JP7231322B2 - LEARNING DEVICE, LEARNING METHOD, LEARNING PROGRAM AND PROGRAM - Google Patents

Description

The present invention relates to a learning device, a learning method, a learning program and a program .

Conventionally, there is known a technique of specifying similar users whose behavior is similar to that of a specific user using user information about the specific user. For example, based on user information about a specific user, generate a behavior prediction model for the specific user's behavior, and use the generated behavior prediction model to identify similar users whose behavior is similar to that of the specific user. There are known techniques for

JP 2016-177649

However, with the conventional technology described above, it is not always possible to identify similar users whose behavior is similar to that of a specific user. For example, in the conventional technology described above, if the user information regarding a specific user is insufficient, the accuracy of the behavior prediction model will be low. Not exclusively.

The present application has been made in view of the above, and aims to provide a learning device, a learning method, a learning program, learning data, and a model that can identify similar users whose behavior is similar to that of a specific user. and

A learning device according to the present application includes a selection unit that selects a second user different from the first user based on similarity with first user information of a first user who has performed a first action, and selection by the selection unit. learning a predictive model of the first action using the second user information of the second user and the first user information when the second user is assumed to have performed the first action and a learning unit.

According to one aspect of the embodiment, it is possible to identify similar users whose behavior is similar to that of a specific user.

FIG. 1 is a diagram illustrating a configuration example of a network system according to an embodiment. FIG. 2 is a diagram illustrating an example of learning processing by the learning device according to the embodiment. FIG. 3 is a diagram illustrating a configuration example of a learning device according to the embodiment; FIG. 4 is a diagram illustrating an example of a purchase history information storage unit according to the embodiment; FIG. 5 is a diagram illustrating an example of a search history information storage unit according to the embodiment; FIG. 6 is a flowchart showing a learning processing procedure by the learning device according to the embodiment. FIG. 7 is a diagram illustrating an example of learning processing according to a modification. FIG. 8 is a hardware configuration diagram showing an example of a computer that implements the functions of the learning device.

A learning device, a learning method, a learning program, learning data, and a mode for implementing a model (hereinafter referred to as "embodiments") according to the present application will be described in detail below with reference to the drawings. Note that the learning device, learning method, learning program, learning data and model according to the present application are not limited to this embodiment. Further, each embodiment can be appropriately combined within a range that does not contradict the processing contents. Also, in each of the following embodiments, the same parts are denoted by the same reference numerals, and overlapping descriptions are omitted.

[1. Network system configuration]
First, the configuration of the network system 1 according to the embodiment will be described using FIG. As shown in FIG. 1, the network system 1 according to the embodiment includes terminal devices 10 ₁ to 10 _n (generally referred to as “10”), an information providing device 20 and a learning device 100 . The terminal device 10, the information providing device 20, and the learning device 100 are each connected to the network N by wire or wirelessly. Although FIG. 1 shows an example in which the network system 1 includes one information providing device 20 and one learning device 100, the network system 1 includes a plurality of information providing devices 20 and , a plurality of learning devices 100 may be included.

The terminal device 10 is an information processing device used by a user. The terminal device 10 may be any type of information processing device including a desktop PC (Personal Computer), a tablet PC, a smart phone, and a PDA (Personal Digital Assistant).

The information providing device 20 is a server device that provides various information to the terminal device 10 . For example, the information providing device 20 provides a website to the terminal device 10 . Websites include shopping sites, news sites, auction sites, weather forecast sites, finance (stock prices) sites, route search sites, map providing sites, travel sites, restaurant introduction sites, survey sites, and web blogs. type of website. Also, the website may be any type of portal site, including content such as shopping, news, auctions, weather forecasts, finance (stock prices), route information, map information, surveys, and the like. However, the term "website" may be interpreted broadly to include "applications." For example, the term "portal" may be interpreted to mean an application portal that runs on a smart phone.

The information providing device 20 also receives user information related to the user of the terminal device 10 through the website. The information providing device 20 then provides the learning device 100 with the user information received through the website.

The learning device 100 is a server device that learns regularities and rules from user information provided from the information providing device 20 . For example, the learning device 100 uses user information provided from the information providing device 20 to learn a predictive model of user behavior.

[2. Learning process]
Next, an example of learning processing by the learning device according to the embodiment will be described with reference to FIG. FIG. 2 is an explanatory diagram showing an example of learning processing by the learning device according to the embodiment.

First, the study device 100 receives user information related to the user of the terminal device 10 from the information providing device 20 . For example, the learning device 100 receives from the information providing device 20 action history information about the user's actions on the website provided by the information providing device 20 . In the example of FIG. 1, the action history information includes user's purchase history information.

Next, learning device 100 selects a user who has performed a specific action from among multiple users included in the user information as a “correct user” to be used for learning a prediction model for the specific action. For example, the learning device 100 refers to the user's purchase history information and selects a user who purchased a specific product as a correct user to be used for learning a prediction model for purchasing the specific product. In the example of FIG. 1, the learning device 100 selects the user who purchased the personal computer as the correct user C1 to be used for learning the prediction model M1 regarding the purchase of the personal computer.

In FIG. 2, for purposes of illustration, multiple users who have purchased personal computers are shown as a set E1 of multiple correct users C1. In the example of FIG. 2, a set E1 of multiple correct users C1 includes user U1, user U2, and user U3. Although not shown in FIG. 2, a set E1 of multiple correct users C1 may include multiple correct users C1 other than user U1, user U2, and user U3.

Subsequently, learning device 100 selects a user who is different from the user who has performed the specific behavior from among the plurality of users included in the user information based on the similarity of the user who has performed the specific behavior with the user information. , as the "assistant user" to use in training the predictive model for that particular behavior. For example, learning device 100 selects a user who has performed an action similar to a specific action as an auxiliary user.

In one example, the learning device 100 refers to a plurality of products included in the user's purchase history information, and selects the assistant user based on whether or not the referenced products have a common upper classification. can be done. In the example of FIG. 1, the learning device 100 reads out the upper class "information terminal" corresponding to the personal computer from the user information, and assigns the user who purchased the product "smartphone" corresponding to the read superclass "information terminal" to the personal computer. is selected as the auxiliary user A1 to be used for learning the prediction model M1 regarding the purchase of .

In FIG. 2, for illustrative purposes, multiple users who have purchased smart phones are shown as a set E2 of multiple auxiliary users A1. In the example of FIG. 2, the set E2 of multiple auxiliary users A1 includes user U4, user U5, user U6 and user U7. Although not shown in FIG. 2, the set E2 of multiple auxiliary users A1 may include multiple auxiliary users A1 other than user U4, user U5, user U6, and user U7.

Subsequently, learning device 100 assigns the first weight to the user information of the correct user. In the example of FIG. 1, the learning device 100 assigns the weight "1" to the user information of the correct user C1 (step S11). Here, the weight is a parameter used when learning a predictive model for user behavior. Specifically, the weight is a parameter that indicates the magnitude of the influence of user information on the learning of the prediction model. As an example, the weight can be multiplied as a correction factor by the learning rate (also called learning factor) in gradient descent.

Subsequently, learning device 100 assigns a second weight, which is smaller than the first weight, to the user information of the assistant user when the assistant user has performed a specific action. In the example of FIG. 1, the learning device 100 assigns a weight smaller than "1" to the user information of the assistant user A1 when the assistant user A1 has purchased a personal computer (step S12).

Subsequently, the learning device 100 learns a prediction model of the specific behavior using the user information of the assistant user and the user information of the correct user when the assistant user performs the specific behavior. In the example of FIG. 1, a predictive model M1 for the purchase of a personal computer is obtained by using the user information of the auxiliary user A1 and the user information of the correct user C1 who purchased the personal computer. is learned (step S13). For example, the learning device 100 learns a prediction model M1 that uses search history information of a given user as an explanatory variable and the probability that the given user will purchase a personal computer as an objective variable.

After that, the learning device 100 provides the learned prediction model to the information providing device 20 . In the example of FIG. 1, the learning device 100 provides the information providing device 20 with a predictive model M1 regarding the purchase of a personal computer.

In this way, the learning device 100 according to the embodiment selects a user who is different from the correct user who performed a specific action based on the similarity with the user information of the correct user who performed the specific action. Select as an auxiliary user to use in training a predictive model for Then, the learning device 100 predicts the specific behavior using the user information of the assistant user when the assistant user has performed the specific behavior and the user information of the correct user who has performed the specific behavior. learn the model.

As a result, the learning device 100 according to the embodiment can identify similar users whose behavior is similar to that of a specific user. More specifically, generally, the accuracy of a predictive model for user behavior depends on the amount of user information about the correct user who performed the behavior. Even if the amount of user information of the correct user is insufficient, the learning device 100 can learn a predictive model of the user's behavior using the user information of the auxiliary user in addition to the user information of the correct user. can. As a result, the learning device 100 can accurately identify similar users whose behavior is similar to that of the correct user even when the amount of user information of the correct user is insufficient. The learning device 100 that implements such learning processing will be described in detail below.

[3. Configuration of learning device]
Next, the configuration of the learning device 100 according to the embodiment will be described using FIG. FIG. 3 is a diagram showing a configuration example of the learning device 100 according to the embodiment. As shown in FIG. 3 , learning device 100 includes communication section 110 , user information storage section 120 , and control section 130 . The providing apparatus 100 has an input unit (for example, keyboard, mouse, etc.) that receives various operations from an administrator or the like who uses the providing apparatus 100, and a display unit (liquid crystal display, etc.) for displaying various information. may

(Communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. As shown in FIG. 1, the communication unit 110 is connected to the network N by wire or wirelessly, and transmits and receives information to and from the terminal device 10 and the information providing device 20 via the network N.

(User information storage unit 120)
The user information storage unit 120 is implemented by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or flash memory, or a storage device such as a hard disk or optical disc. As shown in FIG. 3 , user information storage unit 120 stores purchase history information storage unit 121 , search history information storage unit 122 , and prediction model information storage unit 123 .

(Purchase history information storage unit 121)
The purchase history information storage unit 121 stores user's purchase history information. For example, the purchase history information storage unit 121 stores the user's purchase history information received by the reception unit 131 . Here, FIG. 4 shows an example of the purchase history information storage unit 121 according to the embodiment. In the example of FIG. 4, the purchase history information storage unit 121 stores "purchase history information" for each "user ID". As an example, the 'purchase history information' includes the items 'merchandise', 'purchase time', 'lower class' and 'higher class'.

“User ID” indicates an identifier for identifying the user of the terminal device 10 . When an HTTP (Hypertext Transfer Protocol) cookie or the like is transmitted and received between the information providing device 20 and the terminal device 10, the user ID may be set in the HTTP cookie.

“Product” indicates the product purchased by the user on the website provided by the information providing device 20 . "Purchase time" indicates the time when the user purchased the product. "Subclass" indicates a subdivision of the product group to which the product belongs. "Superior classification" indicates the product group to which the product belongs. That is, FIG. 4 shows that the user identified by the user ID "U1" purchased the product "PC X1" belonging to the lower class "PC" and the higher class "Information terminal" at "June 20, 2017 at 11:59 pm. It shows that it was purchased in minutes.

(Search history information storage unit 122)
The search history information storage unit 122 stores user search history information. For example, the search history information storage unit 122 stores user search history information received by the reception unit 131 . Here, FIG. 5 shows an example of the search history information storage unit 122 according to the embodiment. In the example of FIG. 5, the search history information storage unit 122 stores "search history information" for each "user ID". As an example, “search history information” includes items “search query” and “search time”.

“Search query” indicates a search query input by the user on the website provided by the information providing device 20 . "Search time" indicates the time when the user entered the search query. That is, FIG. 5 shows that the user identified by the user ID "U1" entered the search query "personal computer" at "June 20, 2017 at 11:59 PM".

(Prediction model information storage unit 123)
The prediction model information storage unit 123 stores learning data selected by the selection unit 132 . Here, the learning data stored in the prediction model information storage unit 123 includes user information of each user and a parameter associated with each user and indicating the degree of influence of each user on learning. Learning data, having an input layer and an output layer, a first element belonging to any layer from the input layer to the output layer other than the output layer, and the first element and the first element and the second element whose value is calculated based on the weight of the first element and the first element, with each element belonging to each layer other than the output layer as the first element for the user information input to the input layer. By inputting to the input layer of the model that performs calculation based on the weight, the output value indicating the calculation result is output from the output layer of the model, and the correct value and the output value are determined according to the parameters corresponding to the user information. It is learning data for making a computer function so as to perform learning based on comparison with. For example, the learning data stored in the prediction model information storage unit 123 includes user information of each of the correct user and the auxiliary user, and parameters associated with the correct user and the auxiliary user, respectively. Learning data containing parameters indicating the degree of influence in each learning, having an input layer and an output layer, and any layer from the input layer to the output layer other than the output layer A first element belonging to, a second element whose value is calculated based on the weight of the first element and the first element, and each element belonging to each layer other than the output layer for user information input to the input layer. As the first element, it is input to the input layer of the model that performs the calculation based on the first element and the weight of the first element, and the output value indicating the calculation result is output from the output layer of the model, corresponding to the user information Learning data for causing a computer to perform learning based on a comparison between a correct value determined according to a parameter to be used and an output value.

Also, the prediction model information storage unit 123 stores the prediction model learned by the learning unit 133 . Here, the prediction model stored in the prediction model information storage unit 123 includes an input layer to which user information associated with a parameter indicating the degree of influence in user learning is input, an output layer, and an output from the input layer An input layer including a first element belonging to a layer other than the output layer, which is any layer up to the layer, and a second element whose value is calculated based on the first element and the weight of the first element; Score used to provide user information by performing an operation based on the first element and the weight of the first element, with each element belonging to each layer other than the output layer as the first element for the information input to It is a model generated by operating a computer so as to be learned based on comparison between the correct value determined according to the parameter corresponding to the user information and the output value, which outputs the value of from the output layer. For example, the prediction model stored in the prediction model information storage unit 123 includes an input layer in which user information associated with a parameter indicating the degree of influence of user information on learning of the prediction model is input, and an output layer. and a first element belonging to a layer other than the output layer, which is any layer from the input layer to the output layer, and a second element whose value is calculated based on the first element and the weight of the first element. , with respect to the information input to the input layer, with each element belonging to each layer other than the output layer as the first element, and performing an operation based on the first element and the weight of the first element to obtain the user information Functioning a computer to learn based on comparison between a correct value determined according to a parameter corresponding to user information and an output value, which outputs a value related to the probability that the corresponding user performs a specific action from the output layer This is a model generated by

Here, it is assumed that the prediction model is realized by a regression model indicated by "y=a1*x1+a2*x2+...+ai*xi". In this case, the first element included in the prediction model corresponds to the input data (xi) such as x1 and x2. Also, the weight of the first element corresponds to the coefficient ai corresponding to xi. Here, the regression model can be viewed as a simple perceptron with an input layer and an output layer. When each model is regarded as a simple perceptron, the first element can be regarded as a node of the input layer, and the second element can be regarded as a node of the output layer.

It is also assumed that the prediction model is realized by a neural network having one or more intermediate layers, such as DNN (Deep Neural Network). In this case, the first element included in the prediction model corresponds to any node of the input layer or the intermediate layer. Also, the second element corresponds to the next node, which is a node to which the value is transmitted from the node corresponding to the first element. Also, the weight of the first element corresponds to the connection coefficient, which is the weight considered for the value transmitted from the node corresponding to the first element to the node corresponding to the second element.

(control unit 130)
The control unit 130 is a controller, and various programs stored in a storage device inside the learning device 100 are transferred to the RAM or the like by a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). It is realized by being executed as a work area. Also, the control unit 130 is a controller, and may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the control unit 130 includes a receiving unit 131, a selecting unit 132, a learning unit 133, and a providing unit 134, and implements or executes information processing functions and actions described below. . Note that the internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be another configuration as long as it performs information processing described later.

(Receiver 131)
The receiving unit 131 receives user information related to the user of the terminal device 10 from the information providing device 20 . For example, the receiving unit 131 receives, from the information providing device 20 , action history information regarding user actions on a website provided by the information providing device 20 . Specifically, the receiving unit 131 receives, from the information providing device 20, action history information including the user's purchase history information and the user's search history information.

The receiving unit 131 also stores the user information received from the information providing device 20 in the user information storage unit 120 . For example, the receiving unit 131 stores the user's purchase history information received from the information providing device 20 in the purchase history information storage unit 121 . Further, for example, the receiving unit 131 stores the user's search history information received from the information providing device 20 in the search history information storage unit 122 .

(Selection unit 132)
The selection unit 132 selects a user who has performed a specific action from among a plurality of users included in the user information as a "correct user" to be used for learning a prediction model for the specific action. Then, based on the similarity with the user information of the user who performed the specific behavior, the selection unit 132 selects a user different from the user who performed the specific behavior from among the plurality of users included in the user information. Select as an "auxiliary user" to use in training a predictive model for that particular behavior. Then, the selection unit 132 stores the user information of the selected correct user and the user information of the selected auxiliary user in the prediction model information storage unit 123 as learning data.

As an example for explanation, in the example shown in FIG. Consider the example of selecting an auxiliary user for training a predictive model.

First, the selection unit 132 refers to the subclass “PC” included in the purchase history information. Then, the selecting unit 132 selects the user who purchased the product belonging to the referred subclass “PC” as the correct user to be used for learning the “prediction model for purchase of PC”. For example, the selection unit 132 selects the user identified by the user ID "U1" as the correct user.

Next, the selection unit 132 refers to the upper class "information terminal" corresponding to the lower class "personal computer". After that, the selecting unit 132 selects users who have purchased products belonging to the referenced upper class “information terminal” from among the users who have not purchased products belonging to the lower class “personal computer” as “purchase of personal computers”. Select as an auxiliary user to use for learning the prediction model. In the example of FIG. 4, the selection unit 132 selects the user who purchased the product belonging to the subclass "smartphone" as the auxiliary user. For example, the selection unit 132 selects the user identified by the user ID "U4" as the auxiliary user.

(Learning unit 133)
The learning unit 133 assigns a first weight to the user information of the correct user selected by the selection unit 132 . Also, the learning unit 133 assigns a second weight smaller than the first weight to the user information of the assistant user selected by the selection unit 132 . Then, the learning unit 133 learns a predictive model of the specific behavior using the user information of the assistant user and the user information of the correct user when the assistant user performs the specific behavior. The learning unit 133 then stores the learned prediction model in the prediction model information storage unit 123 . The prediction model learning performed by the learning unit 133 can include, but is not limited to, linear regression, logistic regression, neural network, support vector machine, decision tree learning, and the like.

As an example for explanation, in the example shown in FIGS. 4 and 5, the learning unit 133 stores the search history information of the assistant user when the assistant user who purchased the smartphone purchased the personal computer, Let us consider an example of learning a "prediction model for purchase of a personal computer" by using the search history information of the correct user.

First, the learning unit 133 assigns a weight of "1" to the search history information of the user who purchased the personal computer. Here, the weight is a parameter used when learning the "prediction model for personal computer purchase". Specifically, the weight is a parameter that indicates the magnitude of the influence on the learning of the "prediction model for personal computer purchase".

In one example, the learning unit 133 uses the search history information stored in the search history information storage unit 122 to apply the gradient descent method to the number of searches for a given user's search query as an explanatory variable, It is possible to learn a predictive model whose objective variable is the probability that a user will purchase a personal computer. In this case, the weight is multiplied by the learning rate in the gradient descent method as a correction factor. For example, the step size of the gradient descent method is adjusted based on the weight multiplied by the learning rate as a correction factor.

Subsequently, the learning unit 133 assigns a predetermined weight smaller than "1" (for example, "0.8") to the search history information of the user who purchased the smart phone. In one example, the learning unit 133 determines a predetermined weight smaller than "1" based on the similarity between the search history information of the user who purchased the personal computer and the search history information of the user who purchased the smart phone. can do. For example, the learning unit 133 uses TF-IDF (Term Frequency-Inverse Document Frequency) to assign weights to search queries included in the search history information, and calculate a query vector whose components are the weights assigned to the search queries. do. Then, the learning unit 133, based on the cosine similarity between the query vector calculated from the search history information of the user who purchased the personal computer and the query vector calculated from the search history information of the user who purchased the smartphone, Determine a predetermined weight that is less than 1″.

After that, the learning unit 133 uses the search history information of the user who purchased the personal computer to make a prediction using the number of searches of a predetermined user's search query as an explanatory variable and the probability that the predetermined user will purchase a personal computer as an objective variable. learn the model. In addition, the learning unit 133 uses the search history information of the auxiliary user who purchased the smartphone as the search history information of the user who purchased the personal computer, sets the number of searches of the predetermined user's search query as an explanatory variable, and uses the search history information of the predetermined user. Learn a predictive model with the probability of purchasing a personal computer as the objective variable.

In some embodiments, the learning unit 133 can multiply the correct value (that is, the teacher output) included in the user information of the correct user selected by the selection unit 132 by the first weight. Also, the learning unit 133 can multiply the correct value included in the user information of the assistant user selected by the selection unit 132 by a second weight smaller than the first weight. For example, the learning unit 133 can multiply the correct value (eg, "1") of the user who purchased the personal computer by the first weight (eg, "1"). Also, for example, the learning unit 133 can multiply the correct value (eg, “1”) of the user who purchased the smart phone by a second weight (eg, “0.8”).

In some embodiments, the learning unit 133 generates user information of the correct user including the correct value multiplied by the first weight and user information of the assistant user including the correct value multiplied by the second weight. can be used to learn a predictive model for a particular behavior. As an illustrative example, a predictive model for personal computer purchases may be expressed as a linear regression model. In this example, the learning unit 133 can learn a linear regression model for personal computer purchases by applying a normal equation to the user information of the correct user and the user information of the auxiliary user.

(Providing unit 134)
The providing unit 134 provides the information providing device 20 with the prediction model learned by the learning unit 133 . Moreover, the output data of the prediction model learned by the learning unit 133 may be provided to the information providing device 20 . For example, the providing unit 134 inputs user information of a predetermined user to the input layer by information processing according to the prediction model. Then, the providing unit 134 propagates the input data to the intermediate layer and the output layer, thereby causing the output layer to output a value related to the probability that a given user will perform a specific action. In the examples of FIGS. 4 and 5, the providing unit 134 may provide the information providing device 20 with data representing the probability that a given user will purchase a personal computer.

In some embodiments, the providing unit 134 can use a model having an arbitrary structure such as the above-described regression model or neural network to calculate a value related to the probability that a given user will perform a specific action. . Specifically, the prediction model has a coefficient set so as to output a value related to the probability that a given user will perform a specific action when user information (for example, search history information) about a given user is input. be. The providing unit 134 can use such a prediction model to calculate a value related to the probability that a given user will perform a specific action. In addition, when the learning unit 133 performs a learning process using a GAN (Generative Adversarial Networks), the prediction model may be a model forming part of the GAN.

[4. Flow of learning processing]
Next, a procedure of learning processing by the learning device 100 according to the embodiment will be described. FIG. 6 is a flowchart showing a learning processing procedure by the learning device 100 according to the embodiment.

As shown in FIG. 6, the receiving unit 131 of the study device 100 receives user information related to the user of the terminal device 10 from the information providing device 20 (step S101).

Subsequently, the selecting unit 132 selects a user who has performed a specific action from among the plurality of users included in the user information received by the receiving unit 131, and selects the user who performed the specific action for learning the prediction model for the specific action. user” (step S102). Subsequently, the selection unit 132 selects the specific action from among the plurality of users included in the user information received by the reception unit 131 based on the similarity with the user information of the user who performed the specific action. A user different from the user who performed the action is selected as an "assistant user" to be used for learning the predictive model for the specific action (step S103).

Subsequently, the learning unit 133 assigns the first weight to the user information of the correct user selected by the selection unit 132 (step S104). Subsequently, the learning unit 133 assigns a second weight smaller than the first weight to the user information of the assistant user when the assistant user selected by the selection unit 132 performs a specific action ( step S105). Subsequently, the learning unit 133 learns a prediction model of the specific behavior using the user information of the assistant user and the user information of the correct user when it is assumed that the assistant user has performed a specific behavior ( step S106).

Subsequently, the providing unit 134 provides the prediction model learned by the learning unit 133 to the information providing device 20 (step S107).

[5. Modification]
The learning device 100 according to the above-described embodiments may be implemented in various forms other than the above embodiments. Therefore, another embodiment of the learning device 100 will be described below.

[5-1. Dynamic Selection of Auxiliary User]
The selection unit 132 of the learning device 100 described above selects the user who performed a behavior similar to the specific behavior performed by the correct user C1 as the assistant user A1. However, without being limited to this example, the selection unit 132 may select an auxiliary user based on similarity to the correct user information related to the characteristics of the correct user based on the user information of the correct user C1. In the following description, FIG. 7 is used to describe the dynamic selection of auxiliary users.

FIG. 7 is a diagram illustrating an example of learning processing according to a modification. First, the selection unit 132 selects a user who has performed a specific action from among a plurality of users included in the user information received by the reception unit 131, and selects a user who has performed a specific action to learn a prediction model for the specific action. ”. In the example of FIG. 7, the selection unit 132 selects the user who purchased the personal computer as the correct user C1 to be used for learning the prediction model M2 regarding the purchase of the personal computer. In FIG. 7, for illustrative purposes, multiple users who have purchased personal computers are shown as a set E1 of multiple correct users C1.

Subsequently, the learning unit 133 uses the user information of the correct user C1 to learn a prediction model for a specific action performed by the correct user C1. In the example of FIG. 7, the learning unit 133 uses the user information of the correct user C1 who purchased the personal computer to learn the prediction model M2 regarding the purchase of the personal computer. Then, the learning unit 133 extracts the feature information of the correct user C1 based on the learned output data of the prediction model M2. In the example of FIG. 7, the learning unit 133 extracts the number of searches for the search queries "personal computer" and "spec" as the characteristic information of the correct user C1.

In one embodiment, the learning unit 133 learns a logistic regression model in which the number of searches of a given user's search query is used as an explanatory variable and the probability that the given user purchases a personal computer is used as an objective variable. Then, the learning unit 133 determines main explanatory variables based on the learned regression coefficients of the logistic regression model. Then, the learning unit 133 extracts the determined main explanatory variables as feature information of the correct user C1.

Subsequently, the selection unit 132 selects a user different from the correct user C1 from among the plurality of users included in the user information as the correct user based on the similarity to the feature information of the correct user C1 extracted by the learning unit 133. It is selected as an "assistant user" to be used for learning a predictive model for a specific action performed by user C1. In the example of FIG. 7 , the selection unit 132 selects correct answers from users included in the search history information stored in the search history information storage unit 122 based on the number of searches for the search queries “personal computer” and “spec”. A user different from the user C1 is selected as an auxiliary user A2 to be used for learning a predictive model M3 regarding the purchase of a personal computer. In FIG. 7, for illustrative purposes, the selected auxiliary users are shown as a set E3 of auxiliary users A2.

Subsequently, the learning unit 133 calculates a similarity score for each auxiliary user A2 based on the similarity between the feature information of the correct user C1 and the user information of the auxiliary user A2 selected by the selection unit 132. In the example of FIG. 7, the learning unit 133 uses the number of searches for the search queries “personal computer” and “spec” by the assistant user A2 as explanatory variables, and the probability that the assistant user A2 will purchase a personal computer as the objective variable. A similarity score for each assistant user A2 is calculated based on the output data of .

Subsequently, the learning unit 133 assigns the first weight to the user information of the correct user C1. In the example of FIG. 7, the learning unit 133 assigns the weight "1" to the user information of the correct user C1.

Subsequently, based on the calculated similarity score for each assistant user A2, the learning unit 133 applies a first weight assign a second weight that is less than In the example of FIG. 7, the learning unit 133 assigns a second weight equal to the similarity score of the assistant user A2 to the user information of the assistant user A2.

Next, the learning unit 133 uses the user information of the assistant user A2 and the user information of the correct user C1 when the assistant user A2 has performed a specific action to create a prediction model of the specific action. Relearn. In the example of FIG. 7, the learning unit 133 uses the user information of the assistant user A2 when the assistant user A2 purchased the personal computer and the user information of the correct user C1 who purchased the personal computer to determine whether the personal computer was purchased. Learn a predictive model M3 for Based on the output data of the learned prediction model M3, the learning unit 133 extracts feature information of users belonging to the union E4 of the set E1 of the correct users C1 and the set E3 of the auxiliary users A2. In the example of FIG. 7, the learning unit 133 extracts the number of searches for the search queries "personal computer" and "price" as feature information of users belonging to the union E4.

Next, the selection unit 132 selects a user from among a plurality of users included in the user information based on the similarity between the feature information of the users belonging to the union E4 of the set E1 of the correct users C1 and the set E3 of the assistant users A2. , correct user C1 and auxiliary user A2 are selected as similar users R1. In the example of FIG. 7, the selection unit 132 selects correct answers from users included in the search history information stored in the search history information storage unit 122, based on the number of searches for the search queries “personal computer” and “price”. A user different from the user C1 and the auxiliary user A2 is selected as the similar user R1. In FIG. 7, for illustrative purposes, the selected similar users are shown as a set E5 of similar users R1.

After that, the learning unit 133 calculates a similarity score for each similar user R1 based on the similarity between the feature information of the users belonging to the union E4 and the user information of the similar user R1 selected by the selection unit 132. do. In the example of FIG. 7, the learning unit 133 uses the number of searches for the similar user R1's search queries "personal computer" and "price" as explanatory variables, and the probability that the similar user R1 will purchase a personal computer as the objective variable. A similarity score for each similar user R1 is calculated based on the output value of .

[5-2. others〕
Further, among the processes described in the above embodiments, all or part of the processes described as being automatically performed can be manually performed, or the processes described as being performed manually can be performed manually. All or part of this can also be done automatically by known methods. In addition, information including processing procedures, specific names, various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified.

Also, each component of each device illustrated is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured.

For example, part or all of the user information storage unit 120 shown in FIG. In this case, the learning device 100 acquires various types of information such as purchase history information and search history information by accessing the storage server.

[5-3. Hardware configuration]
Also, the providing apparatus 100 according to the above-described embodiments is implemented by a computer 1000 configured as shown in FIG. 8, for example. FIG. 8 is a diagram illustrating an example of a hardware configuration; A computer 1000 is connected to an output device 1010 and an input device 1020, and an arithmetic device 1030, a primary storage device 1040, a secondary storage device 1050, an output IF (Interface) 1060, an input IF 1070, and a network IF 1080 are connected via a bus 1090. have

The arithmetic device 1030 operates based on programs stored in the primary storage device 1040 and the secondary storage device 1050, programs read from the input device 1020, and the like, and executes various processes. The primary storage device 1040 is a memory device such as a RAM that temporarily stores data used by the arithmetic device 1030 for various calculations. The secondary storage device 1050 is a storage device in which data used for various calculations by the arithmetic device 1030 and various databases are registered, and is implemented by a ROM (Read Only Memory), HDD, flash memory, or the like.

The output IF 1060 is an interface for transmitting information to be output to the output device 1010 that outputs various types of information such as a monitor and a printer. It is realized by a connector conforming to a standard such as HDMI (registered trademark) (High Definition Multimedia Interface). Also, the input IF 1070 is an interface for receiving information from various input devices 1020 such as a mouse, keyboard, scanner, etc., and is realized by, for example, USB.

Note that the input device 1020 includes, for example, optical recording media such as CDs (Compact Discs), DVDs (Digital Versatile Discs), PDs (Phase change rewritable discs), magneto-optical recording media such as MOs (Magneto-Optical discs), and tapes. It may be a device that reads information from a medium, a magnetic recording medium, a semiconductor memory, or the like. Also, the input device 1020 may be an external storage medium such as a USB memory.

Network IF 1080 receives data from other devices via network N and sends the data to arithmetic device 1030, and also transmits data generated by arithmetic device 1030 via network N to other devices.

The arithmetic device 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070 . For example, arithmetic device 1030 loads a program from input device 1020 or secondary storage device 1050 onto primary storage device 1040 and executes the loaded program.

For example, when the computer 1000 functions as the learning device 100, the arithmetic device 1030 of the computer 1000 performs the functions of the control unit 130 by executing a program or data (eg, prediction model) loaded on the primary storage device 1040. Realize

[6. effect〕
As described above, the learning device 100 according to the embodiment has the selection unit 132 and the learning unit 133 . Based on the similarity with the first user information of the first user (for example, the correct user) who performed the first action (for example, a specific action), the selection unit 132 selects a second user different from the first user ( For example, select Auxiliary User. The learning unit 133 uses the first user information and the second user information of the second user selected by the selection unit 132 to perform the first action to perform the first action. learn a prediction model for

As a result, even when the amount of user information of the first user is insufficient, the learning device selects the second Using the user information of the two users, it is possible to learn a highly accurate prediction model for the behavior of the first user. As a result, learning device 100 can identify similar users whose behavior is similar to that of the specific user even when the amount of user information on the specific user is insufficient.

In addition, in the learning device 100 according to the embodiment, the selection unit 132 selects the second user who performed the second action similar to the first action, which is the first user information.

As a result, the learning device 100 can learn a highly accurate prediction model for the behavior of the first user using the user information of the second user whose behavioral tendency is similar to that of the first user. As a result, learning device 100 can identify similar users whose behavior is similar to that of the specific user even when the amount of user information on the specific user is insufficient.

Further, in the learning device 100 according to the embodiment, when the learning unit 133 assigns the first weight to the first user information and the second user selected by the selection unit 132 performs the first action, A second weight smaller than the first weight is assigned to the second user information of the second user to learn the prediction model.

As a result, learning device 100 adjusts the influence of the second user information of the second user on the learning of the prediction model using the second weight so as to match the learning of the prediction model for the behavior of the first user. can be corrected. As a result, learning device 100 can identify similar users whose behavior is similar to that of the specific user even when the amount of user information on the specific user is insufficient.

Further, in the learning device 100 according to the embodiment, the selection unit 132 selects the second user based on the similarity with the first user information related to the characteristics of the first user based on the first user information.

Thus, learning device 100 can learn a predictive model of the behavior of the first user using the user information of the second user whose features are similar to those of the first user. As a result, learning device 100 can identify similar users whose behavior is similar to that of a specific user.

In addition, in the learning device 100 according to the embodiment, the learning unit 133 assigns the first weight to the first user information, and also assigns the first user information and the second user information related to the characteristics of the first user based on the first user information. A second weight is assigned to the second user information of the second user when the second user selected by the selection unit 132 performs the first action based on the similarity between the user information and the user information. , to learn the prediction model.

Thereby, the learning device 100 can determine the influence of the second user's second user information on the learning of the prediction model so as to improve the accuracy of the prediction model. As a result, learning device 100 can identify similar users whose behavior is similar to that of a specific user.

As described above, some of the embodiments of the present application have been described in detail based on the drawings. It is possible to carry out the invention in other forms with modifications.

Further, the learning device 100 described above may be implemented by a plurality of server computers, and some functions may be implemented by calling an external platform or the like using an API (Application Programming Interface) or network computing. Flexible to change.

In addition, "means" described in the claims can be read as "section, module, unit" or "circuit". For example, the acquisition means can be read as an acquisition unit or an acquisition circuit.

1 network system 10 terminal device 20 information providing device 100 learning device 121 purchase history information storage unit 122 search history information storage unit 123 prediction model information storage unit 131 reception unit 132 selection unit 133 learning unit 134 provision unit

Claims (10)

extracting features of first user information including search history information of a first user who has purchased a predetermined object; a selection unit that selects a second user different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning unit that learns a model for predicting accuracy ;
A learning device comprising: The learning unit
Allocating a first weight to the first user information, allocating a second weight smaller than the first weight to the second user information of the second user selected by the selection unit, and assigning a second weight to the model to learn the
2. The learning device according to claim 1, wherein: extracting features related to the action history of the first user based on first user information including search history information of the first user who purchased a predetermined target, and extracting features related to the action history of the user information including the search history information; a selection unit that selects a second user who is similar to the feature related to the action history of the first user information and is different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning unit that learns a model for predicting accuracy ;
A learning device comprising: The learning unit
assigning a first weight to the first user information, assigning a second weight to the second user information of the second user selected by the selection unit, and learning the model;
4. The learning device according to claim 3, characterized in that: A computer implemented learning method comprising:
extracting features of first user information including search history information of a first user who has purchased a predetermined object; a selection step of selecting a second user different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning step of learning a model to predict accuracy ;
A learning method comprising: extracting features of first user information including search history information of a first user who has purchased a predetermined object; a selection procedure for selecting a second user different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning procedure for learning a model to predict accuracy ;
A learning program characterized by causing a computer to execute A computer implemented learning method comprising:
extracting features related to the action history of the first user based on first user information including search history information of the first user who purchased a predetermined target, and extracting features related to the action history of the user information including the search history information; a selection step of selecting a second user who is similar to the characteristics related to the action history of the first user information and is different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning step of learning a model to predict accuracy ;
A learning method comprising: extracting features related to the action history of the first user based on first user information including search history information of the first user who purchased a predetermined target, and extracting features related to the action history of the user information including the search history information; A selection procedure for selecting a second user who is similar to the characteristics related to the action history of the first user information and is different from the first user;
Similarity between the first user information and second user information including search history information of the second user is calculated by a predetermined method, and based on the characteristics of the first user information , the A predetermined weight is assigned to the first user information, a weight corresponding to the similarity calculated by the predetermined method is assigned to the second user information, and the user corresponding to the first user information purchases the predetermined object. a learning procedure for learning a model to predict accuracy ;
A learning program characterized by causing a computer to execute extracting features of first user information including search history information of a first user who has purchased a predetermined object; selecting a second user different from the first user;
A predetermined weight is assigned to the first user information based on the characteristics of the first user information , and the first user information calculated by a predetermined method and the second user information including search history information of the second user; A weight corresponding to the similarity is assigned to the second user information using the similarity, and a behavior prediction model is learned that predicts the probability that the user corresponding to the first user information will purchase the predetermined target. For a computer with training data that allows
having an input layer and an output layer, a first element belonging to any layer from the input layer to the output layer and belonging to a layer other than the output layer; and the first user information and the second user information input to the input layer, each element belonging to each layer other than the output layer is assigned to the first By inputting an element into the input layer of the action prediction model that performs a predetermined calculation based on the first element and a predetermined weight of the first element, an output value indicating a calculation result is input to the action prediction model. To perform learning based on a comparison between a correct value output from the output layer and a correct value determined according to a predetermined weight corresponding to the first user information and the second user information and the output value,
A program to make it work. extracting features of first user information including search history information of a first user who has purchased a predetermined object; selecting a second user different from the first user;
A first user who assigns a predetermined weight to the first user information as a behavior prediction model for predicting the probability that the user corresponding to the user information will purchase the predetermined object based on the characteristics of the first user information . information and the similarity between the first user information calculated by a predetermined method and the second user information including the search history information of the second user, the second user information corresponding to the similarity an input layer to which weighted second user information is input;
an output layer;
a first element belonging to any layer from the input layer to the output layer and belonging to a layer other than the output layer;
The behavior prediction model including a second element whose value is calculated based on the first element and a predetermined weight of the first element,
With respect to the first user information and the second user information input to the input layer, each element belonging to each layer other than the output layer is defined as the first element, and a predetermined weight is applied to the first element and the first element. By performing a predetermined calculation based on and, the score value used for providing information of each user is output from the output layer, and the predetermined weight corresponding to the first user information and the second user information As the action prediction model learned based on the comparison of the score value and the correct value determined according to
A program that makes a computer run.

Images