality of stores by the history acquisition unit and relates to a predetermined pre-purchase behavior of the user A step of obtaining an action history including a history and a history relating to the purchase of the user, and whether or not the user has made a purchase and initial purchase for each of the plurality of stores based on the action history by the preprocessing means. Obtain the initial period from the previous action to the first purchase, as well as the predetermined marketing variables for the user A step of setting an initial value of the determination power parameter λ of the user by the initial setting means, and the user taking the pre-purchase action based on the action history by the initial period determination means. In addition, for stores that have not yet been purchased, the initial period value is determined according to the initial value of the determination parameter λ or the exponential distribution representing the probability distribution of the initial period determined using the previously determined determination parameter λ. Determining a candidate value λ _tmp of the determination parameter λ according to a predetermined probability distribution of the determination parameter λ by the candidate determination unit, and the parameter determination unit by the parameter determination unit Probability distribution of the determination parameter λ, the value of the initial period determined by the initial period determination means, and the preprocessing The determination parameter based on the likelihood of the determination parameter λ calculated based on the initial period determined by the stage and the prior distribution of the determination parameter λ based on the predetermined marketing variable for the user adopting either one of the initial value of λ or the previously determined determination parameter λ and the candidate λ _tmp of the determination parameter λ determined by the candidate determination unit; and the iterative determination unit, Repeating the determination by the initial period determination means, the determination by the candidate determination means, and the adoption by the parameter determination means until the predetermined number of repetitions, and the value of the determination parameter λ by the parameter determination means by the parameter calculation means Based on each of the results of adoption, the value of the determination parameter λ is calculated. A store where the user has taken the pre-purchase behavior and has not yet been purchased based on the value of the determination parameter λ calculated by the parameter calculation means by the initial purchase estimation means Estimating the probability that the user will make an initial purchase within a specified period and the expected value of the date of the initial purchase.

Further, the program of the present invention is a program for estimating the initial purchase of an unpurchased user, the computer collects the history of the user's predetermined pre-purchase behavior collected for each of a plurality of stores, and History acquisition means for acquiring an action history including a history related to the purchase of the user, based on the action history, for each of the plurality of stores, whether or not the user purchases, and the first time from the first pre-purchase action to the first purchase A pre-processing means for obtaining a predetermined marketing variable related to the user, an initial setting means for setting an initial value of the determination power parameter λ of the user, and the user performs the pre-purchase action based on the action history. For stores that have been taken and have not yet been purchased, the initial value of the determination parameter λ or previously determined An initial period determining means for determining a value of the initial period according to an exponential distribution representing a probability distribution of the initial period, which is determined using a determination parameter λ, according to a predetermined probability distribution of the determination parameter λ, Candidate determination means for determining a candidate λ _tmp for the value of the determination parameter λ, the probability distribution of the determination parameter λ, the value of the initial period determined by the initial period determination means, and the value obtained by the preprocessing means Based on the likelihood of the determination parameter λ calculated based on the initial period and the prior distribution of the determination parameter λ based on the predetermined marketing variable for the user, the initial value or the previous time of the determination parameter λ The determined decision parameter λ and the candidate value of the decision parameter λ determined by the candidate determination unit Any employed to parameter determining means one of _tmp, before repeating a predetermined number of times, determined by the first period determining means, determination by the candidate determination unit, and repetition determination means for repeating the adoption by the parameter determining means, by the parameter determining means Based on each of the adoption results of the value of the determination power parameter λ, parameter calculation means for calculating the value of the determination power parameter λ, and based on the value of the determination power parameter λ calculated by the parameter calculation means, As the initial purchase estimation means for estimating the probability that the user will make an initial purchase within a specified period and the expected value of the date of the initial purchase for a store where the user has taken the pre-purchase behavior and has not yet purchased. It is a program to make it function.

  As described above, according to the initial purchase estimation apparatus, method, and program of the present invention, an exponential distribution that represents a probability distribution of the initial period from the initial pre-purchase behavior to the initial purchase, which is determined using the determination parameter. To determine the value of the initial period, and using the determined initial period value and the likelihood of the determination parameter λ calculated based on the initial period obtained for the purchased store, A store where the user has taken a pre-purchase action and has not yet purchased, based on the calculated value of the determination parameter λ, by repeatedly determining the value and calculating the value of the determination parameter λ By estimating the probability of making an initial purchase and the expected value of the day of the initial purchase, it is possible to estimate the initial purchase of an unpurchased user

It is a block diagram which shows the structure of the initial purchase estimation system of embodiment of this invention. It is a block diagram which shows the functional structure of the initial purchase estimation apparatus of embodiment of this invention. It is a figure which shows the content of input data. It is a graph which shows the exponential distribution of a first period. (A) It is a figure which shows the prediction result of initial purchase, and (B) is a figure which shows the coefficient matrix of a marting variable. It is a flowchart which shows the content of the initial purchase estimation processing routine in the initial purchase estimation apparatus of embodiment of this invention. It is a flowchart which shows the flow of the process which estimates the parameter in the initial purchase estimation apparatus of embodiment of this invention.

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

<System configuration>
As shown in FIG. 1, an initial purchase estimation system according to an embodiment of the present invention includes a user terminal 1, a web server group 3, a browsing history storage device 4, and an initial purchase estimation device 10. 1, the web server group 3, the browsing history storage device 4, and the initial purchase estimation device 10 are connected to each other via a communication network 2 such as the Internet. In the initial purchase estimation system shown in FIG. 1, only one user terminal 1 is shown, but actually, a plurality of user terminals are connected via the communication network 2 to the web server group 3 and the browsing history storage device. 4 and the initial purchase estimation apparatus 10 are connected.

  The user terminal 1 is a terminal device used by a specific user (for example, a user # 1 of “user ID: # 1”). For example, a personal computer, a workstation, a home game machine, an Internet TV, a PDA (PDA ( Personal digital assistants), mobile phones, PHS (Personal Handyphone System) and other portable terminals. In addition, the user terminal 1 transmits a web page browsing request to the web server group 3 via the communication network 2, or a web page based on data transmitted by the web server group 3 in response to the browsing request is transmitted to the user PC 1. Software (browser) for displaying on other monitors is incorporated.

  As shown in FIG. 1, when the user # 1 starts a browser and inputs a URL (Uniform Resource Locator) via an input unit (not shown), the user terminal 1 browses the web page of the URL. The request is transmitted to the web server group 3. Then, the web server that can provide the information resource corresponding to the URL input by the user # 1 in the web server group 3 transmits the web page data of the URL as shown in FIG. Thereby, the user terminal 1 displays the received data on a monitor (not shown).

  Here, as shown in FIG. 1, the user terminal 1 has a browsing history acquisition unit 1 a, and the browsing history acquisition unit 1 a is a result of the user starting a browser and making a browsing request. The browsing history data of the web page is acquired along the time series. The browsing history acquisition unit 1a is realized by client software, a browser plug-in, or the like.

  And the browsing history acquisition part 1a transmits the acquired browsing history to the browsing history preservation | save apparatus 4 via the communication network 2, as shown in FIG. Note that the timing at which the browsing history acquisition unit 1a transmits the browsing history to the browsing history storage device 4 can be arbitrarily set by the administrator of the initial purchase estimation system. For example, the browsing history acquisition unit 1a transmits a plurality of browsing histories of the user # 1 to the browsing history storage device 4 every predetermined period (for example, every other day).

  The browsing history storage device 4 shown in FIG. 1 collects the browsing history of each user from the user terminal 1 and other user terminals connected via the communication network 2, and browses the collected browsing history. Store in a history storage unit (not shown). For example, the collected browsing history is classified for each user and stored in the browsing history storage unit.

  Here, the browsing history includes, for example, browsing site information, initial browsing date and time of the browsing site, presence / absence of purchasing experience at the browsing site, and user ID. The browsing history includes the URL of the web page being browsed by the web browser and the current date and time (browsing date and time), and from the URL and browsing date and time of the web page included in the browsing history collected for a certain user, The first browsing date and time of the user's browsing site may be obtained, or the presence / absence of purchasing experience at the browsing site of the user may be obtained from the URL of the Web page included in the browsing history collected for a certain user.

  The initial purchase estimation apparatus 10 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory) that stores a program for executing an initial purchase estimation processing routine described later. It consists of As shown in FIG. 2, this computer functionally includes a condition input unit 12, a browsing history acquisition unit 14, a preprocessing unit 16, a parameter estimation unit 18, a statistical value calculation unit 20, and a display unit 22. It can be expressed in configuration. The parameter estimation unit 18 is an example of an initial setting unit, an initial period determination unit, a candidate determination unit, a parameter determination unit, an iterative determination unit, and a parameter calculation unit, and the statistical value calculation unit 20 is an initial purchase estimation unit. It is an example.

  The condition input unit 12 accepts input of condition information related to the initial purchase to be estimated from the user terminal 1. In the condition information, for example, at least one estimation target user, at least one estimation target site, and a prediction period t are specified. In this embodiment, a case where a plurality of estimation target users and a plurality of estimation target sites are designated will be described as an example.

  The browsing history acquisition unit 14 acquires the browsing history of each of the designated estimation target users and the browsing history related to each of the estimation target sites from the browsing history storage device 4.

  The preprocessing unit 16 generates data serving as input data for the initial purchase estimation process based on the acquired browsing history. As shown in FIG. 3, the generated data includes data obtained for the period from the first browsing to the first purchase for every combination of the estimation target user and the estimation target site. For example, the preprocessing unit 16 calculates a period from when user2 first browses at site A to when purchasing for the first time. For those that have not been viewed, the data value is null. For those that are viewed but not purchased, the data value is Y. The site is an example of a store, browsing the site is an example of pre-purchase behavior, and the browsing history is an example of an action history.

  The generated data includes various marketing variables obtained for each estimation target user. Examples of marketing variables include PV, average purchase price, number of browsing of designated sites, demographic information, etc., which can be arbitrarily set by the service provider. Note that dummy variables that are all 1 may be set.

  In addition, the data that is generated includes data obtained by determining the presence or absence of purchase experience for every combination of the estimation target user and the estimation target site. For example, the purchase experience of user2 at site B is calculated by the preprocessing unit 16.

  The parameter estimation unit 18 estimates a parameter λ indicating the determination power for each estimation target user according to a Markov chain Monte Carlo method (MCMC).

  In the parameter estimation process, as shown in FIG. 4, the index distribution of the period from the first browsing to the first purchase is used, which indicates that the user having a larger parameter λ (decision power) tends to have a shorter period from the first purchase. Thus, for each estimation target user, the parameter λ of the estimation target user is estimated by the MCMC method based on the history of the estimation target user regarding the site with purchase experience. In the MCMC algorithm, the parameter λ of each user is sampled by the independent MH algorithm every MCMC cycle. In the parameter estimation process, a coefficient matrix β representing the degree of association (coefficient) with each marketing variable λ shown in FIG. 5B is simultaneously estimated. According to the coefficient matrix β of the marketing variable, for example, based on the coefficient of m1 (for example, viewing price), it can be determined that the purchase decision is faster as the m1 is higher, and new customers are expanded by the approach of lowering the viewing price. Can do.

  The statistical value calculation unit 20 uses the parameter λ estimated for each estimation target user, as shown in FIG. 5A, as a useful statistical value for each estimation target user, in the unpurchased estimation target site. Calculate initial purchase probability and expected purchase date. For example, the probability that user2 makes an initial purchase at site A during a specified period t and the expected value of the point (day) at which user2 makes the initial purchase at site A are calculated.

  The display unit 22 transmits the calculation result by the statistical value calculation unit 20 and the coefficient matrix β of the marching variable to the user terminal 1 that has performed the condition input. The user terminal 1 displays the received data on a monitor (not shown).

<Operation of the initial purchase estimation system>
When condition information including the estimation target user, the estimation target site, and the prediction period t is input to the initial purchase estimation apparatus 10 of the present embodiment by the user terminal 1, the initial purchase estimation apparatus 10 illustrated in FIG. An initial purchase estimation processing routine is executed.

  First, in step S101, the condition information input by the user terminal 1 is acquired. In step S <b> 102, the browsing history related to the estimation target user and the estimation target site included in the condition information acquired in step S <b> 101 is acquired from the browsing history storage device 4.

  In step S103, on the basis of the browsing history acquired in step S102, the period from the initial browsing to the initial purchase regarding the estimation target user and the estimation target site, the marketing variable, and the purchase presence / absence are calculated as input data for the estimation process. To do.

  In the next step S104, the parameter λ for each estimation target user is estimated using the calculation result in step S103, and at the same time, the coefficient matrix β of the marketing variable is obtained.

  Here, the processing in step S104 is realized by the processing routine shown in FIG.

In step S111, the initial value of parameter λ _i is determined for all estimation target users i. The initial value may be arbitrarily set by the service provider. In this embodiment, the initial value of λ _i for all users i is set to 0.5. A variable k representing the number of repetitions is set to 0. Hereinafter, every cycle λ is generated, and λ in the kth cycle is represented as λ _k . That is, λ ₀ = 0.5.

  In the next step S112, k is incremented by one. In step S113, one of the estimation target users is set as the calculation target user i.

In step S114, based on the data calculated in step S103, the user i browses and sets one of the unpurchased sites as the site to be calculated. In the next step S115, for the site to be calculated, λ = λ _{i, k-1} is substituted into the following equation (1), and random x is calculated by an inverse function method or the like based on the formed exponential distribution equation. Let y _k be generated. However, y _k generates random so that y _k > T holds. T is the period from the first viewing to the observation end point.

  However, F (x) is the probability density of the period x from the first viewing to the first purchase.

  In step S116, it is determined whether or not the processes of steps S114 and S115 have been executed for all the sites that user i has browsed and have not purchased. If there is a site that the user i has browsed and has not yet purchased and the processing of steps S114 and S115 is not performed, the process returns to step S114, and the site is calculated. Set as. On the other hand, if the processes of steps S114 and S115 have been executed for all the sites that user i has browsed and have not purchased, the process proceeds to step S117.

  By the processing in steps S113 to S116, the value in the “Y” portion of the “first viewing to first purchase” row for user i, which is the input data calculated in step S103, is obtained.

In step S117, random numbers are generated based on a proposed distribution (log normal distribution) assumed for λ as shown in the following equation (2), and set to λ _tmp .

  However, the parameters μ and σ may be arbitrarily set by the service provider. Although the case where the proposed distribution is a lognormal distribution has been described as an example, other distributions such as a uniform distribution and a normal distribution may be applied.

  In step S118, α is obtained according to the following equation (3). However, when k = 1, the following equation (5) is omitted.

(If (k = 1)
→ α = (Q (λ _{i, k-1} ) / Q (λ _tmp )) * (L (λ _tmp ) / L (λ _{i, k-1} ))
otherwise
→ α = (Q (λ _{i, k-1} ) / Q (λ _tmp )) * (L (λ _tmp ) / L (λ _{i, k-1} )) * (Prior (λ _tmp ) / Prior (λ _{i , k-1} )))
... (3)

  However, L (λ) is a likelihood formula and is represented by the following formula (4).

However, ρ represents all the sites browsed by the user i, z _site represents the period from the first viewing to the first purchase when there is purchase, and y _k when purchase is not performed.

  Prior (λ) is a prior distribution and is expressed by the following equation (5).

Where d _i is user i's marketing variable matrix n rows and 1 column (m1, m2... Mn) ^t , β is a 1 row and n column coefficient matrix, and S is a variance value.

Then, in step S119, generates uniform random numbers R 0-1, if R <alpha is satisfied, lambda _i, the lambda _tmp employed to _k, lambda _{i, k} otherwise the lambda _{i , k−1} and the value of λ _{i, k} is stored in the memory.

In the next step S120, it is determined whether or not the processing in steps S113 to S119 has been executed for all users to be estimated. If there is a user who has not executed the processes in steps S113 to S119, the process returns to step S113, and the user is set as the calculation target user i. On the other hand, when the processes of steps S113 to S119 are executed for all the estimation target users, it is determined that the values of λ _k for all the estimation target users are estimated, and the process proceeds to step S121.

In step S121, λ _{i, k} of all users is used, and β (n rows and 1 column) and S (1 row and 1 column) are updated by Bayesian regression.

  In the next step S122, it is determined whether or not the number of repetitions k is larger than a predetermined upper limit number (for example, 14000). If k> 14000, the process proceeds to step S123. The process returns to step S112.

In step S123, the average value of the employed λ _{i, k} (4000 <k ≦ 14000) is calculated for all users i, set as the parameter λ _i , and the processing routine is terminated. However, the numbers 4000 and 14000 may be set by the service provider.

Then, in step S105 of the initial purchase estimation processing routine, for each estimation target user, for each of the unpurchased sites A, using the parameter λ _i of the estimation target user estimated in step S104, the following (6 ) Predicts the probability that an unpurchased user will make an initial purchase at site A in period t.

However, the period t is a value included in the condition information acquired in the above step S101, and v is the period from the first viewing on the site A to the observation point.

Further, for each estimation target user, for each unpurchased site A, using the parameter λ _i of the estimation target user estimated in step S104, an unpurchased user can The expected value of the day within the period t in which the initial purchase is made at A is predicted.

However, E (λ _i ) is an expected value (date based on the observation point) on the day when the unpurchased user makes the first purchase.

  In step S106, the prediction result predicted in step S105 and the marketing variable count matrix β obtained in step S104 are transmitted to the user terminal 1 that has input the condition information. Is output, and the initial purchase estimation processing routine is terminated.

  Based on the data displayed on the user terminal 1, for example, the user determines which user is likely to make the first purchase in the future and what is the cause of hesitation for the first purchase. As a result, it is determined that Mr. A has not yet purchased, but is likely to purchase in the near future, or who generally browses the B brand site often leads to purchase.

  As described above, according to the initial purchase estimation apparatus according to the embodiment of the present invention, from the initial pre-purchase behavior to the initial purchase, which is determined using the determination parameter λ for the non-purchased user to be estimated. The value of the initial period x for the unpurchased site is determined according to the exponential distribution representing the probability distribution of the initial period, and the initial period x value for the determined unpurchased site and the site with purchasing experience are determined. The value of the decision parameter λ is determined using the likelihood of the decision parameter λ, the proposed distribution assumed for the decision parameter λ, and the prior distribution based on the marting variable. Repeatedly, the value of the determination parameter λ is calculated. Based on the calculated value of the determination parameter λ, the user is taking a browsing action and has not yet purchased. For that site, by estimating the expected value of the day of the probability and the initial purchase, which for the first time purchase, can be estimated for the initial purchase of the non-purchase of the user.

  In addition, by adopting the hypothesis that “the period from the initial browsing to the initial purchase follows the exponential distribution”, it is possible to predict the initial purchase of potential customers, and users who have not yet made purchases on the EC site will make future purchases. In addition to estimating the probability of performing, it is possible to suggest a factor (marketing variable) that the potential customer cannot make a purchase.

  In addition, various marketing variables are incorporated into the MCMC algorithm as prior distributions, and the factors that prevent potential customers from purchasing based on the behavior of existing customers are estimated. Therefore, it is possible to quantitatively measure how much each marketing variable affects the initial purchase.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made without departing from the gist of the present invention.

  For example, the initial purchase for an unpurchased user in an actual store may be estimated. In this case, the pre-purchase behavior may be a visit, not a browsing, and the first period may be defined as a period from the first visit to the first purchase.

  In addition, browsing may be replaced with other behavior as long as the behavior always occurs before purchase. For example, cart input, top page browsing, etc. may be pre-purchase actions.

  Moreover, you may make it estimate the attendance regarding the non-learning user in E-learning using the estimation process demonstrated by said embodiment. In this case, pre-purchase behavior (browsing) may be replaced with attendance at the nth step and purchase may be replaced with attendance at the mth step (where n <m) in stepwise e-learning.

  Moreover, although the case where the browsing history is acquired in the user terminal has been described as an example, the browsing history is not limited to this, and the browsing history is not limited to the web server, the proxy server, or the server equipped with the DPI (deep packet inspection) function. May be obtained.

  Although the above-mentioned initial purchase estimation apparatus has a computer system inside, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

1 user terminal 1a browsing history acquisition unit 2 communication network 3 web server group 4 browsing history storage device 10 initial purchase estimation device 12 condition input unit 14 browsing history acquisition unit 16 preprocessing unit 18 parameter estimation unit 20 statistical value calculation unit 22 display unit

Claims (3)

An initial purchase estimation device for estimating an initial purchase of an unpurchased user,
History acquisition means for acquiring a history of the user's predetermined pre-purchase behavior collected for each of a plurality of stores, and a behavior history including a history of the user's purchase,
Based on the behavior history, for each of the plurality of stores, pre-processing means for obtaining presence / absence of purchase by the user and a first period from the first pre-purchase behavior to the first purchase and obtaining a predetermined marketing variable for the user When,
An initial setting means for setting an initial value of the determination parameter λ of the user;
Based on the behavior history, the store where the user has taken the pre-purchase behavior and has not yet purchased is determined using the initial value of the determination parameter λ or the previously determined determination parameter λ. The initial period determination means for determining the value of the initial period according to an exponential distribution representing the probability distribution of the initial period;
Candidate determination means for determining a candidate λ _tmp of the value of the determination parameter λ according to a predetermined probability distribution of the determination parameter λ;
The probability distribution of the determination parameter λ, the likelihood of the determination parameter λ calculated based on the initial period value determined by the initial period determination unit and the initial period determined by the preprocessing unit, And an initial value of the determination parameter λ or a previously determined determination parameter λ, and the candidate determination unit, based on a prior distribution of the determination parameter λ based on the predetermined marketing variable regarding the user Parameter determination means that employs one of the candidate values λ _tmp of the determination parameter λ;
Iterative determination means that repeats the determination by the initial period determination means, the determination by the candidate determination means, and the adoption by the parameter determination means until a predetermined number of repetitions,
Parameter calculation means for calculating the value of the determination power parameter λ based on each of the adoption results of the value of the determination power parameter λ by the parameter determination means;
Based on the value of the determination parameter λ calculated by the parameter calculation means, the user has made the initial purchase within a specified period for a store where the user has taken the pre-purchase action and has not yet purchased. Initial purchase estimation means for estimating the probability of performing and the expected value of the day of initial purchase;
Initial purchase estimation device. Including history acquisition means, preprocessing means, initial setting means, initial period determination means, candidate determination means, parameter determination means, iterative determination means, parameter calculation means, and initial purchase estimation means, and estimates the initial purchase of unpurchased users An initial purchase estimation method for an initial purchase estimation device,
Acquiring a history of the user's predetermined pre-purchase behavior collected for each of a plurality of stores by the history acquisition means, and a behavior history including a history of the user's purchase;
Based on the behavior history, the pre-processing means obtains the presence / absence of purchase by the user and the initial period from the first pre-purchase behavior to the first purchase for each of the plurality of stores, and predetermined marketing related to the user Obtaining a variable;
Setting an initial value of the determination power parameter λ of the user by the initial setting means;
The initial value of the determination parameter λ or the previously determined determination power for the store where the user has taken the pre-purchase behavior and has not yet been purchased based on the behavior history by the initial period determination means. Determining a value for the initial period according to an exponential distribution representing a probability distribution for the initial period, defined using a parameter λ;
Determining a candidate value λ _tmp of the determination parameter λ according to a predetermined probability distribution of the determination parameter λ by the candidate determination unit;
The determination power calculated by the parameter determination means based on the probability distribution of the determination parameter λ, the value of the initial period determined by the initial period determination means, and the initial period determined by the preprocessing means. Based on the likelihood of the parameter λ and the prior distribution of the determination parameter λ based on the predetermined marketing variable for the user, the initial value of the determination parameter λ or the previously determined determination parameter λ, and the candidate Adopting any one of the candidate values λ _tmp of the determination parameter λ determined by the determining means;
Repeating the determination by the initial period determination means, the determination by the candidate determination means, and the adoption by the parameter determination means until the repetition determination means repeats a predetermined number of times;
Calculating the value of the determination parameter λ by the parameter calculation unit based on each result of adoption of the value of the determination parameter λ by the parameter determination unit;
Based on the value of the determination parameter λ calculated by the parameter calculation means by the initial purchase estimation means, the store is in a specified period for a store that has taken the pre-purchase action and has not yet been purchased. Estimating the probability that the user will make the first purchase and the expected value of the day of the first purchase;
The initial purchase estimation method. A program for estimating initial purchases of unpurchased users,
Computer
History acquisition means for acquiring a history of the user's predetermined pre-purchase behavior collected for each of a plurality of stores, and a behavior history including a history of the user's purchase,
Based on the behavior history, for each of the plurality of stores, pre-processing means for obtaining presence / absence of purchase by the user and a first period from the first pre-purchase behavior to the first purchase and obtaining a predetermined marketing variable for the user ,
An initial setting means for setting an initial value of the determination parameter λ of the user;
Based on the behavior history, the store where the user has taken the pre-purchase behavior and has not yet purchased is determined using the initial value of the determination parameter λ or the previously determined determination parameter λ. , An initial period determination means for determining a value of the initial period according to an exponential distribution representing a probability distribution of the initial period,
Candidate determining means for determining a candidate λ _tmp of the value of the determination parameter λ according to a predetermined probability distribution of the determination parameter λ,
The probability distribution of the determination parameter λ, the likelihood of the determination parameter λ calculated based on the initial period value determined by the initial period determination unit and the initial period determined by the preprocessing unit, And an initial value of the determination parameter λ or a previously determined determination parameter λ, and the candidate determination unit, based on a prior distribution of the determination parameter λ based on the predetermined marketing variable regarding the user Parameter determination means that employs one of the candidate values λ _tmp of the determination parameter λ,
Iterative determination means that repeats the determination by the initial period determination means, the determination by the candidate determination means, and the adoption by the parameter determination means until a predetermined number of repetitions,
Based on each result of adoption of the value of the determination parameter λ by the parameter determination unit, parameter calculation means for calculating the value of the determination parameter λ, and the determination parameter λ calculated by the parameter calculation unit Based on the value, for the store where the user has taken the pre-purchase behavior and has not yet purchased, the probability that the user will make the first purchase within the specified period and the expected value on the day of the first purchase will be estimated A program for functioning as an initial purchase estimation tool.