Information processing device and information processing method

Abstract

To enable accurate estimation of important explanatory variables for a target variable. [Solution] The information processing device of the present application includes a generation unit that generates a first decision tree model that predicts target data from first explanatory data corresponding to each of a first type of explanatory variables among the multiple types of explanatory variables based on a plurality of set data that is a pair of explanatory data corresponding to each of a plurality of types of explanatory variables and target data corresponding to a target variable, and generates a second decision tree model that predicts target data from second explanatory data corresponding to a second type of explanatory variables different from the first type of explanatory variables among the multiple types of explanatory variables based on set data corresponding to each of the leaf nodes of the generated first decision tree model, and an acquisition unit that acquires second variable importance generated when the generation unit generates the second decision tree model.

Description

[Technical field]

[0001] The present invention relates to an information processing device and an information processing method. [Background technology]

[0002] Conventionally, a technique for estimating explanatory variables important to a target variable is known. For example, a first value indicating the importance of each of a plurality of explanatory variables to a predetermined target variable is calculated using each of a plurality of m different methods. Then, a technique for selecting n explanatory variables based on the plurality of first values ​​corresponding to each of the m different methods is known. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2021-152854 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the above-described conventional techniques, a first value indicating the importance of each of a plurality of explanatory variables with respect to a predetermined objective variable is merely calculated using a plurality of different methods, and therefore it is not necessarily possible to accurately estimate explanatory variables important to the objective variable.

[0005] An object of the present application is to enable accurate estimation of explanatory variables important to a response variable. [Means for solving the problem]

[0006] The information processing device according to the present application includes a generation unit that generates a first decision tree model that predicts the target data from first explanatory data corresponding to each of a first type of explanatory variables among a plurality of types of explanatory variables based on a plurality of set data which is a pair of explanatory data corresponding to each of the plurality of types of explanatory variables and target data corresponding to a target variable, and generates a second decision tree model that predicts the target data from second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variables among the plurality of types of explanatory variables based on the set data corresponding to each of the leaf nodes of the generated first decision tree model, and an acquisition unit that acquires second variable importance generated when the generation unit generates the second decision tree model. Effect of the Invention

[0007] According to one aspect of the embodiment, it is possible to accurately estimate explanatory variables important to a dependent variable. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram for explaining the Net Promoter Score (NPS) according to the embodiment. [Diagram 2] FIG. 2 is a diagram for explaining the problem of variable importance in the comparison technique. [Diagram 3] FIG. 3 is a diagram for explaining an overview of information processing according to the embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of the configuration of the information processing device according to the embodiment. [Diagram 5] FIG. 5 is a diagram for explaining an example of information processing according to the embodiment. [Figure 6] FIG. 6 is a diagram for explaining an example of information processing according to the embodiment. [Figure 7] FIG. 7 is a flowchart showing a processing procedure performed by the information processing device according to the embodiment. [Figure 8]FIG. 8 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, a detailed description will be given of an information processing device and an information processing method according to the present application (hereinafter, referred to as an "embodiment") with reference to the drawings. Note that the information processing device and the information processing method according to the present application are not limited to the embodiment. In addition, the same components in the following embodiments are given the same reference numerals, and duplicated descriptions are omitted.

[0010] (Embodiment) 1. Introduction FIG. 1 is a diagram for explaining NPS (registered trademark) (Net Promoter Score) according to an embodiment. NPS is an index that quantifies customer loyalty for a given product or service. In other words, NPS can be said to be an index that quantifies customer satisfaction with a given product or service. For example, a survey is conducted for customers who have purchased a product, asking "To what extent would you like to recommend this product?" and asking the customer to select a score from 0 to 10 indicating the degree to which they would like to recommend the product to others. Then, the customer selects and answers a score corresponding to the degree to which they would like to recommend the product to others. Hereinafter, the score selected by the customer in the customer's response to the survey may be referred to as "NPS data".

[0011] Here, customers who select a score between 0 and 6 are presumed to be customers who are not very willing to recommend the product to others, and are therefore considered to be product critics. Customers who select a score between 7 and 8 are presumed to be customers who are only moderately willing to recommend the product to others, and are therefore considered to be product neutrals. Customers who select a score between 9 and 10 are presumed to be customers who are highly willing to recommend the product to others, and are therefore considered to be product promoters. NPS is calculated by subtracting the percentage (%) of customers who are detractors from the percentage (%) of customers who are promoters.

[0012] FIG. 2 is a diagram for explaining the problem of variable importance related to the comparison technique. In the comparison technique, a decision tree model is generated that is trained to output the NPS data of a customer corresponding to a target variable when customer data corresponding to an explanatory variable is input based on a plurality of set data that is a combination of customer data related to a customer and NPS data of the customer. Here, the customer data includes attribute information related to the attributes of the customer (hereinafter, may be referred to as "customer attribute information"). For example, the customer attribute information may include information indicating demographic attributes such as the customer's generation (teens, twenties, etc.; age may also be used), sex (male, female, etc.; hereinafter, male or female), occupation, place of residence, etc. In addition, the customer data includes non-attribute information (hereinafter, may be referred to as "customer non-attribute information") that is information about the customer and is a type of information different from the customer's attributes. For example, the customer non-attribute information may include information indicating whether or not the customer has a credit card of a specific company or information indicating whether or not the customer has subscribed to a specific service. The comparison technology generates a decision tree model trained to output customer NPS data corresponding to the objective variable when customer data including both customer attribute information and non-attribute information is input at once, and obtains the variable importance generated when the decision tree model is generated.

[0013] FIG. 2 shows how the variable importance of explanatory variables corresponding to attribute information on attributes such as age and sex of a customer and the variable importance of explanatory variables corresponding to non-attribute information on non-attributes different from attributes such as whether or not the customer has subscribed to a store smartphone support service or whether or not the customer has subscribed to a credit card of Company A are output at the same time. FIG. 2 also shows a problem with the variable importance related to the comparison technique, for example, that there is a correlation between explanatory variables corresponding to attribute information such as age of a customer and explanatory variables corresponding to non-attribute information such as whether or not the customer has subscribed to a credit card of Company A. For example, there is a correlation such that the older the customer is, the more likely the customer is to subscribe to a credit card of Company A. In this way, in the comparison technique, the correlation of explanatory variables (also called multicollinearity) is included in the dataset used to train the decision tree model, so that it was sometimes difficult to accurately estimate explanatory variables important to the objective variable.

[0014] FIG. 3 is a diagram for explaining an overview of information processing according to an embodiment. In FIG. 3, an information processing device according to an embodiment acquires a plurality of set data, which are combinations of explanatory data, which is customer data, and target data, which is customer NPS data. The information processing device also accepts, from a user, designation of explanatory variables corresponding to customer attribute information and explanatory variables corresponding to customer non-attribute information. The information processing device also generates a first decision tree model M1 that has been trained to output customer NPS data when customer attribute information corresponding to each of the explanatory variables accepted as explanatory variables corresponding to the customer attribute information is input. The information processing device also generates second decision tree models M2-1 to M2-N (N is a natural number) that have been trained to output customer NPS data included in each of the set data corresponding to each of the leaf nodes of the first decision tree model M1 when customer non-attribute information corresponding to each of the explanatory variables accepted as explanatory variables corresponding to the customer non-attribute information is input among customer data included in each of the set data corresponding to each of the leaf nodes of the first decision tree model M1. When there is no need to distinguish between the second decision tree models M2-1 to M2-N, they are referred to as the second decision tree model M2.

[0015] As described in FIG. 3, the information processing device according to the embodiment performs learning corresponding to explanatory variables corresponding to customer attribute information and learning corresponding to explanatory variables corresponding to customer non-attribute information in two stages. Specifically, the information processing device generates a first decision tree model that learns only the degree of contribution (e.g., importance) of explanatory variables corresponding to customer attribute information to the objective variable by learning a first decision tree model only with explanatory data corresponding to explanatory variables corresponding to customer attribute information. Next, the information processing device learns a second decision tree model that corresponds to explanatory variables corresponding to customer non-attribute information different from explanatory variables corresponding to customer attribute information based on set data classified into each of the leaf nodes of the first decision tree model that learns only the degree of contribution of explanatory variables corresponding to customer attribute information to the objective variable. That is, the information processing device generates a second decision tree model that learns the degree of contribution (e.g., importance) of explanatory variables corresponding to customer non-attribute information to the objective variable after eliminating the influence of the contribution of explanatory variables corresponding to customer attribute information to the objective variable.

[0016] This allows the information processing device to calculate the variable importance indicating the degree of contribution of the explanatory variable corresponding to the non-attribute information of the customer to the objective variable without being influenced by the explanatory variable corresponding to the attribute information of the customer. That is, the information processing device can calculate the variable importance of the explanatory variable corresponding to the non-attribute information of the customer while eliminating the influence of the correlation between the explanatory variable corresponding to the attribute information of the customer and the explanatory variable corresponding to the non-attribute information of the customer. This allows the information processing device to accurately estimate the explanatory variable that is important for the objective variable from the explanatory variables corresponding to the non-attribute information of the customer based on the variable importance of the explanatory variable corresponding to the non-attribute information of the customer while eliminating the influence of the explanatory variable corresponding to the attribute information of the customer. Therefore, the information processing device can accurately estimate the explanatory variable that is important for the objective variable.

[0017] In the following, a customer is referred to as a "subject to be analyzed." In addition, in the following, customer data is referred to as "information on a subject to be analyzed." In addition, in the present embodiment, a case will be described in which the objective data corresponding to the objective variable is NPS data, but the objective data corresponding to the objective variable is not limited to NPS data. For example, the objective data corresponding to the objective variable may be any data as long as it is data on a subject that a user wishes to analyze.

[0018] 2. Configuration of Information Processing Device 4 is a diagram showing an example of the configuration of an information processing device according to the embodiment. The information processing device 100 according to the embodiment includes a communication unit 110, a storage unit 120, and a control unit .

[0019] (Communication unit 110) The communication unit 110 is realized by a NIC (Network Interface Card), an antenna, etc. Specifically, the communication unit 110 is connected to a network N (not shown) by wire or wirelessly, and transmits and receives information to and from a terminal device used by a user.

[0020] (Storage unit 120) The storage unit 120 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. Specifically, the storage unit 120 stores an information processing program according to the embodiment.

[0021] (Control unit 130) The control unit 130 is a controller, and is realized, for example, by a CPU (Central Processing Unit), MPU (Micro Processing Unit), etc., executing various programs (for example, the information processing program according to the embodiment) stored in a storage device inside the information processing device 100 using a RAM as a working area. The control unit 130 is also a controller, and is realized, for example, by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0022] The control unit 130 has a receiving unit 131, a generating unit 132, an acquiring unit 133, and a providing unit 134 as functional units, and may realize or execute the information processing action described below. Note that the internal configuration of the control unit 130 is not limited to the configuration shown in Fig. 4, and may be other configurations as long as they perform the information processing described below. Also, each functional unit indicates a function of the control unit 130, and does not necessarily have to be physically distinct.

[0023] (Reception Department 131) The reception unit 131 receives various information. Specifically, the reception unit 131 receives from the user a designation of a first type of explanatory variable and a second type of explanatory variable different from the first type of explanatory variable. For example, the reception unit 131 displays the first type of explanatory variable in a selectable state from among a plurality of types of explanatory variables on the screen of a terminal device used by the user. For example, the reception unit 131 displays, as an example of the first type of explanatory variable, an explanatory variable having a low priority of acquiring variable importance by the user in a selectable state. In addition, for example, the reception unit 131 displays, as an example of the first type of explanatory variable, an explanatory variable corresponding to attribute information on the attributes of a subject to be analyzed (hereinafter, may be referred to as "attribute information of a subject") in a selectable state. For example, the attribute information of a subject may include information indicating demographic attributes such as the subject's age (teens, twenties, etc.; age may also be used), sex (male, female, etc.; hereinafter, male or female), occupation, place of residence, etc.

[0024] Next, the reception unit 131 receives information on explanatory variables selected by the user as the first type explanatory variable from among multiple types of explanatory variables from the terminal device. The reception unit 131 receives information on explanatory variables selected by the user as the first type explanatory variable, thereby receiving the designation of the first type explanatory variable from the user. For example, the first type explanatory variable is an explanatory variable corresponding to attribute information on the attributes of a subject to be analyzed. Alternatively, the first type explanatory variable may be an explanatory variable designated by the user as an explanatory variable with a low priority for acquiring variable importance.

[0025] The reception unit 131 may receive information on explanatory variables that are not selected by the user as the first type explanatory variable from among the multiple types of explanatory variables, from the terminal device. The reception unit 131 may receive information on explanatory variables that are not selected by the user as the first type explanatory variable as information on the second type explanatory variable. That is, the reception unit 131 may receive information on explanatory variables that are not selected by the user as the first type explanatory variable, thereby receiving a designation of the second type explanatory variable from the user.

[0026] Furthermore, the reception unit 131 displays, on the screen of the terminal device used by the user, explanatory variables of a second type in a selectable state from among multiple types of explanatory variables. For example, the reception unit 131 displays explanatory variables of a high priority for obtaining variable importance by the user as an example of the second type of explanatory variables in a selectable state. Furthermore, for example, the reception unit 131 displays, as an example of the second type of explanatory variables, explanatory variables corresponding to non-attribute information (hereinafter, may be referred to as "non-attribute information of the subject"), which is information on the subject to be analyzed and is a type of information different from the attributes of the subject, in a selectable state. For example, the non-attribute information of the subject may include information indicating non-attributes other than information indicating attributes such as the subject's age, sex, occupation, place of residence, etc. For example, the non-attribute information of the subject may include information indicating whether the subject has a credit card of a specific company or information indicating whether the subject subscribes to a specific service, etc., as information indicating non-attributes.

[0027] Next, the reception unit 131 receives information on explanatory variables selected by the user as second type explanatory variables from among multiple types of explanatory variables from the terminal device. The reception unit 131 receives information on explanatory variables selected by the user as second type explanatory variables, thereby receiving designation of the second type explanatory variables from the user. For example, the second type explanatory variable is information on a subject to be analyzed, and is an explanatory variable corresponding to non-attribute information, which is information of a type different from the subject's attributes. Alternatively, the second type explanatory variable may be an explanatory variable designated by the user as an explanatory variable with a high priority for acquiring variable importance.

[0028] The reception unit 131 may receive information on explanatory variables that are not selected by the user as the second type explanatory variable from among the multiple types of explanatory variables, from the terminal device. The reception unit 131 may receive information on explanatory variables that are not selected by the user as the second type explanatory variable as information on the first type explanatory variable. That is, the reception unit 131 may receive information on explanatory variables that are not selected by the user as the second type explanatory variable, thereby receiving a designation of the first type explanatory variable from the user.

[0029] (Generation unit 132) The generating unit 132 generates various information. Specifically, the generating unit 132 generates a first decision tree model that predicts the target data from first explanatory data corresponding to each of a first type of explanatory variables among the multiple types of explanatory variables, based on multiple set data that is a pair of explanatory data corresponding to each of a multiple types of explanatory variables and target data corresponding to a target variable, and generates a second decision tree model that predicts the target data from second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variable among the multiple types of explanatory variables, based on set data corresponding to each of the leaf nodes of the generated first decision tree model.

[0030] Here, the first decision tree model or the second decision tree model according to the present embodiment is a machine learning model generated using a known decision tree algorithm such as CART, CHAID, or C5.0. For example, the first decision tree model or the second decision tree model may be a classification tree or a regression tree. In addition, the first decision tree model or the second decision tree model may be a machine learning model generated by an ensemble method based on a decision tree algorithm. For example, the first decision tree model or the second decision tree model may be a machine learning model generated by an ensemble method combining a decision tree algorithm and a random forest, or a decision tree algorithm and gradient boosting.

[0031] More specifically, the storage unit 120 stores a plurality of set data which are pairs of explanatory data corresponding to each of a plurality of types of explanatory variables and objective data corresponding to an objective variable. For example, the storage unit 120 stores a plurality of set data which are pairs of explanatory data which is subject data related to a subject to be analyzed and objective data which is NPS data which is a response to a questionnaire for the subject (hereinafter, may be referred to as "subject's NPS data"). Note that the subject data includes both attribute information of the subject and non-attribute information of the subject. In addition, the generation unit 132 acquires a plurality of set data which are pairs of explanatory data corresponding to each of a plurality of types of explanatory variables and objective data corresponding to an objective variable. For example, the generation unit 132 acquires a plurality of set data which are pairs of explanatory data which is subject data and objective data which is the subject's NPS data. For example, the generation unit 132 acquires a plurality of set data by referring to the storage unit 120.

[0032] Next, when the generating unit 132 acquires a plurality of set data, the generating unit 132 generates a first decision tree model that predicts target data from first explanatory data corresponding to each of the first type of explanatory variables among the plurality of types of explanatory variables based on the plurality of set data. Specifically, the generating unit 132 generates a first decision tree model that is trained to predict target data included in each of the plurality of set data from first explanatory data corresponding to each of the explanatory variables accepted by the accepting unit 131 as the first type of explanatory variables among the explanatory data included in each of the plurality of set data. For example, the generating unit 132 generates a first decision tree model that is trained to predict NPS data of a subject from attribute information of the subject accepted by the accepting unit 131 as the first type of explanatory variables among the subject data included in each of the plurality of set data.

[0033] For example, the generating unit 132 generates a first decision tree model trained to output target data included in each of the plurality of set data when first explanatory data corresponding to each explanatory variable accepted as a first type explanatory variable by the accepting unit 131 is input among explanatory data included in each of the plurality of set data. For example, the generating unit 132 generates a first decision tree model trained to output NPS data of a subject included in each of the plurality of set data when attribute information of a subject is input among subject data included in each of the plurality of set data. In addition, when generating the first decision tree model, the generating unit 132 stores information that can identify the first decision tree model and information related to the first decision tree model in the storage unit 120 in association with each other.

[0034] For example, when a group of attribute information of a subject is input, the generating unit 132 generates a first decision tree including a root node corresponding to the group of attribute information of the subject. For example, when a group of attribute information of a subject is input, the generating unit 132 generates a first branching condition for dividing the group of attribute information of the subject into two groups based on whether the gender of the subject is male or female. For example, when the generating unit 132 generates the first branching condition, the generating unit 132 divides the group of attribute information of the subject into two groups: a group of attribute information of a subject whose gender is male (hereinafter, may be referred to as a "group of attribute information of male subjects") and a group of attribute information of a subject whose gender is female (hereinafter, may be referred to as a "group of attribute information of female subjects"). In addition, the generating unit 132 generates a first decision tree including a male node corresponding to the group of attribute information of a male subject, an edge branching from the root node to the male node, a female node corresponding to the group of attribute information of a female subject, and an edge branching from the root node to the female node.

[0035] Next, when the generating unit 132 divides the group of attribute information of the subjects into two groups, a group of attribute information of male subjects and a group of attribute information of female subjects, the generating unit 132 generates a second condition for dividing the group of attribute information of male subjects into two groups, for example, based on whether the subjects are in their 50s or older. For example, when the generating unit 132 generates the second condition, the generating unit 132 divides the group of attribute information of male subjects into two groups, a group of attribute information of male subjects whose age is 50s or older (hereinafter, may be referred to as "a group of attribute information of male subjects whose age is 50s or older") and a group of attribute information of male subjects whose age is younger than 50s (hereinafter, may be referred to as "a group of attribute information of male subjects whose age is younger than 50s"). In addition, the generation unit 132 generates a first decision tree that includes a male node in his 50s or older corresponding to a group of attribute information of male subjects in their 50s or older, an edge branching from the male node to the male node in his 50s or older, a male node under his 50s corresponding to a group of attribute information of male subjects under his 50s, and an edge branching from the male node to the male node under his 50s.

[0036] Furthermore, when the generating unit 132 divides the group of attribute information of the subjects into two groups, a group of attribute information of male subjects and a group of attribute information of female subjects, the generating unit 132 generates a third condition for dividing the group of attribute information of female subjects into two groups, for example, based on whether the subjects are in their 40s or older. For example, when the generating unit 132 generates the third condition, the generating unit 132 divides the group of attribute information of female subjects into two groups, a group of attribute information of female subjects whose age is 40s or older (hereinafter, may be referred to as "a group of attribute information of female subjects whose age is 40s or older") and a group of attribute information of female subjects whose age is younger than 40s (hereinafter, may be referred to as "a group of attribute information of female subjects whose age is younger than 40s"). In addition, the generation unit 132 generates a first decision tree that includes a women's node in their 40s or older corresponding to a group of attribute information of female subjects in their 40s or older, an edge branching from the women's node to the women's node in their 40s or older, a women's node under their 40s corresponding to a group of attribute information of female subjects under their 40s, and an edge branching from the women's node to the women's node under their 40s.

[0037] Here, the generation of a predetermined branch condition by the generating unit 132 corresponds to a high degree of contribution of the explanatory variable corresponding to the predetermined branch condition to the objective variable. In other words, the generation of a predetermined branch condition by the generating unit 132 means that the explanatory data corresponding to the explanatory variable corresponding to the predetermined branch condition is important in predicting the objective data corresponding to the objective variable. That is, the generation of a predetermined branch condition by the generating unit 132 corresponds to a high importance of the explanatory variable corresponding to the predetermined branch condition. Also, high importance of the explanatory variable corresponds to high variable importance corresponding to the explanatory variable. Here, the variable importance is an index representing the degree of contribution of the explanatory variable to the objective variable. In other words, the variable importance is an index representing how much the explanatory data corresponding to the explanatory variable has influenced in predicting the objective data corresponding to the objective variable. That is, high variable importance means that the explanatory data corresponding to the explanatory variable has a large influence in predicting the objective data corresponding to the objective variable.

[0038] For example, when the generating unit 132 generates a first branching condition regarding whether the subject's gender is male or female, it means that the explanatory variable corresponding to the subject's gender has a high importance. Also, when the explanatory variable corresponding to the subject's gender has a high importance, it means that the explanatory data corresponding to the subject's gender has a large influence on predicting the subject's NPS data.

[0039] Furthermore, the generation unit 132 generates a second branching condition regarding whether the male subject is in his 50s or older, which means that the explanatory variable corresponding to the male subject's age is highly important. Furthermore, the high importance of the explanatory variable corresponding to the male subject's age means that the explanatory data corresponding to the male subject's age has a large influence on predicting the male subject's NPS data.

[0040] Furthermore, the generation unit 132 generates a third branching condition regarding whether the female subject is in her 40s or older, which means that the explanatory variable corresponding to the female subject's age is highly important. Furthermore, the high importance of the explanatory variable corresponding to the female subject's age means that the explanatory data corresponding to the female subject's age has a large influence on predicting the NPS data of the female subject.

[0041] Furthermore, the generating unit 132 generates the first variable importance when generating the first decision tree model. For example, the generating unit 132 may calculate the first variable importance corresponding to each of the first type of explanatory variables by counting the number of times the explanatory variable is used when the branching condition of the decision tree is generated. Furthermore, the generating unit 132 may calculate the first variable importance corresponding to each of the first type of explanatory variables based on the reduction amount of Gini impurity in the explanatory data corresponding to each node of the decision tree. In this way, the generating unit 132 generates the first variable importance corresponding to each of the first type of explanatory variables by calculating the first variable importance corresponding to each of the first type of explanatory variables when generating the first decision tree model. Furthermore, when the generation unit 132 generates the first variable importance corresponding to each of the first type of explanatory variables, the generation unit 132 stores in the memory unit 120 information capable of identifying the first decision tree model, information capable of identifying each of the first type of explanatory variables, and the first variable importance corresponding to each of the first type of explanatory variables in association with each other.

[0042] Next, when the generating unit 132 generates the first decision tree model, the generating unit 132 generates a second decision tree model that predicts target data from second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variable among the multiple types of explanatory variables, based on set data corresponding to each of the leaf nodes of the generated first decision tree model. Here, the set data corresponding to each of the leaf nodes of the first decision tree model is, in other words, set data divided by each of the branching conditions corresponding to each node from the root node to the leaf node of the first decision tree model.

[0043] Specifically, the generating unit 132 generates a second decision tree model trained to predict target data included in each of the set data corresponding to each of the leaf nodes of the first decision tree model from second explanatory data corresponding to each of the explanatory variables accepted by the accepting unit 131 as the second type explanatory variable among explanatory data included in each of the set data corresponding to each of the leaf nodes of the first decision tree model. For example, when second explanatory data corresponding to each of the explanatory variables accepted by the accepting unit 131 as the second type explanatory variable among explanatory data included in each of the set data corresponding to each of the leaf nodes of the first decision tree model is input, the generating unit 132 generates a second decision tree model trained to output target data that is the NPS data of the subject when second explanatory data corresponding to non-attribute information of the subject is input among explanatory data that is subject data. In this way, the generating unit 132 generates second decision tree models corresponding to each of the leaf nodes of the first decision tree model based on the set data corresponding to each of the leaf nodes of the first decision tree model. When the generating unit 132 generates the second decision tree model, the generating unit 132 stores in the storage unit 120 information capable of identifying the second decision tree model, information on the second decision tree model, and information capable of identifying the first decision tree model from which the second decision tree model was generated, in association with each other.

[0044] For example, the leaf nodes of the above-mentioned first decision tree model are assumed to be four nodes: a male node in his 50s or older corresponding to a group of attribute information of male subjects in their 50s or older, a male node under his 50s corresponding to a group of attribute information of male subjects under his 50s, a female node in his 40s or older corresponding to a group of attribute information of female subjects in their 40s or older, and a female node under his 40s corresponding to a group of attribute information of female subjects under his 40s. In this case, the generation unit 132 generates each of the four second decision tree models M21 to M24 based on the set data corresponding to each of the male node in his 50s or older, the male node under his 50s, the female node in his 40s or older, and the female node under his 40s.

[0045] For example, the generating unit 132 generates the second decision tree model M21 based on set data corresponding to a group of attribute information of male subjects aged 50 or older. For example, the generating unit 132 generates the second decision tree model M21 trained to predict NPS data of subjects from non-attribute information of subjects accepted by the accepting unit 131 as a second type of explanatory variable, among subject data included in each of the set data corresponding to the group of attribute information of male subjects aged 50 or older.

[0046] Furthermore, the generating unit 132 generates a second decision tree model M22 based on set data corresponding to a group of attribute information of male subjects under 50. For example, the generating unit 132 generates the second decision tree model M22 trained to predict NPS data of subjects from non-attribute information of subjects accepted as a second type of explanatory variable by the accepting unit 131 among subject data included in each of the set data corresponding to the group of attribute information of male subjects under 50.

[0047] Furthermore, the generating unit 132 generates a second decision tree model M23 based on set data corresponding to a group of attribute information of female subjects aged 40 or older. For example, the generating unit 132 generates a second decision tree model M23 trained to predict NPS data of subjects from non-attribute information of subjects accepted as a second type of explanatory variable by the accepting unit 131 among subject data included in each of the set data corresponding to the group of attribute information of female subjects aged 40 or older.

[0048] Furthermore, the generating unit 132 generates a second decision tree model M24 based on set data corresponding to a group of attribute information of female subjects under 40. For example, the generating unit 132 generates the second decision tree model M24 trained to predict NPS data of subjects from non-attribute information of subjects accepted as a second type of explanatory variable by the accepting unit 131 among subject data included in each of the set data corresponding to the group of attribute information of female subjects under 40.

[0049] Furthermore, the generating unit 132 generates the second variable importance when the second decision tree model is generated. For example, the generating unit 132 may calculate the second variable importance corresponding to each of the second type of explanatory variables by counting the number of times the explanatory variable is used when the branching condition of the decision tree is generated. Furthermore, the generating unit 132 may calculate the second variable importance corresponding to each of the second type of explanatory variables based on the amount of reduction in Gini impurity in the explanatory data corresponding to each node of the decision tree. In this manner, the generating unit 132 generates the second variable importance by calculating the second variable importance corresponding to each of the second type of explanatory variables when the second decision tree model is generated. Furthermore, when the generating unit 132 generates the second variable importance corresponding to each of the second type of explanatory variables, the generating unit 132 stores in the storage unit 120 information capable of identifying the second decision tree model, information capable of identifying each of the second type of explanatory variables, and the second variable importance corresponding to each of the second type of explanatory variables in association with each other.

[0050] (Acquisition part 133) The acquisition unit 133 acquires various information. Specifically, the acquisition unit 133 acquires a first variable importance generated when the generation unit 132 generates the first decision tree model. For example, the acquisition unit 133 refers to the storage unit 120 and acquires a first variable importance corresponding to each of the first type of explanatory variables associated with information that can identify the first decision tree model when the generation unit 132 generates the first decision tree model. In addition, the acquisition unit 133 acquires a second variable importance generated when the generation unit 132 generates the second decision tree model. For example, the acquisition unit 133 refers to the storage unit 120 and acquires a second variable importance corresponding to each of the second type of explanatory variables associated with information that can identify the second decision tree model when the generation unit 132 generates the second decision tree model.

[0051] FIG. 5 is a diagram for explaining an example of information processing according to the embodiment. FIG. 5 explains information processing E1 according to the embodiment. In FIG. 5, the acquisition unit 133 acquires a first predicted value by the first decision tree model M1 generated by the generation unit 132. For example, the acquisition unit 133 refers to the storage unit 120 to acquire information on the first decision tree model M1 generated by the generation unit 132. The acquisition unit 133 also refers to the storage unit 120 to acquire set data for verification. Here, the set data for verification is a set of explanatory data for verification corresponding to each of a plurality of types of explanatory variables and objective data for verification corresponding to the objective variable. The set data for verification is the same type of set data as the plurality of set data (hereinafter, may be referred to as "set data for learning") used when generating the first decision tree model and the second decision tree model. When acquiring information on the first decision tree model M1 and the verification set data, the acquiring unit 133 acquires a first predicted value by inputting, to the first decision tree model M1, first explanatory data corresponding to each of the explanatory variables accepted as the first type of explanatory variable by the accepting unit 131, among the explanatory data included in each of the verification set data. Next, when acquiring the first predicted value, the acquiring unit 133 calculates a first error, which is an error between the verification target data included in the verification set data and the first predicted value, and verifies the prediction accuracy of the first decision tree model M1 based on the calculated first error.

[0052] 5, the acquiring unit 133 acquires a first variable importance corresponding to a first decision tree model M1 whose first error is below a first threshold. For example, the acquiring unit 133 calculates a first error corresponding to each of the multiple different first decision tree models M1 generated by the generating unit 132. Next, the acquiring unit 133 selects a first decision tree model M1 whose first error is below a first threshold from among the multiple different first decision tree models M1, and acquires a first variable importance corresponding to the selected first decision tree model M1.

[0053] Moreover, the acquiring unit 133 acquires a second predicted value by the second decision tree model generated by the generating unit 132. In FIG. 5, the generating unit 132 generates second decision tree models M2-1 to M2-3 corresponding to each of the leaf nodes of the first decision tree model M1 based on set data corresponding to each of the leaf nodes of the first decision tree model M1. For example, the acquiring unit 133 refers to the storage unit 120 to acquire information on the second decision tree models M2-1 to M2-3 generated by the generating unit 132. Furthermore, the acquiring unit 133 acquires set data for verification corresponding to each of the leaf nodes of the first decision tree model M1. For example, the acquiring unit 133 acquires set data for verification corresponding to each of the leaf nodes corresponding to the second decision tree models M2-1 to M2-3. For example, when acquiring information on the second decision tree model M2-1 and verification set data corresponding to the leaf node corresponding to the second decision tree model M2-1, the acquiring unit 133 inputs, to the second decision tree model M2-1, second explanatory data corresponding to each of the explanatory variables accepted as the second type of explanatory variable by the accepting unit 131, among the explanatory data included in each of the verification set data, to the second decision tree model M2-1, thereby acquiring a second predicted value by the second decision tree model M2-1. Furthermore, when acquiring the second predicted value by the second decision tree model M2-1, the acquiring unit 133 calculates a second error by the second decision tree model M2-1, which is an error between the verification target data included in the verification set data and the second predicted value by the second decision tree model M2-1, and verifies the prediction accuracy of the second decision tree model M2-1 based on the calculated second error by the second decision tree model M2-1.

[0054] Furthermore, when acquiring information on the second decision tree model M2-2 and verification set data corresponding to the leaf node corresponding to the second decision tree model M2-2, the acquiring unit 133 inputs, to the second decision tree model M2-2, second explanatory data corresponding to each of the explanatory variables accepted as the second type of explanatory variable by the accepting unit 131, among the explanatory data included in each of the verification set data, to the second decision tree model M2-2, thereby acquiring a second predicted value by the second decision tree model M2-2. Furthermore, when acquiring the second predicted value by the second decision tree model M2-2, the acquiring unit 133 calculates a second error by the second decision tree model M2-2, which is an error between the verification target data included in the verification set data and the second predicted value by the second decision tree model M2-2, and verifies the prediction accuracy of the second decision tree model M2-2 based on the calculated second error by the second decision tree model M2-2.

[0055] Furthermore, when acquiring information on the second decision tree model M2-3 and verification set data corresponding to the leaf node corresponding to the second decision tree model M2-3, the acquiring unit 133 inputs, to the second decision tree model M2-3, second explanatory data corresponding to each of the explanatory variables accepted as the second type of explanatory variable by the accepting unit 131, among the explanatory data included in each of the verification set data, to the second decision tree model M2-3, thereby acquiring a second predicted value by the second decision tree model M2-3. Furthermore, when acquiring the second predicted value by the second decision tree model M2-3, the acquiring unit 133 calculates a second error by the second decision tree model M2-3, which is an error between the verification target data included in the verification set data and the second predicted value by the second decision tree model M2-3, and verifies the prediction accuracy of the second decision tree model M2-3 based on the calculated second error by the second decision tree model M2-3.

[0056] Also, in FIG. 5, the acquisition unit 133 acquires a second variable importance corresponding to a second decision tree model in which the second error is below the second threshold. Specifically, the acquisition unit 133 acquires a second variable importance corresponding to each of the second decision tree models in which the second errors by the second decision tree models corresponding to each of the leaf nodes of the first decision tree model generated by the generation unit 132 are below the second threshold. For example, the acquisition unit 133 calculates a second error corresponding to each of the multiple different second decision tree models M2-1 generated by the generation unit 132. Next, the acquisition unit 133 selects a second decision tree model M2-1 in which the second error is below the second threshold from among the multiple different second decision tree models M2-1, and acquires a second variable importance corresponding to the selected second decision tree model M2-1. The acquisition unit 133 also calculates a second error corresponding to each of the multiple different second decision tree models M2-2 generated by the generation unit 132. Then, the acquisition unit 133 selects a second decision tree model M2-2 whose second error is below a second threshold from among the multiple different second decision tree models M2-2, and acquires a second variable importance corresponding to the selected second decision tree model M2-2. The acquisition unit 133 also calculates a second error corresponding to each of the multiple different second decision tree models M2-3 generated by the generation unit 132. Then, the acquisition unit 133 selects a second decision tree model M2-3 whose second error is below a second threshold from among the multiple different second decision tree models M2-3, and acquires a second variable importance corresponding to the selected second decision tree model M2-3.

[0057] Moreover, the acquiring unit 133 acquires the second variable importance corresponding to each of the second decision tree models generated by the generating unit 132, and calculates the total importance, which is the variable importance corresponding to each of the second type of explanatory variables, by summing the acquired second variable importance for each of the second type of explanatory variables. In FIG. 5, the acquiring unit 133 acquires the second variable importance corresponding to each of the second decision tree models M2-1 to M2-3 generated by the generating unit 132, and calculates the total importance, which is the variable importance corresponding to each of the second type of explanatory variables, by summing the acquired second variable importance for each of the second type of explanatory variables. For example, the acquiring unit 133 calculates the total importance, which is the variable importance corresponding to each of the second type of explanatory variables, by summing the second variable importance corresponding to the second decision tree model M2-1, the second variable importance corresponding to the second decision tree model M2-2, and the second variable importance corresponding to the second decision tree model M2-3 for each of the second type of explanatory variables.

[0058] Furthermore, the acquiring unit 133 acquires second predicted values ​​corresponding to each of the second decision tree models generated by the generating unit 132, and combines the acquired second predicted values ​​to calculate a combined predicted value that is an overall predicted value of the second decision tree models generated by the generating unit 132. For example, the acquiring unit 133 calculates the combined predicted value by calculating an average of the second predicted values ​​corresponding to each of the second decision tree models generated by the generating unit 132.

[0059] FIG. 6 is a diagram for explaining an example of information processing according to an embodiment. The generating unit 132 generates a plurality of different first decision tree models by varying parameters when generating a decision tree model, and generates a plurality of different second decision tree models corresponding to each of the generated plurality of different first decision tree models. FIG. 6 shows a state in which the information processing shown in FIG. 5 is repeated N times (N is a natural number of 2 or more) by varying a seed value when generating a decision tree model each time as a parameter when generating a decision tree model. Specifically, the generating unit 132 generates N different first decision tree models by varying a seed value when generating a first decision tree model each time. In addition, the generating unit 132 generates a plurality of different second decision tree models corresponding to each of the generated N different first decision tree models by varying a seed value when generating a second decision tree model each time.

[0060] Furthermore, the acquiring unit 133 calculates, for each of the first type of explanatory variables, an average of the first variable importance corresponding to each of the multiple different first decision tree models generated by the generating unit 132. In Fig. 6, the acquiring unit 133 calculates, for each of the first type of explanatory variables, an average of the first variable importance corresponding to each of the N different first decision tree models generated by the generating unit 132.

[0061] Furthermore, the acquiring unit 133 calculates, for each second type explanatory variable, an average of the total importance corresponding to each of the multiple different second decision tree models generated by the generating unit 132. In Fig. 6, the acquiring unit 133 calculates, for each second type explanatory variable, an average of the total importance corresponding to each of the N different second decision tree models generated by the generating unit 132.

[0062] Furthermore, the acquiring unit 133 calculates an average of the first predicted values ​​corresponding to each of the multiple different first decision tree models generated by the generating unit 132. In FIG. 6, the acquiring unit 133 calculates an average of the first predicted values ​​corresponding to each of the N different first decision tree models generated by the generating unit 132.

[0063] In addition, the acquiring unit 133 calculates an average of the combined predicted values ​​corresponding to each of the multiple different second decision tree models generated by the generating unit 132. In FIG. 6, the acquiring unit 133 calculates an average of the combined predicted values ​​corresponding to each of the N different second decision tree models generated by the generating unit 132.

[0064] (Provider 134) The providing unit 134 provides various information. Specifically, the providing unit 134 provides the user with information on the average of the first variable importance, the average of the total importance, the average of the first predicted value, and the average of the combined predicted value calculated by the acquiring unit 133. For example, the providing unit 134 displays information on the average of the first variable importance, the average of the total importance, the average of the first predicted value, and the average of the combined predicted value on the screen of the terminal device.

[0065] 3. Processing Procedure Fig. 7 is a flowchart showing a processing procedure by the information processing device according to the embodiment. In Fig. 7, the generating unit 132 of the information processing device 100 generates a plurality of different first decision tree models by varying parameters when generating the decision tree models (step S101). Next, the generating unit 132 of the information processing device 100 generates a plurality of different second decision tree models corresponding to each of the generated plurality of different first decision tree models by varying parameters when generating the decision tree models (step S102).

[0066] Furthermore, the acquisition unit 133 of the information processing device 100 acquires a first variable importance corresponding to each of the multiple different first decision tree models and a second variable importance corresponding to each of the multiple different second decision tree models (step S103). For example, the acquisition unit 133 of the information processing device 100 acquires a total importance as the second variable importance corresponding to each of the multiple different second decision tree models. Then, the acquisition unit 133 of the information processing device 100 calculates an average of the first variable importance and an average of the second variable importance (step S104). For example, the acquisition unit 133 of the information processing device 100 calculates an average of the total importance as the average of the second variable importance.

[0067] 4. Modifications In the above embodiment, the first type of explanatory variables are explanatory variables corresponding to attribute information, and the second type of explanatory variables are explanatory variables corresponding to non-attribute information. However, the combination of the first type of explanatory variables and the second type of explanatory variables is not limited to the combination of explanatory variables corresponding to attribute information and explanatory variables corresponding to non-attribute information. For example, when it is desired to predict the number of customers at an ice cream shop, the temperature on the day of sale, the amount of precipitation, the price of ice cream, the amount of advertisements related to ice cream, etc. can be considered as explanatory variables that affect the number of customers. In this case, for example, a user who wishes to predict the number of customers at an ice cream shop wishes to know the influence of explanatory variables that can be manually controlled, such as the price of ice cream and the amount of advertisements related to ice cream, on the number of customers, rather than the influence of explanatory variables that cannot be manually controlled, such as the temperature and precipitation on the day of sale. In this case, explanatory variables that cannot be manually controlled, such as the temperature and precipitation on the day of sale, are examples of explanatory variables that have a low priority for acquiring variable importance by the user. Also, explanatory variables that can be manually controlled, such as the price of ice cream and the amount of advertisements related to ice cream, are examples of explanatory variables that have a low priority for acquiring variable importance by the user. That is, the first type of explanatory variables may be explanatory variables that cannot be manually controlled, such as the temperature or amount of precipitation on the day of sale, while the second type of explanatory variables may be explanatory variables that can be manually controlled, such as the price of ice cream or the amount of advertising related to ice cream.

[0068] In the above embodiment, the case where the reception unit 131 receives the designation of the first type explanatory variable and the second type explanatory variable different from the first type explanatory variable from the user has been described, but the process of receiving the first type explanatory variable and the second type explanatory variable is not limited to this. Specifically, the storage unit 120 stores variable information in which information indicating the first type explanatory variable and / or information indicating the second type explanatory variable is in one-to-one correspondence with each of the multiple types of explanatory variables. Instead of receiving the designation of the first type explanatory variable and the second type explanatory variable from the user, the reception unit 131 refers to the variable information in the storage unit 120 and receives the explanatory variable in one-to-one correspondence with the information indicating the first type explanatory variable as the first type explanatory variable, and receives the explanatory variable in one-to-one correspondence with the information indicating the second type explanatory variable as the second type explanatory variable.

[0069] 5. Effects As described above, the information processing device 100 according to the embodiment includes the generating unit 132 and the acquiring unit 133. The generating unit 132 generates a first decision tree model that predicts the target data from the first explanatory data corresponding to each of the first type of explanatory variables among the multiple types of explanatory variables based on multiple set data that are pairs of explanatory data corresponding to each of the multiple types of explanatory variables and target data corresponding to the target variable, and generates a second decision tree model that predicts the target data from the second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variable among the multiple types of explanatory variables based on set data corresponding to each of the leaf nodes of the generated first decision tree model. The acquiring unit 133 acquires the second variable importance generated when the generating unit 132 generates the second decision tree model.

[0070] In this way, the information processing device 100 performs learning corresponding to the first type of explanatory variables and learning corresponding to the second type of explanatory variables in two stages. Specifically, the information processing device 100 generates a first decision tree model that has learned only the degree of contribution (e.g., importance) of the first type of explanatory variables to the objective variable by learning the first decision tree model with only explanatory data corresponding to the first type of explanatory variables. Next, the information processing device 100 learns a second decision tree model that corresponds to a second type of explanatory variables different from the first type of explanatory variables based on set data classified into each of the leaf nodes of the first decision tree model that has learned only the degree of contribution of the first type of explanatory variables to the objective variable. That is, the information processing device 100 generates a second decision tree model that has learned the degree of contribution (e.g., importance) of the second type of explanatory variables to the objective variable after eliminating the influence of the contribution of the first type of explanatory variables to the objective variable. Thereby, the information processing device 100 can calculate the variable importance indicating the degree of contribution of the second type explanatory variable to the objective variable without being influenced by the first type explanatory variable. That is, the information processing device 100 can calculate the variable importance of the second type explanatory variable while eliminating the influence of the correlation between the first type explanatory variable and the second type explanatory variable. Thereby, the information processing device 100 can accurately estimate the explanatory variable important to the objective variable from among the second type explanatory variables based on the variable importance of the second type explanatory variable while eliminating the influence of the first type explanatory variable, for example. Therefore, the information processing device 100 can accurately estimate the explanatory variable important to the objective variable. In addition, since the information processing device 100 can accurately estimate the explanatory variable important to the objective variable, it can contribute to the achievement of Goal 9 of the Sustainable Development Goals (SDGs), "Build resilience and promote innovation and infrastructure."

[0071] In addition, the acquisition unit 133 acquires second variable importance corresponding to each of the second decision tree models generated by the generation unit 132, and calculates a total importance, which is the variable importance corresponding to each of the second type of explanatory variables, by summing the acquired second variable importance for each of the second type of explanatory variables.

[0072] This enables the information processing device 100 to accurately estimate explanatory variables that are important to the target variable from among the explanatory variables of the second type, based on the total importance, which is the variable importance corresponding to each of the explanatory variables of the second type.

[0073] Furthermore, the generating unit 132 generates a plurality of different first decision tree models by varying parameters when generating the decision tree models, and generates a plurality of different second decision tree models corresponding to each of the generated plurality of different first decision tree models. The acquiring unit 133 calculates an average of the total importance corresponding to each of the plurality of different second decision tree models generated by the generating unit 132 for each of the second type explanatory variables.

[0074] This enables the information processing device 100 to accurately estimate explanatory variables that are important to the objective variable from among the explanatory variables of the second type, based on the average of the total importance calculated for each explanatory variable of the second type.

[0075] In addition, the acquisition unit 133 acquires a first predicted value by the first decision tree model generated by the generation unit 132, verifies the prediction accuracy of the first decision tree model based on a first error, which is the error between the verification target data included in the verification set data and the first predicted value, and acquires a second predicted value by the second decision tree model generated by the generation unit 132, and verifies the prediction accuracy of the second decision tree model based on a second error, which is the error between the verification target data and the second predicted value.

[0076] As a result, the information processing device 100 can acquire variable importance corresponding to a decision tree model with high prediction accuracy, and can accurately estimate explanatory variables that are important to the target variable from among the explanatory variables based on the more accurate variable importance.

[0077] Moreover, the acquiring unit 133 acquires a second variable importance corresponding to a second decision tree model in which the second error is lower than a second threshold value.

[0078] As a result, the information processing device 100 obtains variable importance corresponding to a decision tree model with high prediction accuracy, and therefore makes it possible to accurately estimate explanatory variables that are important to the target variable from among the explanatory variables based on the more accurate variable importance.

[0079] In addition, the first type of explanatory variables are explanatory variables that correspond to attribute information regarding the attributes of the subjects being analyzed, and the second type of explanatory variables are explanatory variables that correspond to non-attribute information, which is information about the subjects and is a different type of information from the attributes of the subjects.

[0080] In this way, the information processing device 100 performs learning corresponding to explanatory variables corresponding to the attribute information of the subject in two stages, learning corresponding to explanatory variables corresponding to the non-attribute information of the subject. Specifically, the information processing device 100 generates a first decision tree model that learns only the degree of contribution (e.g., importance) of explanatory variables corresponding to the attribute information of the subject to the objective variable by learning a first decision tree model only with explanatory data corresponding to explanatory variables corresponding to the attribute information of the subject. Next, the information processing device 100 learns a second decision tree model that corresponds to explanatory variables corresponding to the non-attribute information of the subject different from explanatory variables corresponding to the attribute information of the subject, based on set data classified into each of the leaf nodes of the first decision tree model that learns only the degree of contribution to the objective variable of explanatory variables corresponding to the attribute information of the subject. That is, the information processing device 100 generates a second decision tree model that learns the degree of contribution (e.g., importance) of explanatory variables corresponding to the non-attribute information of the subject to the objective variable after eliminating the influence of the contribution of explanatory variables corresponding to the attribute information of the subject to the objective variable. Thereby, the information processing device 100 can calculate the variable importance indicating the degree of contribution of the explanatory variable corresponding to the non-attribute information of the subject to the objective variable without being influenced by the explanatory variable corresponding to the attribute information of the subject. That is, the information processing device 100 can calculate the variable importance of the explanatory variable corresponding to the non-attribute information of the subject while eliminating the influence of the correlation between the explanatory variable corresponding to the attribute information of the subject and the explanatory variable corresponding to the non-attribute information of the subject. Thereby, the information processing device 100 can accurately estimate the explanatory variable that is important for the objective variable from the explanatory variables corresponding to the non-attribute information of the subject based on the variable importance of the explanatory variable corresponding to the non-attribute information of the subject while eliminating the influence of the explanatory variable corresponding to the attribute information of the subject, for example. Therefore, the information processing device 100 can accurately estimate the explanatory variable that is important for the objective variable.

[0081] In addition, the first type of explanatory variables are explanatory variables designated by the user as explanatory variables with a low priority for obtaining variable importance, and the second type of explanatory variables are explanatory variables designated by the user as explanatory variables with a high priority for obtaining variable importance.

[0082] In this way, the information processing device 100 performs learning in two stages: learning corresponding to explanatory variables designated by the user as explanatory variables with low priority for acquiring variable importance (hereinafter, may be referred to as "low priority explanatory variables"), and learning corresponding to explanatory variables designated by the user as explanatory variables with high priority for acquiring variable importance (hereinafter, may be referred to as "high priority explanatory variables"). Specifically, the information processing device 100 generates a first decision tree model that has learned only the degree of contribution (e.g., importance) of the low priority explanatory variables to the objective variable by learning the first decision tree model with only explanatory data corresponding to the low priority explanatory variables. Next, the information processing device 100 learns a second decision tree model that corresponds to the high priority explanatory variables based on the set data classified into each of the leaf nodes of the first decision tree model that has learned only the degree of contribution of the low priority explanatory variables to the objective variable. That is, the information processing device 100 generates a second decision tree model that learns the degree of contribution (for example, importance) of the high priority explanatory variable to the objective variable after eliminating the influence of the contribution of the low priority explanatory variable to the objective variable. Thereby, the information processing device 100 can calculate the variable importance indicating the degree of contribution of the high priority explanatory variable to the objective variable in a state where it is not influenced by the low priority explanatory variable. That is, the information processing device 100 can calculate the variable importance of the high priority explanatory variable in a state where the influence of the correlation between the low priority explanatory variable and the high priority explanatory variable is eliminated. Thereby, the information processing device 100 can accurately estimate an explanatory variable that is important for the objective variable from among the high priority explanatory variables based on the variable importance of the high priority explanatory variable in a state where the influence of the low priority explanatory variable is eliminated. Therefore, the information processing device 100 can accurately estimate an explanatory variable that is important for the objective variable.

[0083] [6. Hardware Configuration] The information processing device 100 according to the embodiment described above is realized by a computer 1000 having a configuration as shown in Fig. 8. Fig. 8 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device. The computer 1000 includes a CPU 1100, a RAM 1200, a ROM 1300, a HDD 1400, a communication interface (I / F) 1500, an input / output interface (I / F) 1600, and a media interface (I / F) 1700.

[0084] The CPU 1100 operates and controls each unit based on a program stored in the ROM 1300 or the HDD 1400. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started up, programs that depend on the hardware of the computer 1000, and the like.

[0085] The HDD 1400 stores programs executed by the CPU 1100, data used by such programs, etc. The communication interface 1500 receives data from other devices via a predetermined communication network and sends it to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the predetermined communication network.

[0086] The CPU 1100 controls output devices such as a display and a printer, and input devices such as a keyboard and a mouse, via the input / output interface 1600. The CPU 1100 acquires data from the input devices via the input / output interface 1600. The CPU 1100 also outputs generated data to the output devices via the input / output interface 1600.

[0087] The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads the program from the recording medium 1800 onto the RAM 1200 via the media interface 1700 and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase change rewritable Disc), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

[0088] For example, when the computer 1000 functions as the information processing device 100 according to the embodiment, the CPU 1100 of the computer 1000 executes programs loaded onto the RAM 1200 to realize the functions of the control unit 130. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, the CPU 1100 may obtain these programs from another device via a predetermined communication network.

[0089] Although some of the embodiments of the present application have been described in detail above with reference to the drawings, these are merely examples, and the present invention can be embodied in other forms that incorporate various modifications and improvements based on the knowledge of those skilled in the art, including the forms described in the Disclosure of the Invention section.

[0090] [7. Other] Furthermore, among the processes described in the above embodiments and modifications, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically by a known method. In addition, the information including the processing procedures, specific names, various data and parameters shown in the above documents and drawings can be changed arbitrarily unless otherwise specified. For example, the various information shown in each drawing is not limited to the illustrated information.

[0091] In addition, each component of each device shown in the figure is a functional concept, and does not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads, usage conditions, etc.

[0092] Furthermore, the above-described embodiments and modifications can be appropriately combined as long as the processing contents are not contradictory. [Explanation of symbols]

[0093] 100 Information processing device 110 Communications Department 120 Storage section 130 Control section 131 Reception 132 Generation part 133 Acquisition Department 134 Provision Department

Claims

1. a generation unit that generates a first decision tree model for predicting the target data from first explanatory data corresponding to each of a first type of explanatory variables among a plurality of types of explanatory variables, based on a plurality of set data which is a pair of explanatory data corresponding to each of a plurality of types of explanatory variables and target data corresponding to a target variable, and generates a second decision tree model for predicting the target data from second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variables among the plurality of types of explanatory variables, based on the set data corresponding to each of the leaf nodes of the generated first decision tree model; an acquisition unit that acquires a second variable importance generated when the generation unit generates the second decision tree model; An information processing device comprising:

2. The acquisition unit is The second variable importance corresponding to each of the second decision tree models generated by the generation unit is obtained, and the obtained second variable importance is summed up for each of the second type of explanatory variables to calculate a total importance which is the variable importance corresponding to each of the second type of explanatory variables. The information processing device according to claim 1 .

3. The generation unit is generating a plurality of different first decision tree models by varying parameters when generating a decision tree model, and generating a plurality of different second decision tree models corresponding to each of the generated plurality of different first decision tree models; The acquisition unit is Calculating an average of the total importance corresponding to each of the plurality of different second decision tree models generated by the generation unit for each of the second type explanatory variables. The information processing device according to claim 2 .

4. The acquisition unit is obtaining a first predicted value by the first decision tree model generated by the generation unit, and verifying prediction accuracy of the first decision tree model based on a first error, which is an error between verification target data included in a verification set data and the first predicted value; A second predicted value by the second decision tree model generated by the generation unit is obtained, and prediction accuracy of the second decision tree model is verified based on a second error, which is an error between the verification target data and the second predicted value. The information processing device according to claim 1 .

5. The acquisition unit is Obtain the second variable importance corresponding to the second decision tree model in which the second error is below a second threshold. The information processing device according to claim 4.

6. The first type of explanatory variable is an explanatory variable corresponding to attribute information regarding attributes of a subject to be analyzed, The second type of explanatory variable is an explanatory variable corresponding to non-attribute information, which is information about the subject and is different from the attributes of the subject. The information processing device according to claim 1 .

7. the first type of explanatory variable is an explanatory variable designated by a user as an explanatory variable with a low priority for acquiring variable importance, The second type of explanatory variable is an explanatory variable designated by the user as an explanatory variable having a high priority for obtaining variable importance. The information processing device according to claim 1 .

8. An information processing method implemented by a program executed by an information processing device, comprising: a generation step of generating a first decision tree model that predicts the target data from first explanatory data corresponding to each of a first type of explanatory variables among the multiple types of explanatory variables, based on a plurality of set data which is a pair of explanatory data corresponding to each of a plurality of types of explanatory variables and target data corresponding to a target variable, and generating a second decision tree model that predicts the target data from second explanatory data corresponding to a second type of explanatory variable different from the first type of explanatory variables among the multiple types of explanatory variables, based on the set data corresponding to each of the leaf nodes of the generated first decision tree model; an acquisition step of acquiring a second variable importance generated when the generating step generates the second decision tree model; An information processing method comprising:

Images