Information processing device and information processing method

Abstract

To provide an information processing apparatus and an information processing method for estimating the behavior of a user corresponding to a second domain from the behavior of a user corresponding to a first domain.SOLUTION: An information processing apparatus 1 includes a communication unit 11, a storage unit 12, and a control unit 13. A data acquisition unit 131 acquires first related behavioral data of a first user and second related behavioral data of a second user from a terminal of an analyzer using the information processing apparatus 1. The data acquisition unit 131 acquires the first related behavioral data, which is related to a purchasing behavior at a first site of the first user, including a first item ID of a first item corresponding to a browsing behavior at the first site and a first user ID of the first user. The data acquisition unit 131 acquires the second related behavioral data, which is related to a purchasing behavior at a second site of the second user, including a second item ID of a second item corresponding to a browsing behavior at the second site and a second user ID of the second user.SELECTED DRAWING: Figure 2

Description

[Technical field]

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

[0002] Conventionally, transfer learning is known, which applies the contents learned in a first domain to learning in a second domain different from the first domain. For example, Patent Document 1 discloses a technology for combining purchasing behavior history data in a specific service corresponding to the first domain and online behavior history data related to web page browsing corresponding to the second domain based on a common user of the first domain and the second domain. By using the combined behavior history data, it is possible to estimate the purchasing behavior of a user corresponding to the first domain from the similarity with the online behavior of the common user, even for a user corresponding to the second domain different from the common user. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6596605 Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional technology has a problem in that it is not possible to match users between data in a first domain and data in a second domain, and if there is no common user, it is not possible to infer the behavior of a user corresponding to the second domain from the behavior of the user corresponding to the first domain.

[0005] Therefore, the present invention has been made in consideration of these points, and aims to infer the behavior of a user corresponding to a second domain from the behavior of a user corresponding to a first domain, without matching users in the data of the first domain and the data of the second domain. [Means for solving the problem]

[0006] An information processing device according to a first aspect of the present invention includes a data acquisition unit that acquires first related behavior data including first item identification information for identifying a first item that is an item corresponding to a first related behavior related to a first behavior of a first user who is a user of a first domain and first user identification information for identifying the first user, and second related behavior data including second item identification information for identifying a second item that is an item corresponding to a second related behavior related to a second behavior of a second user who is a user of a second domain different from the first domain and second user identification information for identifying the second user; a first construction unit that constructs a first context estimation model using the first related behavior data, the first construction unit having a first reception unit that receives input of a first user vector corresponding to the first user identification information and a first item vector corresponding to the first item identification information, a first conversion unit that converts the input vector into a first feature vector of a predetermined number of dimensions indicating features of the vector, and a first output unit that outputs first context information that is information related to the first related behavior of the first user based on the converted first feature vector; a second construction unit that constructs a second context estimation model using the second related behavior data, the second context estimation model having a second user vector corresponding to the second user identification information and a second item vector corresponding to the second item identification information and expressed in the same feature space as the first item vector, a second conversion unit that converts the input vector into a second feature vector of the predetermined number of dimensions indicating features of the vector, and a second output unit that outputs second context information, which is information regarding the second related behavior of the second user and has the same expression format as the first context information, based on the converted second feature vector; a third construction unit that constructs a first behavior estimation model using the first related behavior data and data indicating an occurrence result of the first behavior with respect to the first related behavior, the third output unit that outputs first behavior prediction information indicating an occurrence probability of the first behavior of the first user, based on the first reception unit, the first conversion unit, and the converted first feature vector;and a fourth construction unit that constructs a second behavior estimation model having the third output unit that outputs second behavior prediction information indicating an occurrence probability of the second behavior of the second user based on the second feature vector.

[0007] The first related behavior data may include time information indicating a time when the first related behavior occurred, and the second related behavior data may include time information indicating a time when the second related behavior occurred, the first construction unit may construct the first context estimation model having the first output unit that outputs, as the first context information, time information indicating the time when the first related behavior of the first user occurred, and the second construction unit may construct the second context estimation model having the second output unit that outputs, as the second context information having the same expression format as the first context information, time information indicating the time when the second related behavior of the second user occurred.

[0008] The first related behavior data may include location information indicating a location where the first related behavior occurred, and the second related behavior data may include location information indicating a location where the second related behavior occurred, the first construction unit may construct the first context estimation model having the first output unit that outputs, as the first context information, location information indicating a location where the first related behavior of the first user occurred, and the second construction unit may construct the second context estimation model having the second output unit that outputs, as the second context information having the same expression format as the first context information, location information indicating a location where the second related behavior of the second user occurred.

[0009] The first item identification information and the second item identification information may each be associated with text information describing the item, and the first construction unit may construct the first context estimation model having the first reception unit that receives input of the first item vector based on the text information associated with the first item identification information, and the first output unit that outputs, as the first context information, the first item vector corresponding to another first item related to the first item indicated by the first item vector, and the second construction unit may construct the second context estimation model having the second reception unit that receives input of the second item vector based on the text information associated with the second item identification information, and the second output unit that outputs, as the second context information, the second item vector corresponding to another second item related to the second item indicated by the second item vector.

[0010] The data acquisition unit may acquire the first related behavior data including the first item identification information corresponding to the first related behavior related to the first behavior and the first user identification information, and may also acquire the second related behavior data including the second item identification information corresponding to a second related behavior related to the second behavior having a predetermined relationship with the first behavior and the second user identification information.

[0011] The information processing device further includes a judgment unit that accepts input of at least one of the first feature vector converted by the first conversion unit from the first user vector and the first item vector, and the second feature vector converted by the second conversion unit from the second user vector and the second item vector, and outputs judgment information indicating whether the input feature vector is the first feature vector or the second feature vector, wherein the judgment unit learns to increase the accuracy rate of the output result of the judgment information, and at least one of the first construction unit and the second construction unit learns to increase the accuracy rate of the judgment result of the judgment unit when the judgment unit judges the feature vector generated by the judgment unit.

[0012] The first receiving unit may accept input of a vector obtained by averaging a plurality of the first user vectors and a first item vector corresponding to the first item identification information corresponding to the plurality of first user vectors, and the second receiving unit may accept input of a vector obtained by averaging a plurality of the second user vectors and a second item vector corresponding to the second item identification information corresponding to the plurality of second user vectors.

[0013] An information processing method according to a second aspect of the present invention includes a step of acquiring, executed by a computer, first related behavior data including first item identification information for identifying a first item that is an item corresponding to a first related behavior related to a first behavior of a first user who is a user of a first domain and first user identification information for identifying the first user, and second related behavior data including second item identification information for identifying a second item that is an item corresponding to a second related behavior related to a second behavior of a second user who is a user of a second domain different from the first domain and second user identification information for identifying the second user, and generating a first context estimation model using the first related behavior data, the first context estimation model having a first reception unit that receives input of a first user vector corresponding to the first user identification information and a first item vector corresponding to the first item identification information, a first conversion unit that converts the input vector into a first feature vector of a predetermined number of dimensions indicating features of the vector, and a first output unit that outputs first context information that is information regarding the first related behavior of the first user based on the converted first feature vector. a step of constructing a second context estimation model using the second related behavior data, the second context estimation model having a second reception unit that receives an input of a second user vector corresponding to the second user identification information and a second item vector that corresponds to the second item identification information and is expressed in the same feature space as the first item vector, a second conversion unit that converts the input vector into a second feature vector of the predetermined number of dimensions that indicates features of the vector, and a second output unit that outputs second context information, which is information about the second related behavior of the second user and has the same expression format as the first context information, based on the converted second feature vector; a step of constructing a first behavior estimation model using the first related behavior data and data indicating an occurrence result of the first behavior with respect to the first related behavior, the first reception unit, the first conversion unit, and a third output unit that outputs first behavior prediction information indicating an occurrence probability of the first behavior of the first user, based on the converted first feature vector;and constructing a second behavior estimation model including the third output unit constructed in the step of constructing the first behavior estimation model, the third output unit outputting second behavior prediction information indicating an occurrence probability of the second behavior of the second user based on the second feature vector. Effect of the Invention

[0014] According to the present invention, it is possible to estimate the behavior of a user corresponding to a second domain from the behavior of a user corresponding to a first domain without matching users in the data of the first domain and the data of the second domain. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a diagram illustrating an overview of an information processing device according to a first embodiment. [Diagram 2] FIG. 1 is a diagram illustrating a configuration of an information processing device according to a first embodiment. [Diagram 3] FIG. 4 is a diagram showing first related behavior data and second related behavior data. [Figure 4] FIG. 13 is a diagram illustrating a first context estimation model. [Diagram 5] FIG. 13 is a diagram illustrating a second context estimation model. [Figure 6] FIG. 11 is a diagram showing a first behavior inference model. [Figure 7] FIG. 13 is a diagram showing a second behavior estimation model. [Figure 8] 4 is a flowchart showing an example of a process flow in the information processing device according to the first embodiment. [Figure 9] FIG. 13 is a diagram illustrating a configuration of an information processing device according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] <First embodiment> [Outline of information processing device] 1 is a diagram illustrating an overview of an information processing device 1 according to a first embodiment. The information processing device 1 is a computer for estimating the occurrence probability of a predetermined behavior of a user based on related behaviors related to the user behavior.

[0017] In the following description, a user of a first domain is referred to as a first user, a specific action of the first user is referred to as a first action, and a related action related to the first action is referred to as a first related action. Also, a user of a second domain, which is a domain different from the first domain, is referred to as a second user, a specific action of the second user is referred to as a second action, and a related action related to the second action is referred to as a second related action.

[0018] The first user is, for example, a user who uses a first site, which is a site of a first domain. The first behavior is a behavior detectable by the information processing device 1, for example, a purchasing behavior of a user in a real store or a virtual store. In this embodiment, the first behavior is the purchasing behavior of the first user in the first site. The first related behavior is, for example, a browsing behavior of the first site by the first user.

[0019] The second user is, for example, a user who uses a site of the second domain. The second behavior is behavior that has a predetermined relationship with the first behavior. The predetermined relationship means, for example, that the type of the first behavior and the type of the second behavior are the same. The second behavior is, for example, purchasing behavior of the second user at a second site, which is a site of the second domain that is a different site from the site corresponding to the first domain. The second related behavior is, for example, browsing behavior of the second user at the second site.

[0020] First, the information processing device 1 acquires first related behavior data corresponding to a first related behavior of a first user of a first domain, data indicating the occurrence result of a first behavior for the first related behavior, and second related behavior data corresponding to a second related behavior of a second user of a second domain. The first related behavior data includes a first user ID for identifying the first user, and a first item ID as first item identification information for identifying an item corresponding to the first related behavior. The second related behavior data includes a second user ID for identifying the second user, and a second item ID as second item identification information for identifying an item corresponding to the second related behavior.

[0021] The information processing device 1 uses the first user ID and the first item ID included in the acquired first related behavior data to construct a first context estimation model, which is a program that outputs first context information, which is information regarding the first related behavior of the first user, in response to input of a first user vector corresponding to the first user ID and a first item vector corresponding to the first item ID.

[0022] The information processing device 1 uses the second user ID and the second item ID included in the acquired second related behavior data to construct a second context estimation model, which is a program that outputs second context information that is information about the second related behavior of the second user and has the same expression format as the first context information, in response to input of a second user vector corresponding to the second user ID and a second item vector corresponding to the second item ID and expressed in the same feature space as the first item vector. This allows the first context estimation model and the second context estimation model to output context information in the same expression format.

[0023] The information processing device 1 uses the first related behavior data and data indicating the occurrence result of the first behavior for the first related behavior to construct a first behavior estimation model that outputs first behavior prediction information indicating the occurrence probability of the first behavior of the first user in response to input of a first user vector corresponding to a first user ID and a first item vector corresponding to a first item ID. The information processing device 1 constructs the first behavior estimation model, which is a program having a part that performs processing common to the first context estimation model, by reusing an input part and an intermediate processing part of the first behavior estimation model.

[0024] The first context estimation model and the first behavior estimation model have common input parts and intermediate processing parts, and the first context estimation model and the second context estimation model output context information in the same expression format, so that the output part of the second context estimation model can be replaced with the output part of the first behavior estimation model.

[0025] Therefore, the information processing device 1 replaces the output unit of the second context estimation model with the output unit of the first behavior estimation model to construct a second behavior estimation model, which is a program that outputs second behavior prediction information indicating the occurrence probability of the second behavior of the second user in response to input of a second user vector corresponding to a second user ID and a second item vector corresponding to a second item ID. In this way, the information processing device 1 can estimate the behavior of the user corresponding to the second domain from the behavior of the user corresponding to the first domain without associating users in the data of the first domain and the data of the second domain.

[0026] Next, a description will be given of the configuration of the information processing device 1. In describing the configuration of the information processing device 1, the description will be given on the assumption that the first behavior is the user's purchasing behavior at a first site which is a site of a first domain, the first related behavior is the user's browsing behavior at the first site, the second behavior is the user's purchasing behavior at a second site which is a site of a second domain, and the second related behavior is the user's browsing behavior at the second site.

[0027] [Configuration example of information processing device 1] Next, the configuration of the information processing device 1 will be described. Fig. 2 is a diagram showing the configuration of the information processing device 1 according to the first embodiment. The information processing device 1 has a communication unit 11, a storage unit 12, and a control unit 13. The control unit 13 has a data acquisition unit 131, a teacher data generation unit 132, a first construction unit 133, a second construction unit 134, a third construction unit 135, a fourth construction unit 136, and a prediction unit 137.

[0028] The communication unit 11 is a communication interface for transmitting and receiving data to and from an external device via a network such as the Internet. The storage unit 12 is, for example, a Read Only Memory (ROM) and a Random Access Memory (RAM), etc. The storage unit 12 stores various programs for causing the information processing device 1 to function. For example, the storage unit 12 stores programs for causing the control unit 13 of the information processing device 1 to function as a data acquisition unit 131, a teacher data generation unit 132, a first construction unit 133, a second construction unit 134, a third construction unit 135, a fourth construction unit 136, and a prediction unit 137.

[0029] The control unit 13 is, for example, a CPU (Central Processing Unit). The control unit 13 executes various programs stored in the storage unit 12 to control functions related to the information processing device 1. The control unit 13 executes the programs stored in the storage unit 12 to function as a data acquisition unit 131, a teacher data generation unit 132, a first construction unit 133, a second construction unit 134, a third construction unit 135, a fourth construction unit 136, and a prediction unit 137.

[0030] The data acquisition unit 131 acquires, for example, from a terminal (not shown) of an analyst using the information processing device 1, first related behavior data corresponding to a first related behavior related to a first behavior of a first user and second related behavior data corresponding to a second related behavior related to a second behavior of a second user. The data acquisition unit 131 acquires first related behavior data including a first item ID of a first item corresponding to a browsing behavior (first related behavior) on a first site related to a purchasing behavior (first behavior) on a first site of the first user, and a first user ID of the first user. In addition, the data acquisition unit 131 acquires second related behavior data including a second item ID of a second item corresponding to a browsing behavior (second related behavior) on a second site related to a purchasing behavior (second behavior) on a second site of the second user, and a second user ID of the second user.

[0031] Specifically, the data acquisition unit 131 acquires first related behavior data in which a first user ID, a first item ID indicating a first item viewed by the first user on the first site, time information indicating the time when the first item was viewed, and location information indicating the location where the first item was viewed are associated with each other. The data acquisition unit 131 also acquires first related behavior data in which a second user ID, a second item ID indicating a second item viewed by the second user on the second site, time information indicating the time when the second item was viewed, and location information indicating the location where the second item was viewed are associated with each other.

[0032] FIG. 3 is a diagram showing the first related behavior data and the second related behavior data. FIG. 3(a) shows the first related behavior data, and FIG. 3(b) shows the second related behavior data. As shown in FIG. 3(a), it can be confirmed that the first user ID, the first item ID, time information indicating the time when the first item was viewed, and location information indicating the place where the first item was viewed are associated with each other in the first related behavior data. Similarly, as shown in FIG. 3(b), it can be confirmed that the second user ID, the second item ID, time information indicating the time when the second item was viewed, and location information indicating the place where the second item was viewed are associated with each other in the second related behavior data. Note that, as shown in FIG. 3(a) and FIG. 3(b), the first user ID and the second user ID are IDs of different systems, and it is assumed that the first user ID and the second user ID cannot be associated with each other.

[0033] Furthermore, the data acquiring unit 131 acquires first behavior data indicating the occurrence result of a purchasing behavior (first behavior) of a first item in response to a browsing behavior (first related behavior) of the first item on the first site. For example, the data acquiring unit 131 acquires, as the first behavior data, purchase history information in which the first user ID of the first user, the first item ID of the first item purchased by the first user on the first site, and time information indicating the time when the first item was purchased are associated with each other.

[0034] The teacher data generating unit 132 generates first teacher data used to construct the first context estimation model based on the first related behavior data. The first teacher data is a combination of an input vector to be input to the first context estimation model and an output vector to be output from the first context estimation model in response to the input of the input vector. The teacher data generating unit 132 generates a plurality of pieces of first teacher data.

[0035] The teacher data generation unit 132 generates, as input vectors used in constructing a first context estimation model, a first user vector corresponding to a first user and a first item content vector as a first item vector corresponding to a first item.

[0036] For example, the teacher data generating unit 132 generates a 1hot vector corresponding to a first user ID included in the first related behavior data as a first user vector. In addition, the teacher data generating unit 132 generates a first item content vector indicating a feature of a first item corresponding to a first item ID associated with the first user ID in the first related behavior data.

[0037] For example, an item ID of a first item and a description as text information explaining the first item in a predetermined language are stored in association with each other in storage unit 12. Teacher data generation unit 132 generates a first item content vector based on vocabulary included in the description associated with the item ID of the first item.

[0038] The first item content vector is, for example, a vector indicating whether each of n vocabulary words is included in the description of the first item, and is expressed as (V 1 , V 2 , …, V n ) is shown. V i (where 1≦i≦n) is 1 if vocabulary i is included in the description of the first item, and is 0 if vocabulary i is not included in the description.

[0039] The teacher data generation unit 132 sets, as an input vector, a set of a first user vector corresponding to a first user ID included in the first related behavior data and a first item content vector generated for an item ID associated with the first user ID.

[0040] The teacher data generation unit 132 generates a first time vector, a first location vector, and a first item context vector in correspondence with the input vector, as output vectors used in constructing a first context estimation model.

[0041] The teacher data generation unit 132 generates a first time vector and a first location vector based on time information and location information associated with the first user ID and first item ID corresponding to the generated input vector in the first related behavior data.

[0042] The first time vector is a vector corresponding to the time indicated by the time information associated with the first item, and is generated based on the time of the time information associated with the item ID, for example. For example, the first time vector has 24 elements corresponding to each of the hours from 0:00 to 23:00. If the time associated with the first item is 19:00, the value of the element corresponding to 19:00 is 1, and the values ​​of the other elements are 0.

[0043] The first location vector is a vector corresponding to a location indicated by the location information associated with the first item. The first location vector is generated, for example, based on which prefecture the location associated with the item ID belongs to. For example, the first location vector has 47 elements corresponding to the number of prefectures. If the location associated with the first item is Tokyo, the value of the element corresponding to Tokyo is 1, and the values ​​of the other elements are 0.

[0044] The teacher data generation unit 132 refers to the first related behavior data and identifies multiple first item IDs that appear before and after the first item ID included in the input vector. The teacher data generation unit 132 generates a first item context vector that includes a first item content vector corresponding to the first item ID included in the input vector and a first item content vector for each of the identified multiple first item IDs.

[0045] For example, the first item context vector includes an item content vector corresponding to the first item ID included in the input vector, two item content vectors that were viewed immediately before the viewing time of the first item indicated by the first item ID, and two item content vectors that were viewed immediately after the viewing time of the first item.

[0046] The teacher data generation unit 132 generates multiple combinations of input vectors and output vectors as teacher data to be used in constructing the second context estimation model, based on the second related behavior data, in the same way as generating teacher data to be used in constructing the first context estimation model.

[0047] The teacher data generation unit 132 generates, as input vectors used in constructing a second context estimation model, a second user vector corresponding to the second user and a second item content vector as a second item vector corresponding to the second item.

[0048] The teacher data generation unit 132 generates, as the second user vector, a one-hot vector corresponding to the second user ID included in the second associated behavior data, in the same manner as the first user vector. In addition, the storage unit 12 stores the item ID of the second item and a description as text information that describes the second item using the same language as the description of the first item, in association with each other. The teacher data generation unit 132 generates the second item content vector based on vocabulary included in the description associated with the item ID of the second item. As with the first item content vector, the second item content vector is calculated as follows: (V 1 , V 2 , …, V n ) As a result, the first item content vector and the second item content vector are represented in a common feature space.

[0049] Furthermore, the teacher data generation unit 132 generates a second time vector, a second location vector, and a second item context vector as output vectors used to construct a second context estimation model in response to the generated input vector. The second time vector, the second location vector, and the second item context vector are generated in the same manner as the first time vector, the first location vector, and the first item context vector, respectively. As a result, the second time vector, the second location vector, and the second item context vector are expressed in a common feature space as the first time vector, the first location vector, and the first item context vector, respectively.

[0050] The first construction unit 133 uses the first related behavior data to construct a first context estimation model that outputs first context information, which is information regarding the browsing behavior, when an input vector indicating the browsing behavior (first related behavior) of the first user is input.

[0051] The input vector indicating the browsing behavior of the first user is a first user vector corresponding to a first user ID indicating the first user, and a first item content vector corresponding to a first item ID indicating a first item browsed by the first user. The first context information is spatio-temporal context information and item context information. The spatio-temporal context information is a first time vector as time information indicating the occurrence time of the browsing behavior of the first user, and a first location vector as location information indicating the occurrence location of the browsing behavior of the first user. The item context information is a vector including a first item content vector of a first item corresponding to the browsing behavior of the first user, and a first item content vector corresponding to another first item related to the first item. The other first items related to the first item are, for example, first items corresponding to browsing behavior before and after the browsing behavior of the first user.

[0052] Fig. 4 is a diagram showing a first context estimation model. The first context estimation model is, for example, a neural network model, and has a first reception unit, a first conversion unit, and a first output unit, as shown in Fig. 4. When a computer executes the first context estimation model, the computer functions as the first reception unit, the first conversion unit, and the first output unit.

[0053] The first reception unit receives an input of a first user vector corresponding to a first user ID and a first item content vector corresponding to a first item ID. The first conversion unit converts the input vectors, that is, the first user vector and the first item content vector, into a first embedding vector, which is a first feature vector with a predetermined number of dimensions indicating features of these vectors. The first output unit outputs an output vector as first context information based on the first embedding vector converted by the first conversion unit.

[0054] The first construction unit 133 uses the input vector generated by the teacher data generation unit 132 and the output vector corresponding to the input vector as teacher data. The first construction unit 133 constructs the first context estimation model by learning the first context estimation model such that, when an input vector included in the teacher data is input to the first context estimation model, an output vector corresponding to the input vector is output from the first context estimation model. The first construction unit 133 stores the constructed first context estimation model in the storage unit 12.

[0055] The second construction unit 134 constructs a second context estimation model that outputs second context information related to a browsing behavior of a second user when an input vector indicating the browsing behavior (second related behavior) of the second user is input, using the second related behavior data. The second construction unit 134 outputs the second context information having the same expression format as the first context information.

[0056] The input vector indicating the second related behavior is a second user vector corresponding to a second user ID indicating a second user, and a second item content vector corresponding to a second item ID indicating a second item viewed by the second user. The second item content vector may be an item content vector corresponding to a first item ID indicating a first item as an item that the second user may view in the future. The second context information is spatio-temporal context information and item context information. The spatio-temporal context information is a second time vector as time information indicating the occurrence time of the browsing behavior of the second user, and a second location vector as location information indicating the occurrence location of the browsing behavior of the second user. The item context information is a vector including a second item content vector of the first item or the second item corresponding to the browsing behavior of the second user, and a second item content vector corresponding to another second item related to the second item. The other second items related to the second item are, for example, second items corresponding to browsing behavior before and after the browsing behavior of the second user.

[0057] Fig. 5 is a diagram showing a second context estimation model. As shown in Fig. 5, the second context estimation model is, for example, a neural network model, and has a similar configuration to the first context estimation model. The second context estimation model has a second reception unit, a second conversion unit, and a second output unit. When a computer executes the second context estimation model, the computer functions as the second reception unit, the second conversion unit, and the second output unit.

[0058] The second reception unit receives an input of a second user vector corresponding to a second user ID, and a second item content vector corresponding to a second item ID and expressed in the same feature space as the first item content vector.

[0059] The second conversion unit converts the input vectors, i.e., the second user vector and the second item content vector, into second embedding vectors, which are second feature vectors with a predetermined number of dimensions indicating the features of these vectors. The second conversion unit converts the second user vector and the second item content vector into second embedding vectors having the same number of dimensions as the first embedding vector.

[0060] The second output unit outputs, based on the second embedding vector converted by the second conversion unit, second context information which is information on a second associated behavior of a second user and has the same expression format as the first context information.

[0061] The second construction unit 134 constructs the second context estimation model by learning the second context estimation model so that, when an input vector corresponding to the second domain generated from the second related behavior data by the teacher data generation unit 132 is input to the second context estimation model, an output vector corresponding to the input vector is output from the second context estimation model. The second construction unit 134 stores the constructed second context estimation model in the storage unit 12.

[0062] The third construction unit 135 uses the first related behavior data and data indicating the occurrence result of the first behavior for the first related behavior included in the first related behavior data to construct a first behavior estimation model that outputs first behavior prediction information indicating the occurrence probability of the first user's purchasing behavior (first behavior) when an input vector indicating the first user's browsing behavior (first related behavior) is input.

[0063] The input vector indicating the first related behavior is the same as the input vector input to the first context estimation model, that is, a first user vector corresponding to a first user ID indicating a first user, and a first item content vector corresponding to a first item ID indicating a first item viewed by the first user.

[0064] Fig. 6 is a diagram showing a first behavior estimation model. The first behavior estimation model is, for example, a neural network model, and has a first reception unit, a first conversion unit, and a third output unit, as shown in Fig. 6. When a computer executes the first behavior estimation model, the computer functions as the first reception unit, the first conversion unit, and the third output unit.

[0065] The first reception unit and the first conversion unit are the same as the first reception unit and the first conversion unit included in the first behavior estimation model, and therefore a description thereof will be omitted. The third output unit outputs first behavior prediction information indicating the occurrence probability of the purchasing behavior (first behavior) of the first user based on the first embedding vector converted by the first conversion unit. The first behavior prediction information is, for example, rating information indicating an evaluation value of the first user purchasing the first item corresponding to the input vector. The rating information may be an explicit evaluation value such as a five-level rating value associated with the browsing behavior or purchasing behavior of the first user for the first item, or an implicit evaluation value indicating whether or not the browsing behavior or purchasing behavior has occurred.

[0066] The third construction unit 135 uses, as teacher data, the input vector generated by the teacher data generation unit 132 and purchase result information indicating the purchase result of the first item corresponding to the first item content vector included in the input vector. The purchase result information is generated based on the purchase history information acquired by the data acquisition unit 131, and is binary information taking either 0 or 1. For example, if the first user purchases the first item, the value of the purchase result information is 1, and if the first user does not purchase the first item, the value of the purchase result information is 0. Note that the purchase result information may be information indicating the probability that the first user purchases the first item, and may be generated by, for example, an analysis by an analyst or the like.

[0067] The third construction unit 135 constructs the first behavior estimation model by learning the first behavior estimation model using the teacher data so that when an input vector included in the teacher data is input to the first behavior estimation model, rating information indicating the probability of purchasing a first item corresponding to the input vector is output from the first behavior estimation model. The third construction unit 135 stores the constructed first behavior estimation model in the storage unit 12.

[0068] The fourth construction unit 136 uses the second context estimation model constructed by the second construction unit 134 and the first behavior estimation model constructed by the third construction unit 135 to construct a second behavior estimation model that outputs second behavior prediction information indicating the occurrence probability of the purchasing behavior (second behavior) of the second user when an input vector indicating the browsing behavior (second related behavior) of the second user is input.

[0069] Specifically, the fourth construction unit 136 constructs a second behavior estimation model having a second reception unit and a second conversion unit constructed by the second construction unit 134, and a third output unit constructed by the third construction unit, which outputs second behavior prediction information indicating the occurrence probability of the second behavior of the second user based on the second embedding vector generated by the second conversion unit.

[0070] FIG. 7 is a diagram showing the second behavior estimation model. The second behavior estimation model is, for example, a neural network model, and as shown in FIG. 7, has the second reception unit and the second conversion unit of the second context estimation model. In addition, the second behavior estimation model has the third output unit of the first behavior estimation model. When the computer executes the second behavior estimation model, the computer functions as the second reception unit, the second conversion unit, and the third output unit. The fourth construction unit 136 stores the constructed second behavior estimation model in the storage unit 12.

[0071] The prediction unit 137 predicts the occurrence probability of the second behavior of the second user. For example, the prediction unit 137 acquires, from the terminal used by the analyst, the second user ID of the second user of the second domain and the second item ID of the second item corresponding to the browsing behavior (second related behavior) of the second user on the second site. The prediction unit 137 acquires the second item ID of the second item, but is not limited to this, and may acquire the first item ID of the first item.

[0072] The prediction unit 137 generates a 1hot vector corresponding to the acquired second user ID as a second user vector. The prediction unit 137 generates a second item content vector corresponding to the acquired first item ID or second item ID. The prediction unit 137 executes the second behavior estimation model stored in the storage unit 12, inputs the generated second user vector and the second item content vector to the second behavior estimation model, and acquires second behavior prediction information from the second behavior estimation model. The prediction unit 137 outputs the acquired second behavior prediction information to the terminal of the analyst. The prediction unit 137 may input a plurality of candidate second item content vectors and output the first item ID or the second item ID having a high occurrence probability among them.

[0073] [Operation flow] FIG. 8 is a flowchart showing an example of a process flow in the information processing device 1 according to the present embodiment. First, the data acquisition unit 131 acquires the first associated behavior data, the second associated behavior data, and the first behavior data (S1). Next, the teacher data generation unit 132 generates teacher data to be used for constructing the first context estimation model, the second context estimation model, and the first behavior estimation model based on the first related behavior data, the second related behavior data, and the first behavior data acquired in S1 (S2).

[0074] Next, the first construction unit 133 constructs a first context estimation model using the teacher data generated by the teacher data generation unit 132 (S3). Next, the second construction unit 134 constructs a second context estimation model using the teacher data generated by the teacher data generation unit 132 (S4). Here, the process related to S4 is executed after the process related to S3 is executed, but this is not limited to the above. The process related to S3 may be executed after the process related to S4 is executed, or may be executed asynchronously in parallel. Next, the third construction unit 135 constructs a first behavior estimation model using the teacher data generated by the teacher data generation unit 132 (S5).

[0075] Next, the fourth construction unit 136 constructs a second behavior estimation model by combining the second reception unit and the second conversion unit of the second context model and the third output unit of the first behavior estimation model (S6).

[0076] Next, the prediction unit 137 acquires a second user ID and a second item ID from the terminal of the analyst (S7). Note that, as described above, the prediction unit 137 may acquire a first item ID as a future action instead of acquiring the second item ID.

[0077] Next, the prediction unit 137 generates a second user vector corresponding to the acquired second user ID, and generates a second item content vector corresponding to the acquired second item ID. Note that, when the prediction unit 137 acquires a first item ID instead of the second item ID, it generates a first item content vector corresponding to the first item ID. Then, the prediction unit 137 inputs the generated second user vector and the second item content vector to a second behavior estimation model, and acquires second behavior prediction information from the second behavior estimation model (S8). Note that, when the prediction unit 137 acquires a first item ID instead of the second item ID, it inputs the generated second user vector and the first item content vector to the second behavior estimation model, and acquires second behavior prediction information from the second behavior estimation model. Next, the prediction unit 137 outputs the second behavior prediction information acquired in S8 to the analyst's terminal (S9).

[0078] [Variations] In the above embodiment, the first reception unit of the first context estimation model receives a first user vector corresponding to one first user, and the second reception unit of the second context estimation model receives a first user vector corresponding to one second user, but this is not limited to the above. The first reception unit may receive an input of a vector obtained by averaging a plurality of first user vectors corresponding to a plurality of first users, and a first item content vector corresponding to a first item ID corresponding to the plurality of first user vectors. Similarly, the second reception unit may receive an input of a vector obtained by averaging a plurality of second user vectors corresponding to a plurality of second users, and a second item content vector corresponding to a second item ID corresponding to the plurality of second user vectors.

[0079] In this case, the first item ID corresponding to the multiple first user vectors is, for example, a first item ID corresponding to a first item that is viewed relatively frequently by multiple first users on a site of the first domain. Also, the second item ID corresponding to the multiple second user vectors is, for example, a first item ID corresponding to a first item that is viewed relatively frequently by multiple second users on a site of the second domain. In this way, the information processing device 1 can acquire context information corresponding to multiple users of each domain, taking into consideration the privacy of the users.

[0080] [Advantages of the first embodiment] As described above, the information processing device 1 according to the present embodiment uses the first related behavior data and data indicating the occurrence result of the first behavior for the first related behavior to construct a first behavior prediction model having a first reception unit that receives input of a first user vector corresponding to a first user ID and a first item content vector corresponding to a first item ID, a first conversion unit that converts the input vector into a first embedding vector with a predetermined number of dimensions indicating the characteristics of the vector, and a third output unit that outputs first behavior prediction information indicating the occurrence probability of the first behavior of the first user based on the converted first embedding vector. Also, the information processing device 1 uses the second related behavior data to construct a second context estimation model having a second reception unit that receives input of a second user vector corresponding to a second user ID and a second item content vector corresponding to the second item ID and expressed in the same feature space as the first item content vector, a second conversion unit that converts the input vector into a second embedding vector with a predetermined number of dimensions indicating the characteristics of the vector, and a second output unit that outputs second context information, which is information related to the second related behavior of the second user and has the same expression format as the first context information, based on the converted second embedding vector. Then, the information processing device 1 constructs a second behavior estimation model having a second reception unit, a second conversion unit, and a third output unit constructed by the third construction unit, the third output unit outputting second behavior prediction information indicating the occurrence probability of the second behavior of the second user based on the second embedding vector. In this way, it is possible to infer the behavior of the user corresponding to the second domain from the behavior of the user corresponding to the first domain without associating users in the data of the first domain and the data of the second domain.

[0081] <Second embodiment> Next, a second embodiment will be described. The information processing device 1 according to the second embodiment is different from the first embodiment in that the first context estimation model and the second context estimation model are adjusted so that the first embedding vector and the second embedding vector are generated as if the embedding vectors generated by the first context estimation model and the second context estimation model were the embedding vectors output from the same context estimation model. The information processing device 1 according to the second embodiment will be described below. Note that the description of the same parts as those in the first embodiment will be omitted as appropriate.

[0082] 9 is a diagram showing a configuration of the information processing device 1 according to the second embodiment. As shown in FIG.

[0083] The determination unit 138 receives at least one of an input of a first embedding vector converted by a first conversion unit of a first context estimation model from a first user vector and a first item content vector, and a second feature vector converted by a second conversion unit of a second context estimation model from a second user vector and a second item content vector, and outputs determination information indicating whether the input embedding vector is the first embedding vector or the second embedding vector. For example, the determination unit 138 determines whether the input embedding vector is the first embedding vector or the second embedding vector based on the feature amount of the input embedding vector, and outputs determination information indicating the determination result.

[0084] In addition, the determination unit 138 learns so as to increase the accuracy rate of the output result of the determination information. For example, the determination unit 138 accepts an input of an embedding vector and obtains correct answer information indicating whether the embedding vector is the first embedding vector or the second embedding vector. The determination unit 138 learns so as to increase the probability that its own determination result will be the correct answer indicated by the correct answer information.

[0085] The first construction unit 133 and the second construction unit 134 learn so that the accuracy rate of the judgment result of the judgment unit 138 when the judgment unit 138 judges the embedding vector generated by the first construction unit 133 is close to 50%. For example, the first construction unit 133 inputs the first embedding vector generated by the first conversion unit to the judgment unit 138, performs a process of acquiring the judgment result from the judgment unit 138 multiple times, and learns so that the accuracy rate of the judgment result output from the judgment unit 138 is close to 50%. Similarly to the first construction unit 133, the second construction unit 134 also inputs the first embedding vector generated by the first conversion unit to the judgment unit 138, performs a process of acquiring the judgment result from the judgment unit 138 multiple times, and learns so that the accuracy rate of the judgment result output from the judgment unit 138 is close to 50%.

[0086] In the present embodiment, the first construction unit 133 and the second construction unit 134 learn so that the accuracy rate of the determination result of the determination unit 138 when the determination unit 138 determines an embedding vector generated by the first construction unit 133 and the second construction unit 134 is close to 50%, but this is not limited to the above. Only one of the first construction unit 133 and the second construction unit 134 may learn so that the accuracy rate of the determination result of the determination unit 138 when the determination unit 138 determines an embedding vector generated by the first construction unit 133 and the second construction unit 134 is close to 50%.

[0087] [Advantages of the second embodiment] As described above, in the information processing device 1 according to the present embodiment, the determination unit 138 learns to increase the accuracy rate of the output result of the determination information, and at least one of the first construction unit 133 and the second construction unit 134 learns to increase the accuracy rate of the determination result when the determination unit 138 determines a feature vector generated by the determination unit 133 or the second construction unit 134. In this manner, the first context estimation model and the second context estimation model are adjusted as if the embedding vectors generated by the first context estimation model and the second context estimation model were the embedding vectors output from the same context estimation model. This reduces the difference between the domains of the embedding vectors, which are intermediate representations of the first context estimation model and the second context estimation model, and improves the estimation accuracy of the second behavior estimation model generated by combining the second reception unit and the second conversion unit of the second context estimation model with the third output unit of the first behavior estimation model.

[0088] Furthermore, this invention will make it possible to contribute to Goal 9 of the United Nations' Sustainable Development Goals (SDGs), which is to "build resilient infrastructure, promote industry, innovation and infrastructure."

[0089] Also, for example, all or part of the device can be configured in any unit by distributing or integrating functionally or physically. Also, new embodiments resulting from any combination of multiple embodiments are included in the embodiments of the present invention. The effect of the new embodiment resulting from the combination also has the effect of the original embodiment. [Explanation of symbols]

[0090] 1. Information processing device 11 Communications Department 12 Storage section 13 Control section 131 Data Acquisition Section 132 Teacher Data Generation Unit 133 First Construction Division 134 Second Construction Division 135 3rd Construction Division 136 4th Construction Division 137 Prediction Department 138 Judgment section

Claims

1. The data acquisition unit acquires first related behavior data including first item identification information for identifying a first item, which is an item corresponding to a first related behavior related to a first behavior of a first user who is a user of a first domain, and first user identification information for identifying the first user, and second related behavior data including second item identification information for identifying a second item, which is an item corresponding to a second related behavior related to a second behavior of a second user who is a user of a second domain different from the first domain, and second user identification information for identifying the second user, the first action and the second action are actions of the same type, and the first related action and the second related action are actions of the same type; a first construction unit that constructs a first context estimation model using the first related behavior data, the first construction unit having a first reception unit that receives an input of a first user vector corresponding to the first user identification information and a first item vector corresponding to the first item identification information and indicating a feature of the first item, a first conversion unit that converts the input vector into a first feature vector of a predetermined number of dimensions that indicates a feature of the vector, and a first output unit that outputs first context information, which is information regarding the first related behavior of the first user, based on the converted first feature vector; a second receiving unit that receives an input of a second user vector corresponding to the second user identification information and a second item vector corresponding to the second item identification information and expressed in the same feature space as the first item vector and indicating features of the second item, using the second related behavior data, a second conversion unit that converts the input vector into a second feature vector of the predetermined number of dimensions that indicates features of the vector, and a second construction unit that constructs a second context estimation model based on the converted second feature vector, the second context estimation model having a second output unit that outputs second context information that is information related to the second related behavior of the second user and has the same expression format as the first context information; a third construction unit that constructs a first behavior estimation model using the first related behavior data and data indicating an occurrence result of the first behavior with respect to the first related behavior, based on the first reception unit, the first conversion unit, and the converted first feature vector, the third construction unit having a third output unit that outputs first behavior prediction information indicating an occurrence probability of the first behavior of the first user; a fourth construction unit that constructs a second behavior estimation model including the second reception unit and the second conversion unit constructed by the second construction unit, and the third output unit constructed by the third construction unit, the third output unit outputting second behavior prediction information indicating an occurrence probability of the second behavior of the second user based on the second feature vector; An information processing device having the above configuration.

2. the first related behavior data includes time information indicating a time of occurrence of the first related behavior, and the second related behavior data includes time information indicating a time of occurrence of the second related behavior, the first construction unit constructs the first context estimation model including the first output unit that outputs, as the first context information, time information indicating an occurrence time of the first related behavior of the first user; the second construction unit constructs the second context estimation model, the second output unit being configured to output time information indicating an occurrence time of the second related behavior of the second user as the second context information having the same expression format as the first context information; The information processing device according to claim 1 .

3. the first related behavior data includes location information indicating a location where the first related behavior occurred, and the second related behavior data includes location information indicating a location where the second related behavior occurred; the first construction unit constructs the first context estimation model including the first output unit that outputs, as the first context information, location information indicating a location where the first related behavior of the first user occurred; the second construction unit constructs the second context estimation model, the second output unit being configured to output location information indicating a location where the second related behavior of the second user occurred, as the second context information having the same expression format as the first context information; 3. The information processing device according to claim 1 or 2.

4. each of the first item identification information and the second item identification information is associated with text information describing the item; the first construction unit constructs the first context estimation model, the first construction unit having the first reception unit that receives an input of the first item vector based on the text information associated with the first item identification information, and the first output unit that outputs, as the first context information, the first item vector corresponding to another first item related to the first item indicated by the first item vector; the second construction unit constructs the second context estimation model, the second construction unit having the second reception unit that receives an input of the second item vector based on the text information associated with the second item identification information, and the second output unit that outputs, as the second context information, the second item vector corresponding to another second item related to the second item indicated by the second item vector. The information processing device according to claim 1 .

5. the data acquisition unit acquires the first related behavior data including the first item identification information corresponding to the first related behavior related to the first behavior and the first user identification information, and acquires the second related behavior data including the second item identification information corresponding to a second related behavior related to the second behavior having a predetermined relationship with the first behavior and the second user identification information. The information processing device according to claim 1 .

6. a determination unit that receives an input of at least one of the first feature vector converted by the first conversion unit from the first user vector and the first item vector and the second feature vector converted by the second conversion unit from the second user vector and the second item vector, and outputs determination information indicating whether the input feature vector is the first feature vector or the second feature vector; The determination unit learns so as to increase the accuracy rate of the output result of the determination information, At least one of the first construction unit and the second construction unit learns so that a rate of accuracy of a judgment result of the judgment unit when the judgment unit judges a feature vector generated by the first construction unit and the second construction unit approaches 50%. The information processing device according to claim 1 .

7. the first reception unit receives an input of a vector obtained by averaging a plurality of the first user vectors and a first item vector corresponding to the first item identification information corresponding to the plurality of first user vectors; the second reception unit receives input of a vector obtained by averaging a plurality of the second user vectors and a second item vector corresponding to the second item identification information corresponding to the plurality of second user vectors; The information processing device according to claim 1 .

8. The computer executes The method includes a step of acquiring first related behavior data including first item identification information for identifying a first item, which is an item corresponding to a first related behavior related to a first behavior of a first user who is a user of a first domain, and first user identification information for identifying the first user, and second related behavior data including second item identification information for identifying a second item, which is an item corresponding to a second related behavior related to a second behavior of a second user who is a user of a second domain different from the first domain, and second user identification information for identifying the second user, the first action and the second action are actions of the same type, and the first related action and the second related action are actions of the same type; constructing a first context estimation model using the first related behavior data, the first context estimation model having a first reception unit that receives an input of a first user vector corresponding to the first user identification information and a first item vector corresponding to the first item identification information and indicating a feature of the first item, a first conversion unit that converts the input vector into a first feature vector of a predetermined number of dimensions that indicates a feature of the vector, and a first output unit that outputs first context information, which is information regarding the first related behavior of the first user, based on the converted first feature vector; a step of constructing a second context estimation model using the second related behavior data, the second context estimation model including a second reception unit that receives an input of a second user vector corresponding to the second user identification information and a second item vector that corresponds to the second item identification information and is expressed in the same feature space as the first item vector and indicates features of the second item, a second conversion unit that converts the input vector into a second feature vector of the predetermined number of dimensions that indicates features of the vector, and a second output unit that outputs second context information, which is information related to the second related behavior of the second user and has the same expression format as the first context information, based on the converted second feature vector; constructing a first behavior estimation model using the first related behavior data and data indicating an occurrence result of the first behavior with respect to the first related behavior, the first receiving unit, the first converting unit, and a third output unit that outputs first behavior prediction information indicating an occurrence probability of the first behavior of the first user based on the converted first feature vector; constructing a second behavior estimation model including the second reception unit and the second conversion unit constructed in the step of constructing the second context estimation model, and the third output unit constructed in the step of constructing the first behavior estimation model, the third output unit outputting second behavior prediction information indicating an occurrence probability of the second behavior of the second user based on the second feature vector; An information processing method comprising the steps of:

Images