Learning device, learning method, and learning program

Abstract

To make effective use of the vast amount of data accumulated as a result of service usage. [Solution] The information processing device according to the present invention comprises an acquisition unit and a learning unit. When the acquisition unit receives conditions for identifying a group of users to be learned, it acquires characteristic information of users included in the user group identified by the conditions from a first storage unit that stores characteristic information indicating the attributes, interests, or usage status of each user, and acquires behavioral information of users included in the user group from a second storage unit that stores behavioral information indicating the service actions of each user. The learning unit trains a model on the relationship between the acquired characteristic information and behavioral information.

Description

[Technical Field]

[0001] The present application relates to a learning device, a learning method, and a learning program. [Background Art]

[0002] Conventionally, various online services have been provided via the Internet, and analysis of enormous data accumulated along with service usage has been performed. For example, a technique for analyzing customers in consideration of seasonal trends by using purchase history data or the like has been proposed (see, for example, Patent Document 1). [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2015-146145 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] However, conventional techniques still leave room for improvement. For example, high-level skills may be required to find and process useful data from the enormous amount of data accumulated along with service usage, which may prevent the promotion of utilization of data.

[0005] The present application has been made in view of the above, and an object thereof is to provide a learning device, a learning method, and a learning program that can promote utilization of the enormous data accumulated along with service usage. [Means for Solving the Problem]

[0006] The information processing device according to the present application comprises an acquisition unit and a learning unit. When the acquisition unit receives conditions for identifying a group of users to be learned, it acquires characteristic information of users included in the user group identified by the conditions from a first storage unit that stores characteristic information indicating the attributes, interests, or usage status of each user, and acquires behavioral information of users included in the user group from a second storage unit that stores behavioral information indicating the service actions of each user. The learning unit trains a model on the relationship between the acquired characteristic information and behavioral information. [Effects of the Invention]

[0007] According to one embodiment of the system, it is possible to make use of the vast amount of data accumulated as a result of using the service. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 shows an example of the configuration of an information processing system according to an embodiment. [Figure 2] Figure 2 is a diagram illustrating the overview of the learning process (part 1) according to the embodiment. [Figure 3] Figure 3 is a diagram illustrating the overview of the learning process (part 2) according to the embodiment. [Figure 4] Figure 4 is a diagram illustrating the overview of the score reproduction process according to the embodiment. [Figure 5] Figure 5 shows an example of hierarchical information according to the embodiment. [Figure 6] Figure 6 shows an example of the configuration of a model creation unit according to an embodiment. [Figure 7] Figure 7 shows an overview of the behavioral information stored in the user behavior database according to this embodiment. [Figure 8] Figure 8 is a diagram showing an overview of the information regarding features stored in the user feature database according to this embodiment. [Figure 9] Figure 9 is a diagram showing an overview of the trained model information stored in the model database according to this embodiment. [Figure 10]Figure 10 shows an example of the configuration of a score reproduction unit according to an embodiment. [Figure 11] Figure 11 is a flowchart showing an example of the procedure (part 1) of the learning process performed by the model creation unit according to the embodiment. [Figure 12] Figure 12 is a flowchart showing an example of the procedure (part 2) of the learning process performed by the model creation unit according to the embodiment. [Figure 13] Figure 13 is a flowchart showing an example of the steps of the score reproduction process performed by the score reproduction unit according to the embodiment. [Figure 14] Figure 14 is a hardware configuration diagram showing an example of a computer that implements the functions of the model creation unit according to the embodiment and modified example. [Modes for carrying out the invention]

[0009] The following describes in detail, with reference to the drawings, the learning device, the learning method, and the learning program according to the present application, and the embodiments for implementing them (hereinafter referred to as "embodiments"). Note that these embodiments do not limit the learning device, the learning method, and the learning program according to the present application. Furthermore, each embodiment can be combined as appropriate, provided that the processing content is not contradictory. Also, the same parts are denoted by the same reference numerals in each of the following embodiments, and redundant descriptions are omitted.

[0010] [Embodiment] [1. Overview of Information Processing Based on the Embodiment] (1-1. System Configuration) The following describes an overview of the information processing system SYS that performs information processing according to the embodiment. Figure 1 is a diagram showing an example of the configuration of the information processing system SYS according to the embodiment. The information processing system SYS shown in Figure 1 has a model creation unit 100, which is an example of a learning device according to the present application.

[0011] As shown in FIG. 1, the information processing system SYS according to the embodiment includes a model creation unit 100 and a score reproduction unit 200. The model creation unit 100 and the score reproduction unit 200 are connected to a predetermined network N such as the Internet by wire or wirelessly. The model creation unit 100 and the score reproduction unit 200 can mutually communicate with other devices such as an operator device 10 via the predetermined network N.

[0012] As the information processing according to the embodiment, the model creation unit 100 executes processing for creating a model that calculates a prediction score related to service usage of a user who uses various online services. The model creation unit 100 is typically a server device, but may be implemented by a mainframe, a workstation, or the like. Further, when the model creation unit 100 is implemented by a server device, it may be implemented by a single server device, or may be implemented by a cloud system or the like in which a plurality of server devices and a plurality of storage devices operate in cooperation.

[0013] The model creation unit 100 includes a user behavior DB 121, a user feature quantity DB 122, and a model DB 123.

[0014] The user behavior DB 121 stores behavior information (i.e., behavior logs) related to actions performed by each user, who is a service user using various online services. For example, examples of the behavior information include online service visit histories and conversion histories in online services. The various online services may include Internet connection, search services, travel information provision services, SNS (Social Networking Service), electronic commerce services, electronic payment services, online games, online banking services, online trading services, accommodation reservation services, ticket reservation services, video distribution services, music distribution services, news distribution services, map information services, route search services, route guidance services, route information services, operation information services, weather information services, and the like.

[0015] The user feature quantity DB 122 stores information related to feature quantities indicating characteristics such as the aforementioned attributes and interests of service users. For example, the information related to feature quantities stored in the user feature quantity DB 122 is aggregated as training data from various types of information recorded in various online services for training a model that calculates a score related to service usage by the aforementioned service users. The information related to feature quantities includes demographic attribute information such as the user's age, gender, address, and family structure, as well as psychographic attribute information such as interests and preferences.

[0016] The model DB 123 stores information of the trained model created by the model creation unit 100. The information of the trained model stored in the model DB 123 is created by causing the model to learn the relationship between the feature quantities associated with the aforementioned service users and the behavior information.

[0017] The score reproduction unit 200 reproduces a score that matches prediction conditions received from an operator using the information processing system SYS (for example, the operator OP shown in FIGS. 2 to 4).

[0018] The score reproduction unit 200 has a model DB 221. The model DB 221 stores information about the trained models created by the score reproduction unit 200. The model DB 221 stores multiple models that calculate scores related to the behavior of service users in various online services (examples of "specified services") when information linked to the aforementioned service users who use various online services is input.

[0019] The operator device 10 is an information processing terminal used by an operator OP who uses the information processing system SYS. For example, the operator device 10 can be implemented as a smartphone, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, or a PDA (Personal Digital Assistant). The operator device 10 can communicate with the model creation unit 100 and the score reproduction unit 200 of the information processing system SYS via the network N. The operator device 10 may also have communication functions for performing short-range wireless communication such as Bluetooth® or wireless LAN (Local Area Network).

[0020] (1-2. Learning Process (Part 1)) The following describes the overview of the learning process (part 1) according to the embodiment, using Figure 2. Figure 2 is a diagram illustrating the overview of the learning process (part 1) according to the embodiment. In the following description, operator OP can sometimes be identified with operator device 10. That is, operator OP can be rephrased as operator device 10.

[0021] Operator OP accesses the model creation unit 100 and specifies a query to the user behavior DB 121 as the behavior condition, which is a model creation condition (step S01).

[0022] When the model creation unit 100 receives a query specified by the operator OP (i.e., when it receives a query from the operator device 10), it retrieves logs of actions that match the specified query from the user behavior DB 121 (step S02).

[0023] Furthermore, the model creation unit 100 obtains the features of the relevant user, who is a service user corresponding to the acquired behavior log, and the features of the non-relevant user, who is a service user not corresponding to the acquired behavior log, from the user feature DB 122 (step S03).

[0024] Furthermore, the model creation unit 100 learns the model to output a high score when the features of the relevant user are input, and a low score when the features of an irrelevant user are input (step S04).

[0025] For example, the model creation unit 100 may train a model using a well-known machine learning method such as GBDT (Gradient Boosting Decision Tree), which is a type of supervised machine learning, and create a trained model. GBDT includes XGBoost (eXtreme Gradient Boosting) and LightGBM (Gradient Boosting Machine). Note that creating a trained model includes not only creating a new trained model but also updating an existing trained model. The model creation unit 100 registers the information of the created trained model in the model DB 123.

[0026] (1-3. Learning Process (Part 2)) The following describes the overview of the learning process (part 2) according to the embodiment, using Figure 3. Figure 3 is a diagram illustrating the overview of the learning process (part 2) according to the embodiment. In the following description, operator OP can sometimes be identified with operator device 10. That is, operator OP can be rephrased as operator device 10.

[0027] Operator OP accesses the model creation unit 100 and specifies the features that will be used as the model creation conditions (step S11).

[0028] When the model creation unit 100 receives the features specified by the operator OP (i.e., when it receives the features from the operator device 10), it identifies the relevant users who have features similar to the specified features, and the non-relevant users who do not have features similar to the specified features, from the user feature database 122 (step S12). Note that similar features include matching features.

[0029] Furthermore, the model creation unit 100 obtains logs of the identified user's actions from the user behavior DB 121 (step S13).

[0030] Furthermore, the model creation unit 100 learns the model to output a high score when the features of the relevant user are input, and a low score when the features of an irrelevant user are input (step S14). The model creation unit 100 registers the information of the created model in the model DB 123.

[0031] The model creation unit 100 may also accept a specification of the model training method from the operator OP as a model creation condition. Furthermore, the model creation unit 100 may train the model at predetermined intervals, such as daily, and update the model information stored in the model DB 123.

[0032] Furthermore, in the learning processes (1) and (2) described above, the model creation unit 100 uses known techniques such as XGBoost to estimate the importance of the behavioral logs and features (hereinafter referred to as "training data") used to create the model for each data type, and extracts the types of training data with high importance (for example, gender or product purchase date and time) based on the estimated importance. Then, the model creation unit 100 retrains the model using only the extracted training data. The model creation unit 100 may also use a bandit algorithm to include new types of training data as additional type training data in the training target. The model creation unit 100 improves the accuracy of the model by repeating this process.

[0033] Furthermore, in the learning processes (1) and (2) described above, the model creation unit 100 may learn a model using training data corresponding to each channel for each channel through which traffic flows into the online service. Channels may include email, advertisements, service domains, etc. Also, in the learning processes (1) and (2) described above, the model creation unit 100 may learn a model using training data corresponding to each layer of layered information, for each layer of layered information that divides the transitions of actions leading to a predetermined action in the online service used by the user into steps.

[0034] (1-4. Score reproduction process) The following describes the overview of the score reproduction process according to the embodiment, using Figure 4. Figure 4 is a diagram illustrating the overview of the score reproduction process according to the embodiment. In the following description, Operator OP can sometimes be identified with Operator Device 10. That is, Operator OP can be rephrased as Operator Device 10.

[0035] Operator OP accesses the score reproduction unit 200 and specifies the prediction conditions to be used for calculating the score (step S21). For example, one example of a prediction condition specified by Operator OP ("new condition") is "users likely to purchase product X".

[0036] When the score reproduction unit 200 receives the prediction conditions specified by the operator OP (i.e., receives them from the operator device 10), it selects a model that matches the received prediction conditions (step S22). For example, based on the comparison result between the content of the prediction conditions and the conditions corresponding to each of the multiple models stored in the model DB 221, the score reproduction unit 200 selects a model from among the multiple models to calculate a score used to reproduce a score that indicates the degree to which each of the conditions included in the prediction conditions is satisfied. The score reproduction unit 200 can, for example, extract each of the conditions included in the prediction conditions using natural language processing and automatically extract a model corresponding to each of the extracted conditions according to a predetermined rule base. Alternatively, the score reproduction unit 200 may extract each of the conditions included in the prediction conditions using natural language processing, present the models corresponding to the extracted conditions to the operator OP, and select the model selected by the operator OP.

[0037] Furthermore, as shown in Figure 4, the score reproduction unit 200 receives a prediction condition from the operator OP, such as "users likely to purchase product X," extracts "product X" and "purchase" as conditions included in the prediction condition, and selects an interest model corresponding to "product X" and a conversion prediction model corresponding to "purchase."

[0038] The interest model takes behavioral information associated with service users as input and outputs an interest score indicating interest in each product or service. The conversion prediction model takes features associated with service users as input and outputs a conversion prediction score that predicts, probabilistically, whether the user will purchase a product within a certain period.

[0039] Furthermore, the score reproduction unit 200 uses the scores calculated by each selected model to reproduce a score that matches the prediction conditions (step S23).

[0040] For example, if the score calculated by each selected model is a probability, the score reproduction unit 200 multiplies the prediction score calculated by the first model for each user and the prediction score calculated by the second model for each user to reproduce a score indicating the likelihood that the user will perform the action in question.

[0041] For example, if it is difficult to multiply the scores calculated by each of the selected models, the score reproduction unit 200 uses the ranking of the prediction scores calculated by the first model among the selected models for each user and the ranking of the prediction scores calculated by the second model among the selected models for each user to reproduce a score that predicts which users are likely to perform the target action. For example, as shown in Figure 4, the score reproduction unit 200 calculates a reproduction score as a combination of the ranking of the CV prediction score output by the CV prediction model for each user who is a service user and the ranking of the interest score output by the interest model for each user who is a service user.

[0042] Furthermore, the score reproduction unit 200 may accept, as a prediction condition, information about the channels through which users access online services. These channels may include email, advertising, service domains, etc. In this case, the score reproduction unit 200 selects a model corresponding to the channels specified as prediction conditions and performs score reproduction.

[0043] Furthermore, the score reproduction unit 200 may accept, as a prediction condition, the estimation of hierarchical information that sequentially divides the transitions of actions leading to a predetermined action in the online service used by the user. Figure 5 is a diagram showing an example of hierarchical information according to the embodiment. As shown in Figure 5, the hierarchical information may include no visit (new), awareness, visit, usage, app download (DL), and continuation. In this case, the score reproduction unit 200 selects a model corresponding to the funnel specified as the prediction condition and performs score reproduction.

[0044] Furthermore, if the score reproduction unit 200 has not identified a model for calculating the score used to reproduce the score indicating the degree to which each condition included in the prediction conditions is met, it may send a training request to the model creation unit 100 to train a model on the relationships between the conditions included in the prediction conditions. The score reproduction unit 200 may then register the model information received from the model creation unit 100 in the model DB 221. [2. Equipment configuration] (2-1. Model creation unit 100) The following describes an example of the functional configuration of each device included in the information processing system SYS according to the embodiment. First, an example of the functional configuration of the model creation unit 100 included in the information processing system SYS will be described. Figure 6 is a diagram showing an example of the configuration of the model creation unit 100 according to the embodiment. As shown in Figure 6, the model creation unit 100 includes a communication unit 110, a storage unit 120, and a control unit 130.

[0045] (Regarding Communications Unit 110) The communication unit 110 is implemented, for example, by a NIC (Network Interface Card). The communication unit 110 is connected to the network N by wire or wireless connection. The model creation unit 100 transmits and receives information to and from other devices such as the operator device 10 and the score reproduction unit 200 via the network N.

[0046] (Regarding memory unit 120) The storage unit 120 stores, for example, programs and data used for control and calculations by the control unit 130. For example, the storage unit 120 can be implemented using semiconductor memory elements such as RAM (Random Access Memory) or flash memory, or storage devices such as hard disks or optical discs. For example, the storage unit 120 has a user behavior DB 121, a user feature DB 122, and a model DB 123. Note that the storage unit 120 is not limited to the example shown in Figure 6, and can appropriately store data necessary for executing the information processing according to the embodiment.

[0047] (User Behavior DB121) The User Behavior DB 121 stores behavioral information (i.e., behavioral logs) related to the actions taken by each user who is a user of various online services. Figure 7 is a diagram showing an overview of the behavioral information stored in the User Behavior DB 121 according to this embodiment.

[0048] As shown in Figure 7, the behavioral information stored in the user behavior DB121 has a "User ID" field and a "Behavioral Information" field. These fields in the behavioral information are interconnected.

[0049] The "User ID" field stores the User ID, which is unique identification information assigned to each user to uniquely identify each user who is a user of various online services.

[0050] The "Behavioral Information" section stores logs of user behavior, such as page views (PV) and conversion data (CV) for various online services.

[0051] (User features DB122) The User Feature Database 122 stores information about features that represent the attributes, interests, and other characteristics of each user who is a user of various online services. Figure 8 is a diagram showing an overview of the feature information stored in the User Feature Database 122 according to this embodiment.

[0052] As shown in Figure 8, the feature information stored in the user feature DB122 has a "User ID" field and a "Feature" field. These fields in the feature information are interconnected.

[0053] The "User ID" field stores the User ID, which is unique identification information assigned to each user to uniquely identify each user who is a service user of various online services. For example, this User ID is the same information as the information stored in the "User ID" field of User Behavior DB121.

[0054] The "Features" section stores information about features that represent each user's attributes, interests, and other characteristics. This feature information is aggregated as training data from various pieces of information recorded in various online services to train a model that calculates service usage scores for service users. Feature information includes demographic attributes such as age, gender, address, and family structure, as well as psychographic attributes such as interests and preferences. Note that feature information may also be data that estimates and scores each user's attributes and interests.

[0055] (Model DB123) The model DB123 stores information about the trained model created by the model creation unit 100. Figure 9 is a diagram showing an overview of the trained model information stored in the model DB123 according to this embodiment.

[0056] As shown in Figure 9, the trained model information stored in Model DB123 includes the fields "Model ID," "Correspondence Conditions," and "Model Information." These fields in the trained model information are interconnected.

[0057] The "Model ID" field stores the Model ID, which is a unique identifier assigned to each model to uniquely identify it.

[0058] The "Correspondence Conditions" field stores information indicating the conditions under which the model is compatible. For example, possible conditions for a model include visit prediction, conversion prediction, and interest estimation. Further detailed conditions, such as travel site visit prediction or conversion prediction via advertising, may also be added to the model's conditions. Additionally, channel and hierarchical information (also known as a funnel) may be included.

[0059] The "Model Information" field stores information about the trained model. For example, if machine learning is performed using a neural network, the trained model information may include various types of information such as connection information, which describes how the nodes in each of the multiple layers that make up the neural network are connected to each other, and connection coefficients, which are multiplied by the numerical values ​​input and output between the connected nodes.

[0060] (Regarding the control unit 130) The control unit 130 is implemented by a CPU (Central Processing Unit) or MPU (Micro Processing Unit) executing various programs stored in the memory device inside the model creation unit 100 using RAM as the working area. Alternatively, the control unit 130 may be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0061] The control unit 130 shown in Figure 6 has an acquisition unit 131 and a learning unit 132, and these units realize or execute the functions and operations of the information processing described below. The control unit 130 may have multiple internal configurations divided into processing units that realize or execute the functions and operations of the information processing described below. Furthermore, the control unit 130 is not limited to the configuration shown in Figure 6, and may have other configurations as long as they perform the information processing described later, and may have other functional units other than those shown in Figure 6.

[0062] (Acquisition part 131) When the acquisition unit 131 receives user conditions regarding the user to be trained, it acquires feature quantities representing the characteristics of users who satisfy the user conditions from the user feature quantity DB 122, a database in which feature quantities representing the characteristics of each user are registered, and acquires action information representing the actions taken by users who satisfy the user conditions from the user action DB 121, a database in which action information representing the actions taken by each user is registered. The acquisition unit 131 then passes the acquired feature quantities and action information to the learning unit 132.

[0063] Furthermore, when the acquisition unit 131 receives a query from the operator OP, who is a user of the model creation unit 100, specifying the user's actions to be learned, it acquires action information corresponding to the actions specified by the query and acquires features associated with the acquired action information.

[0064] Furthermore, when the acquisition unit 131 receives a specification of user characteristics to be learned from the operator OP, who is a user of the model creation unit 100, it acquires feature quantities corresponding to the specified characteristics and acquires behavioral information associated with the acquired feature quantities.

[0065] Furthermore, the behavioral information acquired by the acquisition unit 131 may include information about the channels through which the user to be learned accesses a predetermined service when using that service. In addition, the behavioral information acquired by the acquisition unit 131 may include hierarchical information that sequentially categorizes the transitions of actions leading up to a predetermined action within a predetermined service used by the user to be learned.

[0066] (Learning Section 132) The learning unit 132 trains a model on the relationship between the features acquired by the acquisition unit 131 and the behavioral information. The learning unit 132 can train the model using known machine learning techniques. For example, the learning unit 132 may train the model using supervised learning, where the behavioral information stored in the user behavior DB 121 is the ground truth data and the features stored in the user feature DB 122 are the learning features.

[0067] The machine learning method used by the learning unit 132 is not particularly limited, but it can utilize methods such as learning using GBDT (Gradient Boosting Decision Tree) such as LightGBM, or preparing training data that links data (input information) with correct information (output information), and then using that training data with a multilayer neural network such as DNN (Deep Neural Network).

[0068] (2-2. Score reproduction unit 200) The following describes an example of the functional configuration of the score reproduction unit 200 provided in the information processing system SYS. Figure 10 is a diagram showing an example of the configuration of the score reproduction unit 200 according to the embodiment. As shown in Figure 10, the score reproduction unit 200 has a communication unit 210, a storage unit 220, and a control unit 230.

[0069] (Regarding Communications Unit 210) The communication unit 210 is implemented, for example, by a NIC (Network Interface Card). The communication unit 210 is connected to the network N by wire or wireless. The score reproduction unit 200 transmits and receives information to and from other devices such as the operator device 10 and the model creation unit 100 via the network N.

[0070] (Regarding memory unit 220) The storage unit 220 stores, for example, programs and data used for control and calculations by the control unit 230. For example, the storage unit 220 can be implemented using semiconductor memory elements such as RAM (Random Access Memory) or flash memory, or storage devices such as hard disks or optical discs. For example, the storage unit 220 has a model DB221. Note that the storage unit 220 is not limited to the example shown in Figure 10, and can appropriately store data necessary for executing the information processing according to the embodiment.

[0071] (Model DB221) The model database 221 stores information about the trained models created by the model creation unit 100. The trained model information stored in the model database 221 is retrieved from the model creation unit 100.

[0072] (Regarding the control unit 230) The control unit 230 is implemented by a CPU (Central Processing Unit) or MPU (Micro Processing Unit) executing various programs stored in the memory device inside the score reproduction unit 200 using RAM as the working area. Alternatively, the control unit 230 may be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

[0073] The control unit 230 shown in Figure 10 includes a reception unit 231, a selection unit 232, a reproduction unit 233, a learning request unit 234, and a registration unit 235. These units realize or execute the information processing functions and operations described below. The control unit 230 may have multiple internal configurations divided into processing units that realize or execute the information processing functions and operations described below. Furthermore, the control unit 230 is not limited to the configuration shown in Figure 10; it may have other configurations as long as they perform the information processing described later, and may have other functional units other than those shown in Figure 10.

[0074] (Reception Desk 231) The reception unit 231 receives the specification of new conditions to be used for score calculation from the operator OP, who is a user of the score reproduction unit 200. For example, the reception unit 231 receives the specification of prediction conditions related to user behavior from the operator OP.

[0075] Furthermore, the reception unit 231 may accept, as a prediction condition, information about the channels through which users access various online services. Also, the reception unit 231 may accept, as a prediction condition, estimation of hierarchical information (also called a funnel) that divides the transitions of actions leading up to a predetermined action in the various online services used by the user into steps.

[0076] (Selection Department 232) The selection unit 232, based on the comparison result between the content of the new conditions received by the reception unit 231 and the conditions corresponding to each of the multiple models stored in the model DB 221, selects models from among the multiple models to calculate a score used to reproduce the degree to which each of the conditions included in the new conditions is satisfied. For example, the selection unit 232, based on the comparison result between the content of the prediction conditions received by the reception unit 231 and the conditions corresponding to each of the multiple models, selects a first model and a second model from among the multiple models as the respective models.

[0077] (Reproduction section 233) The reproduction unit 233 reproduces a score indicating the degree to which new conditions are met, using the scores calculated by each model selected by the selection unit 232. For example, the reproduction unit 233 multiplies the score calculated by the first model for each user and the score calculated by the second model for each user to reproduce a score indicating the likelihood that each user will perform actions on various online services. Alternatively, for example, the reproduction unit 233 uses the ranking of the prediction scores calculated by the first model for each user and the ranking of the prediction scores calculated by the second model for each user to reproduce a score for predicting which users are most likely to perform actions on various online services.

[0078] (Learning Request Department 234) If the selection unit 232 has not selected any models to calculate the scores used to reproduce the degree to which each of the conditions included in the new conditions is satisfied, the learning request unit 234 sends a learning request to the model creation unit 100 to train a model on the relationships between the conditions included in the new conditions.

[0079] (Registration Section 235) The registration unit 235 registers the model information received from the model creation unit 100 (an example of an "external device") into the model DB 221.

[0080] [3. Processing procedure according to the embodiment] (3-1. Learning Process Procedure (Part 1)) The following describes the procedure (1) of the learning process performed by the model creation unit 100 according to the embodiment. First, the procedure (1) of the learning process performed by the model creation unit 100 according to the embodiment will be described using Figure 11. Figure 11 is a flowchart of an example of the procedure (1) of the learning process performed by the model creation unit 100 according to the embodiment. The procedure shown in Figure 11 is performed by the control unit 130 of the model creation unit 100. The procedure shown in Figure 11 is repeatedly performed while the model creation unit 100 is running.

[0081] As shown in Figure 11, the acquisition unit 131 receives a query from the operator OP specifying the user behavior to be learned (step S101). For example, this query is a query to the user behavior DB 121.

[0082] When the learning unit 132 receives a query specified by the operator OP, it retrieves logs of actions that match the specified query from the user behavior DB 121 (step S102).

[0083] Furthermore, the learning unit 132 obtains the features of the relevant user, who is a service user corresponding to the acquired action log, and the features of the non-relevant user, who is a service user not corresponding to the acquired action log, from the user feature DB 122 (step S103).

[0084] Furthermore, the learning unit 132 learns the relationship between behavioral information and features using a predetermined model (step S104). For example, the learning unit 132 uses an arbitrary machine learning method to train a model that outputs a high score when the features of the relevant user are input, and a low score when the features of an irrelevant user are input.

[0085] Furthermore, the learning unit 132 uses known techniques to estimate the importance of the behavior logs and features used to create the model for each type of data, and based on the estimated importance, extracts the types of behavior logs or features with high importance (for example, gender or product purchase date and time) (step S105).

[0086] Furthermore, the learning unit 132 retrains the model using only the extracted training data (step S106).

[0087] Furthermore, the learning unit 132 registers the retrained model in the model DB 123 (step S107), and then terminates the learning process procedure shown in Figure 11.

[0088] (3-2. Learning Process Procedure (Part 2)) The following describes the procedure (part 2) of the learning process performed by the model creation unit 100 according to the embodiment. First, the procedure (part 2) of the learning process performed by the model creation unit 100 according to the embodiment will be described using Figure 12. Figure 12 is a flowchart of an example of the procedure (part 2) of the learning process performed by the model creation unit 100 according to the embodiment. The procedure shown in Figure 12 is performed by the control unit 130 of the model creation unit 100. The procedure shown in Figure 12 is repeatedly performed while the model creation unit 100 is running. Note that the procedure (part 2) of the learning process shown in Figure 12 differs from the procedure (part 1) of the learning process shown in Figure 11 in steps S201 to S203.

[0089] As shown in Figure 12, the acquisition unit 131 receives the specification of the user's features to be learned from the operator OP (step S201).

[0090] When the learning unit 132 receives the feature specified by the operator OP, it identifies the relevant users who have feature quantities similar to the specified feature quantities, and the non-relevant users who do not have feature quantities similar to the specified feature quantities, from the user feature quantity DB 122 (step S202).

[0091] Furthermore, the learning unit 132 obtains logs of the identified user's actions from the user behavior DB 121 (step S203).

[0092] Furthermore, the learning unit 132 learns the relationship between behavioral information and features using a predetermined model (step S204). For example, the learning unit 132 uses an arbitrary machine learning method to train a model that outputs a high score when the features of the relevant user are input, and a low score when the features of an irrelevant user are input.

[0093] Furthermore, the learning unit 132 uses known techniques to estimate the importance of the behavior logs and features used to create the model for each type of data, and based on the estimated importance, extracts the types of behavior logs or features with high importance (for example, gender or product purchase date and time) (step S205).

[0094] Furthermore, the learning unit 132 retrains the model using only the extracted training data (step S206).

[0095] Furthermore, the learning unit 132 registers the retrained model in the model DB 123 (step S107), and then terminates the learning process procedure shown in Figure 11.

[0096] (3-3. Procedure for score reproduction) The following describes the procedure for score reproduction processing performed by the score reproduction unit 200 according to the embodiment. First, the procedure for score reproduction processing performed by the score reproduction unit 200 according to the embodiment will be described using Figure 13. Figure 13 is a flowchart showing an example of the procedure for score reproduction processing performed by the score reproduction unit 200 according to the embodiment. The procedure shown in Figure 13 is performed by the control unit 230 of the score reproduction unit 200. The procedure shown in Figure 13 is repeatedly performed while the score reproduction unit 200 is running.

[0097] As shown in Figure 13, the reception unit 231 receives the operator OP's specification of the prediction conditions to be used for score calculation (step S301).

[0098] The selection unit 232, based on the comparison result between the content of the prediction conditions specified by operator OP and the conditions corresponding to each of the multiple models stored in the model DB 221, selects from among the multiple models to calculate a score used to reproduce the score that indicates the degree to which each of the conditions included in the prediction conditions is satisfied (step S302).

[0099] The reproduction unit 233 uses the scores calculated by each model selected by the selection unit 232 to reproduce a score indicating the degree to which the new conditions are met (step S303), and then terminates the score reproduction process procedure shown in Figure 13.

[0100] [4. Variant] The information processing apparatus, information processing method, and information processing program according to this application may be implemented in various different forms other than the embodiments described above. Modifications of the above embodiments will be described below.

[0101] In the above embodiment, an example was described in which the information processing system SYS according to the embodiment is configured to include a model creation unit 100 and a score reproduction unit 200, but the system is not limited to this example. For example, the information processing system SYS may be configured to include a single information processing device in which the model creation unit 100 and the score reproduction unit 200 are functionally and physically integrated.

[0102] [5. Effects] The model creation unit 100 according to this embodiment includes an acquisition unit 131 and a learning unit 132. When the acquisition unit 131 receives user conditions regarding the user to be learned, it acquires feature quantities that represent the characteristics of a user who satisfies the conditions from the user feature quantity DB 122, in which feature quantities that represent the characteristics of each user are registered, and acquires action information that represents the actions taken by a user who satisfies the user conditions from the user action DB 121, in which action information that represents the actions taken by each user is registered. The learning unit 132 trains the model on the relationship between the feature quantities acquired by the acquisition unit 131 and the action information.

[0103] Furthermore, when the acquisition unit 131 receives a query from the operator OP, who is a user of the model creation unit 100, specifying the user's actions to be learned, it acquires action information corresponding to the action specified by the query from the user action DB 121 and acquires the features associated with the acquired action information from the user feature DB 122.

[0104] Furthermore, when the acquisition unit 131 receives a specification of user characteristics to be learned from the operator OP, who is a user of the model creation unit 100, it acquires feature quantities corresponding to the specified characteristics from the user feature quantity DB 122 and acquires behavioral information associated with the acquired feature quantities from the user behavior DB 121.

[0105] Thus, the model creation unit 100 according to this embodiment can effectively utilize the big data accumulated in various online services by using the processing performed by each of the above-described parts, or any combination of the processing performed by each of the parts, to learn a model that outputs a service usage score for each service user, using data accumulated in connection with the provision of various online services as training data.

[0106] [6. Hardware Configuration] Furthermore, the model creation unit 100 according to the embodiments and modified examples described above is implemented by a computer 1000 having a configuration such as that shown in Figure 14. Figure 14 is a hardware configuration diagram showing an example of a computer that implements the functions of the model creation unit according to the embodiments and modified examples.

[0107] Computer 1000 is connected to an output device 1010 and an input device 1020, and has a configuration in which an arithmetic unit 1030, a primary storage device 1040, a secondary storage device 1050, an output interface 1060, an input interface 1070, and a network interface 1080 are connected by a bus 1090.

[0108] The arithmetic unit 1030 operates based on programs stored in the primary storage device 1040 and the secondary storage device 1050, as well as programs read from the input device 1020, and executes various processes. The primary storage device 1040 is a memory device, such as RAM, that temporarily stores data used by the arithmetic unit 1030 for various calculations. The secondary storage device 1050 is a storage device where data used by the arithmetic unit 1030 for various calculations and various databases are registered, and is implemented using ROM (Read Only Memory), HDD, flash memory, etc.

[0109] The output IF1060 is an interface for transmitting information to be output to output devices 1010, such as monitors and printers, and is implemented using connectors of standards such as USB (Universal Serial Bus), DVI (Digital Visual Interface), and HDMI (High Definition Multimedia Interface). The input IF1070 is an interface for receiving information from various input devices 1020, such as mice, keyboards, and scanners, and is implemented using, for example, USB.

[0110] The input device 1020 may also be a device that reads information from, for example, an optical recording medium such as a CD (Compact Disc), DVD (Digital Versatile Disc), or PD (Phase Change Rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), tape media, magnetic recording media, or semiconductor memory. Furthermore, the input device 1020 may be an external storage medium such as a USB memory stick.

[0111] Network IF1080 receives data from other devices via network N and sends it to the arithmetic unit 1030, and also transmits data generated by the arithmetic unit 1030 to other devices via network N.

[0112] The arithmetic unit 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070. For example, the arithmetic unit 1030 loads a program from the input device 1020 or the secondary storage device 1050 onto the primary storage device 1040 and executes the loaded program.

[0113] For example, if the computer 1000 functions as a model creation unit 100 according to an embodiment or modification, the arithmetic unit 1030 of the computer 1000 performs the same function as the control unit 130 by executing a program (for example, a learning program) loaded on the primary storage device 1040. That is, the arithmetic unit 1030 works in cooperation with the program (for example, a learning program) loaded on the primary storage device 1040 to perform the processing by the model creation unit 100 according to the embodiment.

[0114] [7. Others] Of the processes described in the above embodiments, 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 known methods. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above documents and drawings can be changed at will unless otherwise specified.

[0115] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions.

[0116] Furthermore, the embodiments described above can be combined as appropriate, provided that the processing content is not contradictory.

[0117] Although embodiments of the present application have been described in detail above with reference to several drawings, these are illustrative examples, and the present invention can be implemented in various other forms with modifications and improvements based on the knowledge of those skilled in the art, starting with the embodiments described in the disclosure section of the invention.

[0118] Furthermore, the terms "section, module, unit" mentioned above can be replaced with "means" or "circuit," etc. For example, a control unit can be replaced with a control means or a control circuit. [Explanation of Symbols]

[0119] SYS Information Processing System N Network 10 Operator device 100 Model Creation Units 110 Communications Department 120 Storage section 121 User Behavior Database 122 User Feature Database 123 Model DB 130 Control Unit 131 Acquisition Department 132 Learning Department 200 Score Reproduction Unit 210 Communications Department 220 Storage section 221 Model DB 230 Control Unit 231 Reception Department 232 Selection Department 233 Reproduction Section 234 Learning Request Department 235 Registration Department

Claims

1. When a condition for identifying a group of users to be studied is received, the acquisition unit acquires characteristic information of users included in the user group identified by the condition from a first storage unit that stores characteristic information indicating the attributes, interests, or usage status of each user, and acquires behavioral information of users included in the user group from a second storage unit that stores behavioral information indicating the service actions of each user. A learning unit that trains a model on the relationship between the acquired feature information and the behavioral information. A learning device characterized by having the following features.

2. The acquisition unit is, When a user submits a query specifying the user's behavior to be learned, the system obtains the behavior information corresponding to the behavior specified by the query, and then obtains the feature information associated with the obtained behavior information. The learning device according to feature 1.

3. The acquisition unit is, When the system receives a user's specification of the user characteristics to be learned, it obtains the characteristic information corresponding to the specified characteristics and retrieves the behavioral information associated with the obtained characteristic information. The learning device according to feature 1.

4. A learning method performed by a computer, When conditions for identifying a group of users to be studied are received, the system acquires characteristic information of users included in the user group identified by the conditions from a first storage unit that stores characteristic information indicating the attributes, interests, or usage status of each user, and acquires behavioral information of users included in the user group from a second storage unit that stores behavioral information indicating the service actions of each user. A learning process in which the relationship between the acquired feature information and the behavioral information is learned in a model, A learning method characterized by including the following.

5. On the computer, When conditions for identifying a group of users to be studied are received, the acquisition procedure includes obtaining characteristic information of users included in the user group identified by the conditions from a first storage unit that stores characteristic information indicating the attributes, interests, or usage status of each user, and obtaining behavioral information of users included in the user group from a second storage unit that stores behavioral information indicating the service actions of each user, and A learning procedure for training a model on the relationship between the acquired feature information and the behavioral information, A learning program characterized by causing the execution of [a specific action].

Images