JP6730225B2 - Program, analysis apparatus and method for extracting factors affecting results - Google Patents

Description

The present invention relates to a factor analysis technique for extracting a factor that influences a result from big data.

Conventionally, there is a technique of extracting an improvement factor in a service based on user evaluation such as satisfaction (see, for example, Patent Document 1). According to this technique, among a plurality of factors, it is possible to calculate the influence on the satisfaction level that differs for each user, and extract the improvement factors in order of priority. Specifically, the main factors for obtaining the purchasing behavior (result) of the user desired by the service provider can be estimated.

FIG. 1 is an explanatory diagram of factor analysis in the conventional technique.

According to FIG. 1, as big data, a questionnaire describing satisfaction with respect to each factor element of the service for each user, and the purchase behavior of the user as a result element are collected in advance. Then, by analyzing the large number of factor elements and result elements, the factor elements that influence the result elements can be analyzed.
In the case where the texts described by the user are collected as the questionnaire, the texts may be clustered into a plurality of categories (see Patent Document 2, for example) and each cluster may be used as a factor element.

Japanese Patent No. 5948292 JP, 2017-027495, A

However, unless the estimated factor element can be changed on the service provider side, the result element cannot be influenced. That is, the result of factor analysis cannot be utilized at all.
For example, suppose that it is possible to reduce the result element "contract cancellation" by increasing the factor element "terminal quality" for communication services. However, the service provider cannot easily improve the quality of the terminal possessed by the user. With that in mind, simply extracting the factor elements that influence the result elements does not suggest any solution.

Here, the inventors of the present application have thought that it may be necessary for the service provider to be able to change even the factor element that affects the result element.

Therefore, an object of the present invention is to provide a program, an analysis device and a method capable of extracting a changeable factor that influences a result from big data.

According to the present invention, a computer mounted on an apparatus that extracts a factor element that influences a result element by using a “factor element” and a “result element” that include a value and are associated with each belonging identifier Is a program that
A factor element group classification means for classifying a plurality of factor elements into a group of "invariant factor elements" whose values are less likely to change in time series and a group of "transformation factor elements" whose values are more likely to change in time series,
Clustering means for classifying the set of belonging identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each belonging identifier,
For each cluster, an evaluation value calculation means for calculating an evaluation value that affects the result element for each transformation factor element,
It is characterized in that the computer is made to function as a factor element extracting means for extracting the change factor element according to the evaluation value of the result element.

According to another embodiment of the program of the present invention,
Factor element group classification means
For each factor element, for each attribution identifier, it is detected whether the value has changed within a predetermined time range,
For each factor element, when the ratio of the number of belonging identifiers whose value has changed with respect to the predetermined number of belonging identifiers is equal to or more than the first threshold value, the factor element is defined as “transformation factor element”, and the others are “ It is also preferable to have the computer function as an "invariant factor".

According to another embodiment of the program of the present invention,
Factor element group classification means
For each factor element, for each belonging identifier, count the frequency of change of the value within a predetermined time range,
For each factor element, if the ratio of the number of attribution identifiers whose transformation frequency is equal to or higher than the second threshold value to the predetermined number of attribution identifiers is equal to or higher than the third threshold value, the relevant factor element is designated as a “transformation factor element”. It is also preferable to cause the computer to function so that the rest are “invariant factor elements”.

According to another embodiment of the program of the present invention,
The evaluation value calculation means calculates the evaluation value Vi for each transformation factor element i for each cluster based on the number of belonging identifiers corresponding to each result element as follows.
n11: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the first value
n12: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the second value
n21: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the first value
n22: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the second value
n11+n12+n21+n22=number N of all belonging identifiers of cluster
It is also preferable to make the computer function as described above.

According to another embodiment of the program of the present invention,
The evaluation value calculation means calculates, for each cluster, the evaluation value V1 or V2 for each change factor element for the result element as the evaluation value Vi as follows: The first value of the result element in the first value of the change factor elements Evaluation value of: P11=n11/(n11+n12)
Evaluation value of the second value of the result element in the first value of the transformation factor element: P12=n12/(n11+n12)
Evaluation value of the first value of the result element in the second value of the transformation factor element: P21=n21/(n21+n22)
Evaluation value of the second value of the result element in the second value of the transformation factor element: P22=n22/(n21+n22)
Evaluation value that affects the first value of the result element for the transformation factor element: V1=P11-P21
Evaluation value that affects the second value of the result element for the transformation factor element: V2=P22-P12
It is also preferable to make the computer function as described above.

According to another embodiment of the program of the present invention,
The evaluation value calculating means uses the Akaike's Information Criterion (AIC) for each cluster as follows, and the influence factor V(s,C ) Is calculated as the evaluation value Vi of the result element for the transformation factor element.
MLL_IM(s,C)＝(n11+n12) log(n11+n12)
＋(n11+n21) log(n11+n21)
＋(n21+n22) log(n21+n22)
＋(n12+n22) log(n12+n22)−2N log N
MLL_DM(s,C)=n11 log n11+n12 log n12+n21 log n21+n22 log n22−N log N
However, N=n11+n12+n21+n22
AIC_IM(s,C) = -2 x MLL_IM(s,C) + 2 x 2
AIC_IM(s,C): Subordinate AIC of the set of transformation factor element s and result element C
AIC_DM(s,C) = -2 x MLL_DM(s,C) + 2 x 3
AIC_DM(s,C): Independent AIC of the set of transformation factor element s and result element C
V(s,C) = AIC_IM(s, C)-AIC_DM(s,C)
V(s,C): It is also preferable to make the computer function as the influence ratio of the set of the change factor element s and the result element C.

According to another embodiment of the program of the present invention,
It is also preferable that the factor element extracting means causes the computer to extract the transformation factor element i having the maximum value max(Vi) from the evaluation values Vi for each transformation factor element i for the result element in each cluster.

According to another embodiment of the program of the present invention,
The attribution identifier is a user identifier,
The factor element is based on the information that can be collected from the user,
It is also preferable to have the computer function such that the result element is based on the presence or absence of user activity.

According to another embodiment of the program of the present invention,
The factor element is a combination of usage fee, communication volume, usage plan, recommendation level, communication quality, and terminal quality as the daily activities of the user,
The resulting element also preferably causes the computer to function as a continuation/cancellation of the contract as a purchasing behavior of the user.

According to another embodiment of the program of the present invention,
The attribution identifier is a sensor identifier,
The factor element is based on the information that can be collected from the sensor,
It is also preferable to cause the computer to function such that the result element is based on the presence or absence of the process result.

According to the present invention, an analysis device for extracting a factor element that influences a result element by using a “factor element” and a “result element” including a value, which are associated with each belonging identifier,
A factor element group classification means for classifying a plurality of factor elements into a group of "invariant factor elements" whose values are less likely to change in time series and a group of "transformation factor elements" whose values are more likely to change in time series,
Clustering means for classifying the set of belonging identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each belonging identifier,
For each cluster, an evaluation value calculation means for calculating an evaluation value that affects the result element for each transformation factor element,
And a factor element extracting means for extracting the change factor element according to the evaluation value of the result element.

According to the present invention, a method of analyzing a device that extracts a factor element that influences a result element by using a “factor element” and a “result element” that include a value and are associated with each other,
The device is
A first step of classifying a plurality of factor elements into an “invariant factor element” group whose values are less likely to change in time series and a “transformation factor element” group whose values are more likely to change value in time series;
A second step of classifying the set of attribution identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each attribution identifier;
A third step of calculating an evaluation value that affects a result element for each transformation factor element for each cluster,
The fourth step of extracting the transformation factor element is executed according to the evaluation value of the result element.

According to the program, analysis device, and method of the present invention, it is possible to extract a changeable factor that influences a result from big data.

It is explanatory drawing of the factor analysis in a prior art. It is a functional block diagram of the analyzer in this invention. It is explanatory drawing of the element information storage part in this invention. It is explanatory drawing of the factor element group classification part in this invention. It is explanatory drawing of the clustering part in this invention. It is explanatory drawing of the evaluation value calculation part in this invention. It is explanatory drawing showing the total number in the group of the discrete binary of a change factor element and the discrete binary of a result element. It is explanatory drawing of the factor analysis in this invention.

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

FIG. 2 is a functional configuration diagram of the analyzer according to the present invention.

The analysis apparatus 1 of the present invention extracts a factor element that influences a result element by using a “factor element” and a “result element” that include a value and are associated with each belonging identifier.
According to FIG. 2, the analysis device 1 includes an element information storage unit 10, a factor element group classification unit 11, a clustering unit 12, an evaluation value calculation unit 13, and a factor element extraction unit 14. These functional components are realized by executing a program that causes a computer mounted on the device to function. The processing flow of these functional components can also be understood as a factor analysis method.

[Element information storage unit 10]
The element information storage unit 10 stores a plurality of factor elements and result elements in association with each other for each belonging identifier. When the attribution identifier is a user identifier, for example, it becomes as follows.
Factor element: Element based on the information that can be collected from the user Result element: Element based on the presence or absence of user action

FIG. 3 is an explanatory diagram of the element information storage unit in the present invention.

According to FIG. 3, as a communication service, a “factor element” based on a questionnaire or contract information and a “result element” based on the purchasing behavior of the user are associated in time series for each user identifier (attribution identifier). Has been. The value of the factor element may be binary, or may be a numerical value represented by 0 to 10 points, for example. When binarizing the numerical values of the factor element and the result element, it is preferable to minimize the variance.
In the case of a communication service, factor factors are, for example, usage charges, communication volume, usage plan, recommendation level (NPS (Net Promoter Score)) to a customer, communication quality, and terminal quality as daily activities of a user. .. The result element is, for example, the continuation/cancellation of the contract as the purchase behavior of the user.

The present invention can be used for various factor analyzes such as the production process of a factory, in addition to the factor analysis by the user as shown in FIG.
For example, when the belonging identifier is a sensor identifier in the production process of a factory, the process will be as follows, for example.
Factor element: Element based on information that can be collected from sensors Result element: Element based on presence/absence of process result In this case, for example, collected from various sensors (factor elements) that affect normal/abnormal (result element) of the product The value given can be analyzed.

[Factor element group classification unit 11]
The factor element group classification unit 11 classifies a plurality of factor elements into the following groups.
"Invariant factor element" group in which the value is difficult to change in time series "Change factor element" group in which the value is easy to change in time series The factor element group classification unit 11 specifically classifies a plurality of factor elements as follows. To do.

FIG. 4 is an explanatory diagram of the factor element group classification unit in the present invention.

(Category 1) First, for each factor element, whether or not the value has changed within a predetermined time range is detected for each attribution identifier.
According to FIG. 4A, for example, regarding the factor element “use fee”, since the user D includes different values in the predetermined time range (January to May 2017), the factor element “use fee” is included. The "charge" is "transformation" (indicated by a circle in the figure).
Further, according to FIG. 4A, for example, with respect to the factor element “usage plan”, since the user D does not include different values within the predetermined time range, the factor element “usage plan” is “invariant”. (Indicated by = in the figure).

Next, regarding each factor element, for example, when all attribution identifiers are classified, if the ratio of the number of attribution identifiers whose values are changed is equal to or larger than the first threshold value, the factor element is changed to “change factor element”. And the rest are “invariant factor elements”.
Number of users whose value has changed/total number of users ≧first threshold value According to FIG. 4B, for example, regarding the factor element “use fee”, the number of users whose value has changed is 3 Since the ratio (0.6) is equal to or greater than the first threshold value (0.5), the factor element “use fee” is set as the “transformation factor element”.
Further, according to FIG. 4B, for example, in the factor element “use plan”, the ratio (0.2) of the number of users 1 whose value has changed to the total number of users 5 is the first threshold ( Since it is smaller than 0.5), the factor element “use plan” is defined as “invariant factor element”.

(Category 2) First, with respect to each factor element, the frequency of change of the value within a predetermined time range is counted for each attribution identifier.
Next, regarding each factor element, for example, when all attribution identifiers are classified, if the ratio of the number of attribution identifiers whose transformation frequency is equal to or higher than the second threshold is equal to or higher than the third threshold, the factor element concerned Is the "transformation factor element" and the others are "invariance factor elements".
Number of users whose change frequency is greater than or equal to the second threshold value/total number of users ≧third threshold value

According to FIG. 4C, finally, the plurality of factor elements are classified as follows.
Factors of change: Usage charge, communication volume, recommendation level Factors of change: Usage plan, communication quality, terminal quality

As another first embodiment, there is a classification method when the value of the element x is categorical. For example, assume that any one of N values (A1, A2,..., AN) is taken. Here, when categorical elements xi(t) of a certain user (attribution identifier) i are observed at different times t0, t1,..., Tn, xi(t0), xi(t1), ..., xi( tn).
m/n>predetermined threshold value Am: the value most often taken among xi(t0), xi(t1), ..., xi(tn)
m: Number of appearances of Am In this case, Am is dominant in the value of xi during t0, t1, to, and tn, and it can be said that the value is not changed. Therefore, for the user i, the element x is the “invariant factor element”.

As another second embodiment, there is also a classification method when comparing the range of possible values of the element x. For example, when the element x has the upper limit value xmax and the lower limit value xmin, it is obtained when observing continuous elements xi(t) of a certain user (attribution identifier) i at different times t0, t1, ..., tn. Further, the variance ρi of xi(t0), xi(t1), ˜, xi(tn) is calculated.
[rho]i/(xmax-xmin)<predetermined threshold value In this case, the element x is the "invariant factor element" for the user i.

[Clustering unit 12]
The clustering unit 12 classifies the set of belonging identifiers into a plurality of clusters based on the degree of similarity of the value of each invariant factor element for each belonging identifier. For example, clustering can be performed according to the degree of similarity using the value of each invariant factor element as a vector for each belonging identifier. That is, the degree of similarity is obtained by comparing columns of values of invariant factor elements.

FIG. 5 is an explanatory diagram of the clustering unit in the present invention.

The clustering may be either hard clustering or soft clustering. Further, as the clustering, finer classification may be performed for each minimum unit of the value of the factor element, or k-means (k-means method) may be used. The k-means is a non-hierarchical clustering, which is classified into k clusters using the average of clusters.
Further, as another embodiment, a hierarchical classification such as the Ward method may be used.

Specifically, when the invariant factor element is, for example, “communication quality”, it may be classified into a user identifier group that is simply considered to be “communication quality” and a user identifier group that is considered to be “poor communication quality”. it can. As a result, for each user identifier group that considers the invariant factor elements "communication quality is good" and "communication quality is poor", as will be described later, each change factor element such as usage charge, communication amount, recommendation degree, and result element. It is possible to evaluate the degree of association with each of “continuation of contract” and “cancellation of contract”.

[Evaluation value calculation unit 13]
The evaluation value calculation unit 13 calculates, for each cluster, an evaluation value that affects the result element for each transformation factor element.

FIG. 6 is an explanatory diagram of the evaluation value calculation unit in the present invention.

According to FIG. 6, the evaluation value of each transformation factor element with respect to the result element is calculated for each cluster. The evaluation value calculation unit 13 calculates, for each transformation factor element i, an evaluation value Vi based on the number of belonging identifiers corresponding to the result element for each cluster.

FIG. 7 is an explanatory diagram showing the total number of sets of the discrete binary values of the transformation factor elements and the discrete binary values of the result elements.
n11: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the first value
n12: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the second value
n21: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the first value
n22: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the second value
n11+n12+n21+n22=number N of all belonging identifiers of cluster

Specifically, the evaluation value calculation unit 13 can use the following evaluation method.

(Evaluation Method 1) For each cluster, the evaluation value V1 or V2 for each transformation factor element with respect to the result element is calculated as the evaluation value Vi as follows.
Evaluation value of the first value of the result element in the first value of the transformation factor element: P11=n11/(n11+n12)
Evaluation value of the second value of the result element in the first value of the transformation factor element: P12=n12/(n11+n12)
Evaluation value of the first value of the result element in the second value of the transformation factor element: P21=n21/(n21+n22)
Evaluation value of the second value of the result element in the second value of the transformation factor element: P22=n22/(n21+n22)
Evaluation value that affects the first value of the result element for the transformation factor element: V1=P11-P21
Evaluation value that affects the second value of the result element for the transformation factor element: V2=P22-P12

For example, when the change factor element has the first value=“low” and the second value=“high” and the result element has the first value=“release” and the second value=“continue”, the evaluation value is calculated. The unit 13 uses the total number to calculate the evaluation value Vi as follows.
Ratio of contract cancellations in the change factor element "low": Pl=n11/(n11+n12)
Ratio of contract “cancellation” in change factor element “high”: Ph=n21/(n21+n22)
Evaluation value of the transformation factor element in the cluster: Vi=Pl−Ph

(Evaluation method 2) For each cluster, the influence ratio V(s,C) in which the transformation factor element s appears biased to the result element C using the Akaike's Information Criterion (AIC) as follows: Is calculated as the evaluation value Vi of the result element for the transformation factor element.
MLL_IM(s,C)＝(n11+n12) log(n11+n12)
＋(n11+n21) log(n11+n21)
＋(n21+n22) log(n21+n22)
＋(n12+n22) log(n12+n22)−2N log N
MLL_DM(s,C)=n11 log n11+n12 log n12+n21 log n21+n22 log n22−N log N
However, N=n11+n12+n21+n22
AIC_IM(s,C) = -2 x MLL_IM(s,C) + 2 x 2
AIC_IM(s,C): Subordinate AIC of the set of transformation factor element s and result element C
AIC_DM(s,C) = -2 x MLL_DM(s,C) + 2 x 3
AIC_DM(s,C): Independent AIC of the set of transformation factor element s and result element C
V(s,C) = AIC_IM(s, C)-AIC_DM(s,C)
V(s,C): Influence ratio of the combination of the transformation factor element s and the result element C

As described above, the higher the degree of association between the transformation factor element and the result element for each cluster, the higher the evaluation value.

When the variance of the invariant factor element for each cluster is small, the result element and the invariant factor element are independent and can be evaluated by the above-described evaluation methods 1 and 2.
On the other hand, among the invariant factor elements for each cluster, if the variance of the elements that are difficult to change is large, the result element and the invariant factor element may be strongly related.
Therefore, it is preferable to calculate Vj for the result element and each invariant factor element j as in the evaluation method 1 described above.

In this case, the evaluation value V′ is calculated as follows, for example.
V'=max(Vi)-max(Vj)
Or
V′={max(Vi), max(Vj)<predetermined threshold}
{0, max(Vj)> predetermined threshold}

[Factor element extraction unit 14]
The factor element extraction unit 14 extracts the transformation factor element according to the evaluation value of the result element. For example, the transformation factor element can be extracted as follows.

(Extraction 1) A transformation factor element is extracted for each cluster. For the cluster, the transformation factor element i having the maximum value max(Vi) is extracted from the evaluation values Vi for each transformation factor element i for the result element. As a result, it is possible to recognize the change factor element that affects the result element according to the user identifier included in the cluster.
In extraction 1, which cluster does the operator request have a high policy effect? , What is the effective factor for the cluster? Suitable for

(Extraction 2) One transformation factor element is extracted for all clusters. Among the evaluation values Vi for each transformation factor element i extracted for each cluster, the transformation factor element i having the maximum value max(Vi) is extracted. This makes it possible to extract the change factor element that affects the result element regardless of the cluster.
What is Extract 2 the most effective measure for all users? , Which element should be changed? Suitable for

As another embodiment of the present invention, in the case of the result element “cancel contract”, even if the target cluster is extracted, the user who wants to take measures is “cancel contract”, so the result element (contract action) There is a demand for wanting to calculate an evaluation value for a user who has not obtained.
Specifically, as shown in FIG. 2, it is determined which cluster the user i, who can observe the “customer information” and the “questionnaire” but has no result element, corresponds to which cluster. The cluster to which the user i belongs can be uniquely determined from the definition of the cluster if all the element values can be observed.
With this, for example, among users who have not yet canceled the contract (result elements have not been obtained), customers whose behavior can be changed can be extracted.
On the other hand, among the elements of “customer information” and “questionnaire”, some users cannot observe the values of some elements.
y: an element that can be detected z: an element that cannot be detected Using the values of user elements that can observe both y and z, z is an objective variable and y is an explanatory variable, such as SVM (Support Vector Machine) A simple machine learning estimator can also be used.

FIG. 8 is an explanatory diagram of factor analysis in the present invention.

According to FIG. 8, for example, with respect to the communication service for the user group included in an arbitrary cluster, the evaluation value of the transformation factor element is calculated with respect to “continuation of contract (result element)”.
Recommendation degree> Communication volume> Usage fee In this case, it is recognized that there is a high degree of relevance between “continuation of contract” and “recommendation degree”. In this case, it can be considered that the communication service provider needs to increase the “recommendation level” in order to “continue the contract”.

As described above in detail, according to the program, the analysis device and the method of the present invention, it is possible to extract a changeable factor that influences the result from the big data.
In particular, according to the present invention, the degree of association between the result element and the invariant factor element that is difficult to change in time series is not evaluated, but only the degree of association with the change factor element that is easily changed in time series is evaluated. Therefore, the service provider can improve the result element by taking measures against the change factor element having a high evaluation value.

With respect to the various embodiments of the present invention described above, various changes, modifications and omissions of the technical idea and scope of the present invention can be easily made by those skilled in the art. The above description is merely an example and is not intended to be limiting. The invention is limited only by the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 1 Analysis device 10 Element information storage unit 11 Factor element group classification unit 12 Clustering unit 13 Evaluation value calculation unit 14 Factor element extraction unit

Claims (12)

A program that causes a computer installed in a device that extracts a factor element that influences a result element, using a “factor element” and a “result element” that include a value and is associated with each attribution identifier,
A factor element group classification means for classifying a plurality of factor elements into a group of "invariant factor elements" whose values are less likely to change in time series and a group of "transformation factor elements" whose values are more likely to change in time series,
Clustering means for classifying the set of belonging identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each belonging identifier,
Evaluation value calculation means for calculating an evaluation value that affects the result element for each transformation factor element for each cluster,
A program that causes a computer to function as factor factor extraction means for extracting a change factor factor according to the evaluation value of the result factor. The factor element group classification means,
For each factor element, for each attribution identifier, it is detected whether the value has changed within a predetermined time range,
For each factor element, when the ratio of the number of belonging identifiers whose value has changed with respect to the predetermined number of belonging identifiers is equal to or more than the first threshold value, the factor element is defined as “transformation factor element”, and the others are “ The program according to claim 1, which causes a computer to function as an "invariant factor element". The factor element group classification means,
For each factor element, for each belonging identifier, count the frequency of change of the value within a predetermined time range,
For each factor element, when the ratio of the number of attribution identifiers whose transformation frequency is equal to or greater than the second threshold value is equal to or greater than the third threshold value with respect to the predetermined number of attribution identifiers, the relevant factor element is changed to “transformation factor element The computer program is caused to function such that "," and the rest are "invariant factor elements". The evaluation value calculation means calculates the evaluation value Vi for each of the transformation factor elements i based on the number of belonging identifiers corresponding to each result element for each cluster as follows.
n11: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the first value
n12: Number of attribution identifiers whose transformation factor element has the first value and whose result element has the second value
n21: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the first value
n22: Number of attribution identifiers whose transformation factor element has the second value and whose result element has the second value
n11+n12+n21+n22=number N of all belonging identifiers of cluster
4. The program according to claim 1, which causes a computer to function as described above. The evaluation value calculation means calculates, for each cluster, the evaluation value V1 or V2 for each change factor element with respect to the result element as the evaluation value Vi as follows. Evaluation value of 1 value: P11=n11/(n11+n12)
Evaluation value of the second value of the result element in the first value of the transformation factor element: P12=n12/(n11+n12)
Evaluation value of the first value of the result element in the second value of the transformation factor element: P21=n21/(n21+n22)
Evaluation value of the second value of the result element in the second value of the transformation factor element: P22=n22/(n21+n22)
Evaluation value that affects the first value of the result element for the transformation factor element: V1=P11-P21
Evaluation value that affects the second value of the result element for the transformation factor element: V2=P22-P12
The program according to claim 4, which causes a computer to function as described above. The evaluation value calculation means uses, for each cluster, the influence ratio V(s, V(s, C) is calculated as the evaluation value Vi of the result element for the transformation factor element
MLL_IM(s,C)＝(n11+n12) log(n11+n12)
＋(n11+n21) log(n11+n21)
＋(n21+n22) log(n21+n22)
＋(n12+n22) log(n12+n22)−2N log N
MLL_DM(s,C)=n11 log n11+n12 log n12+n21 log n21+n22 log n22−N log N
However, N=n11+n12+n21+n22
AIC_IM(s,C) = -2 x MLL_IM(s,C) + 2 x 2
AIC_IM(s,C): Subordinate AIC of the set of transformation factor element s and result element C
AIC_DM(s,C) = -2 x MLL_DM(s,C) + 2 x 3
AIC_DM(s,C): Independent AIC of the set of transformation factor element s and result element C
V(s,C) = AIC_IM(s, C)-AIC_DM(s,C)
V(s,C): The program according to claim 4, which causes a computer to function as an influence ratio of a set of a transformation factor element s and a result element C. The factor factor extraction means causes a computer to function so as to extract the transformation factor element i having the maximum value max(Vi) among the evaluation values Vi for each transformation factor element i for the result element in each cluster. The program according to any one of claims 1 to 6. The attribution identifier is a user identifier,
The factor element is based on information that can be collected from the user,
The program according to any one of claims 1 to 7, wherein the result element causes the computer to function so as to be based on the presence or absence of a user action. The factor element is a combination of a usage charge, communication volume, usage plan, recommendation level, communication quality, and terminal quality as a daily activity of a user,
The program according to claim 8, wherein the result element causes a computer to function as continuation/cancellation of a contract as a purchase behavior of a user. The attribution identifier is a sensor identifier,
The factor is based on information that can be collected from the sensor,
The program according to any one of claims 1 to 7, wherein the result element causes a computer to function so as to be based on the presence or absence of a process result. An analysis device for extracting a factor element that influences a result element by using a “factor element” and a “result element” including a value, which are associated with each attribution identifier,
A factor element group classification means for classifying a plurality of factor elements into a group of "invariant factor elements" whose values are less likely to change in time series and a group of "transformation factor elements" whose values are more likely to change in time series,
Clustering means for classifying the set of belonging identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each belonging identifier,
Evaluation value calculation means for calculating an evaluation value that affects the result element for each transformation factor element for each cluster,
And a factor element extracting means for extracting a change factor element according to the evaluation value of the result element. A method of analyzing a device, which is associated with each attribution identifier, and uses a "factor element" and a "result element" that include a value to extract a factor element that affects the result element,
The device is
A first step of classifying a plurality of factor elements into an “invariant factor element” group whose values are less likely to change in time series and a “transformation factor element” group whose values are more likely to change value in time series;
A second step of classifying the set of attribution identifiers into a plurality of clusters based on the similarity of the values of the respective invariant factor elements for each attribution identifier;
A third step of calculating an evaluation value that affects the result element for each transformation factor element for each cluster,
And a fourth step of extracting a transformation factor element according to the evaluation value of the result element.

Images