Information processing device and information processing method

Abstract

To enable matching of first data and second data of the same user with high accuracy. [Solution] The information processing device 1 has a feature acquisition unit 133 that inputs first location history data and acquires first features that are features of each of a plurality of first users, and inputs second location history data and acquires second features that are features of each of a plurality of second users, for a learning model trained so that a positive example distance, which is the distance between a feature corresponding to first location history data of an anchor user and a feature corresponding to second location history data of a user other than the anchor user, is smaller than a negative example distance, which is the distance between a feature corresponding to the first location history data of the anchor user and a feature corresponding to second location history data of a user other than the anchor user, and a matching unit 134 that matches a plurality of first location history data with a plurality of second location history data based on the distance between the plurality of first features and the plurality of second features.

Description

[Technical field]

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

[0002] Conventionally, data of the same user included in a first data group collected by a first business operator and a second data group collected by a second business operator are associated with each other. For example, in Patent Document 1, the first data group includes first data of a plurality of users, and the second data group includes second data of a plurality of users, and the first data and the second data are matched based on whether or not the user's location indicated by the location history information included in the first data is similar to the user's location indicated by the location history information included in the second data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2014-044699 A Summary of the Invention [Problem to be solved by the invention]

[0004] When multiple users are present in the same location, the technique described in Patent Document 1 has a problem in that it is not possible to match the first data and the second data of the same user with high accuracy.

[0005] The present invention has been made in consideration of these points, and has an object to enable accurate matching of first data and second data of the same user. [Means for solving the problem]

[0006] An information processing device according to a first aspect of the present invention includes a location history data acquisition unit that acquires a first data group including a plurality of first location history data associating first user identification information for identifying each of a plurality of first users collected by a first operator with the location history of each of the first users, and a second data group including a plurality of second location history data associating second user identification information for identifying each of a plurality of second users collected by a second operator with the location history of each of the second users, and a learning model that outputs, in response to an input of location history data, a feature amount indicating a feature of a location history indicated by the location history data, wherein a positive example distance, which is a distance between the feature amount output in response to an input of the first location history data corresponding to an anchor user who is a user used as a reference, and the feature amount output in response to an input of the second location history data of the same user as the anchor user, is greater than or equal to a positive example distance corresponding to a previous user corresponding to the anchor user. the learning model is trained so that a feature value obtained by inputting the first location history data of each of the multiple first users into the learning model, the learning model having been trained so that a feature value obtained by inputting the first location history data of each of the multiple first users is smaller than a negative example distance, which is a distance between a feature value output in response to an input of the first location history data of a user different from the anchor user and a feature value output in response to an input of the second location history data of a user different from the anchor user; and a feature acquisition unit that acquires a first feature value that is the feature value of each of the multiple first users by inputting the second location history data of each of the multiple second users into the learning model; and a matching unit that matches the multiple first location history data with the multiple second location history data based on a distance between the first feature value of each of the multiple first users acquired by the feature acquisition unit and the second feature value of each of the multiple second users.

[0007] The learning model may be a model trained using a triplet loss that maximizes the distance between the positive example distance and the negative example distance as a loss function.

[0008] The location history data included in the first data group or the second data group is associated with associated data related to the location history, and the learning model may be a learning model that has been further trained so that the similarity between two feature values ​​output for each input of two location history data corresponding to two of the associated data belonging to different clusters is smaller than a similarity indicating the degree of similarity between two feature values ​​output for each input of two location history data corresponding to two of the associated data included in the same cluster when the first data group or the second data group is clustered based on the associated data.

[0009] The user different from the anchor user may be a user whose associated data belongs to a cluster different from a cluster to which the associated data corresponding to the anchor user belongs.

[0010] An information processing method according to a second aspect of the present invention includes a step of acquiring, executed by a computer, a first data group including a plurality of first location history data associating first user identification information for identifying each of a plurality of first users collected by a first operator with the location history of each of the first users, and a second data group including a plurality of second location history data associating second user identification information for identifying each of a plurality of second users collected by a second operator with the location history of each of the second users, and a learning model that outputs, in response to an input of location history data, a feature amount indicating a feature of a location history indicated by the location history data, wherein a positive example distance, which is a distance between a feature amount output in response to an input of the first location history data corresponding to an anchor user who is a user set as a reference, and a feature amount output in response to an input of the second location history data of the same user as the anchor user, is greater than or equal to a positive example distance of the anchor user. The method includes the steps of: acquiring first features that are the features of each of the multiple first users by inputting the first location history data of each of the multiple first users into the learning model, which is trained so that a distance between a feature output in response to an input of the first location history data corresponding to a user and a feature output in response to an input of the second location history data of a user different from the anchor user is smaller than a negative example distance, which is a distance between a feature output in response to an input of the first location history data corresponding to a user and a feature output in response to an input of the second location history data of a user different from the anchor user; and acquiring second features that are the features of each of the multiple second users by inputting the second location history data of each of the multiple second users into the learning model; and matching the multiple first location history data with the multiple second location history data based on a distance between the acquired first feature of each of the multiple first users and the second feature of each of the multiple second users. Effect of the Invention

[0011] According to the present invention, it is possible to achieve an effect of matching the first data and the second data of the same user with high accuracy. [Brief description of the drawings]

[0012] [Figure 1]FIG. 1 is a diagram illustrating an overview of an information processing device. [Diagram 2] FIG. 2 is a diagram illustrating a functional configuration of an information processing device. [Diagram 3] FIG. 4 is a diagram showing an example of a first data group and a second data group. [Figure 4] 4 is a flowchart showing a process flow in an information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] [Overview of information processing device 1] 1 is a diagram illustrating an overview of an information processing device 1. The information processing device 1 is a computer that performs matching between first position history data indicating the position histories of multiple first users and second position history data indicating the position histories of multiple second users.

[0014] The information processing device 1 is operated by an aggregation service provider that provides a service of aggregating data and providing the aggregated data, and is communicatively connected to external devices such as a first device 2 and a second device 3 via a communication network (not shown) such as the Internet or a mobile phone line.

[0015] The first device 2 is, for example, a computer operated by a first business operator. The first device 2 manages a first data group including a plurality of first location history data in which a first user ID (Identification), which is first user identification information for identifying each of a plurality of first users collected by the first business operator, is associated with the location history of each of the first users. The second device 3 is, for example, a computer operated by a second business operator. The second device 3 manages a second data group including a plurality of second location history data in which a second user ID, which is second user identification information for identifying each of a plurality of second users collected by the second business operator, is associated with the location history of each of the second users.

[0016] Among the multiple second users and the multiple first users, there are some identical users. However, the first user identification information and the second user identification information corresponding to the identical user are different identification information, and it is not possible to match the location history data of the identical user by matching the first user identification information and the second user identification information.

[0017] The information processing device 1 acquires a first data group from the first device 2, and acquires a second data group from the second device 3. In order to match the first position history data with the second position history data, the information processing device 1 can use a learning model that outputs, in response to input of position history data, a feature amount indicating the characteristics of the position history indicated by the position history data.

[0018] The learning model is trained so that the positive example distance, which is the distance between the feature output in response to an input of first location history data corresponding to an anchor user, who is a reference user, and the feature output in response to an input of second location history data of the same user as the anchor user, is smaller than the negative example distance, which is the distance between the feature output in response to an input of first location history data corresponding to the anchor user, and the feature output in response to an input of second location history data of a user different from the anchor user.

[0019] The information processing device 1 inputs the first location history data of each of the multiple first users into the learning model to obtain a first feature amount that is a feature amount of each of the multiple first users, and inputs the second location history data of each of the multiple second users into the learning model to obtain a second feature amount that is a feature amount of each of the multiple second users. Then, the information processing device 1 matches the multiple first location history data with the multiple second location history data based on the distance between the first feature amount of each of the multiple first users and the second feature amount of each of the multiple second users. In this way, by performing matching based on the feature amount output from the learning model that is trained so that the distance between the first feature amount and the second feature amount of the same user is small and the distance between the first feature amount and the second feature amount of different users is large, the information processing device 1 can accurately match the first data and the second data of the same user.

[0020] [Functional configuration of information processing device 1] Next, a description will be given of the functional configuration of the information processing device 1. FIG.

[0021] As shown in FIG. 2, the information processing device 1 includes a communication unit 11, a storage unit 12, and a control unit 13. The communication unit 11 is a communication interface for transmitting and receiving data to and from the first device 2, the second device 3, and the like via a communication network.

[0022] The storage unit 12 is a storage medium that stores various data, and includes a ROM (Read Only Memory), a RAM (Random Access Memory), a hard disk, an SSD (Solid State Drive), a flash memory, etc. The storage unit 12 stores a program executed by the control unit 13. The storage unit 12 stores a program that causes the control unit 13 to function as a position history data acquisition unit 131, a generation unit 132, a feature acquisition unit 133, a matching unit 134, and an output unit 135.

[0023] The control unit 13 is, for example, a CPU (Central Processing Unit). The control unit 13 executes a program stored in the storage unit 12 to function as a position history data acquisition unit 131, a generation unit 132, a feature amount acquisition unit 133, a matching unit 134, and an output unit 135.

[0024] The location history data acquisition unit 131 acquires a first data group including a plurality of first location history data associating a first user ID for identifying each of a plurality of first users collected by a first operator with the location history of each of the first users, and a second data group including a plurality of second location history data associating a second user ID for identifying each of a plurality of second users collected by a second operator with the location history of each of the second users.

[0025] FIG. 3 is a diagram showing an example of a first data group and a second data group. FIG. 3(A) shows the first data group, and FIG. 3(B) shows the second data group. As shown in FIG. 3(A), the first location history data in which the first user ID and the location history are associated is associated with data indicating the number of visits of the first user to the point A specified based on the location history and the number of visits of the first user to the point B as related data related to the location history. Also, as shown in FIG. 3(B), the second location history data in which the second user ID and the location history are associated is associated with data indicating whether the first user specified based on the location history has visited a supermarket and data indicating whether the second user has visited a park as related data related to the location history.

[0026] Here, the location history is, for example, information that associates a date and time with latitude and longitude information that indicates the user's location corresponding to the date and time, and indicates the location of each of a plurality of dates and times. Also, the associated data is, for example, data with fewer dimensions than the location history.

[0027] In addition, in this embodiment, as shown in Figures 3(A) and (B), the first user ID and the second user ID are data of different systems, and the first user ID and the second user ID of the same user are different. As a result, by associating the first user ID and the second user ID, it is not possible to match the location history data of the same user.

[0028] The generation unit 132 generates a learning model that outputs a feature amount indicating a feature of a location history indicated by the location history data in response to the input of the location history data. For example, the generation unit 132 generates the learning model by performing a first learning and a second learning.

[0029] The first learning is learning that reduces the distance between the feature quantities of the same user included in the first position history data and the second position history data, and increases the distance between the feature quantities of different users. For example, the generating unit 132 accepts, from a user terminal (not shown) used by a user who integrates the first data group and the second data group, designation of the first position history data corresponding to an anchor user who is a reference user, among the first position history data of each of the multiple users included in the first data group. In addition, the generating unit 132 accepts, from the user, designation of the second position history data corresponding to the anchor user among the second position history data of each of the multiple users included in the second data group, and designation of the second position history data corresponding to a user different from the anchor user. The first position history data and the second position history data corresponding to the anchor user are called positive example data, and a combination of the first position history data corresponding to the anchor user and the second position history data corresponding to another user is called negative example data.

[0030] The generation unit 132 performs learning so that a positive example distance, which is a distance between a feature amount output by the learning model for an input of first position history data corresponding to an anchor user and a feature amount output by the learning model for an input of second position history data of the same user as the anchor user, is smaller than a negative example distance, which is a distance between a feature amount output by the learning model for an input of first position history data corresponding to the anchor user and a feature amount output by the learning model for an input of second position history data of a user different from the anchor user. Specifically, the generation unit 132 performs the first learning so that a triplet loss that maximizes the distance between the positive example distance and the negative example distance is set as a loss function L1, and the triplet loss output by the loss function L1 is minimized.

[0031] Here, the feature quantity that the learning model outputs in response to the input of the first location history data corresponding to the anchor user is denoted as F, and the feature quantity that the learning model outputs in response to the input of the second location history data of the same user as the anchor user is denoted as F. + The feature quantity that the learning model outputs for the input of the second location history data of a user different from the anchor user is F - , and the distance margin is m. The generation unit 132 then trains the learning model so that the loss function L1 shown in the following formula (1) is minimized when a plurality of positive example data and negative example data are input to the learning model to acquire features. Here, the d function is a distance function, for example, the Euclidean distance. The max function is a function that returns the maximum value among a plurality of values ​​included in the max function.

number

[0032] The second learning is learning that reduces the distance between the features of multiple users who behave similarly in the first data group or the second data group, and increases the distance between the features of multiple users who behave differently. An example of performing the second learning using the first data group will be described below.

[0033] For example, the generating unit 132 clusters the first data group based on the related data included in the first data group. The generating unit 132 then performs the second learning so that the similarity of two feature amounts output for each input of two position history data corresponding to two related data belonging to different clusters is smaller than the similarity indicating the degree of similarity between two feature amounts output for each input of two position history data corresponding to two related data included in the same cluster when the first data group is clustered based on the related data. Here, it is assumed that the greater the similarity between the two feature amounts, the more similar the two feature amounts are.

[0034] Specifically, P(i) is the index set of users in the same cluster as the i-th user, A(i) is the index set of all users other than the i-th user, and z i , z p , z a are the feature quantities of the i-th, p-th, and a-th users, respectively, and τ is a temperature parameter for adjusting the contrast loss output by the loss function. The generation unit 132 then performs learning of the learning model so that the contrast loss output by the loss function L2 shown in the following formula (2) is minimized.

[0035]

number

[0036] In addition, when performing the first learning, it is preferable that the user different from the anchor user is a user whose associated data belongs to a cluster different from the cluster to which the associated data corresponding to the anchor user belongs. In response to this, the generation unit 132 accepts, from the user terminal of the user who integrates the first data group and the second data group, designation of the second location history data corresponding to the anchor user among the second location history data of each of the multiple users included in the second data group, and accepts designation of the second location history data corresponding to a user who belongs to a cluster different from the cluster to which the anchor user belongs when the associated data associated with the second location history data is clustered as a user different from the anchor user.

[0037] In this way, during the first learning, the information processing device 1 learns with negative example data based on the features of other users who behave differently from the anchor user, and can learn in a way that brings out the differences between the features of this anchor user and other users, thereby improving learning efficiency compared to when learning with negative example data based on the features of users similar to the anchor user.

[0038] In addition, when the location history information is location history information in a space where multiple users with the same hobbies and interests gather, such as an event venue, it may be difficult to distinguish between the actions of multiple users. In response to this case, when performing the first learning, a user other than the anchor user may be a user whose related data belongs to the same cluster as the cluster to which the related data corresponding to the anchor user belongs. In this way, the information processing device 1 can learn to distinguish between features even for multiple users who behave in a similar manner.

[0039] The feature amount acquiring unit 133 acquires a first feature amount which is a feature amount of each of the multiple first users by inputting the first position history data of each of the multiple first users acquired by the position history data acquiring unit 131 to the learning model generated by the generation unit 132. In addition, the feature amount acquiring unit 133 acquires a second feature amount which is a feature amount of each of the multiple second users by inputting the second position history data of each of the multiple second users acquired by the position history data acquiring unit 131 to the learning model generated by the generation unit 132.

[0040] The matching unit 134 matches the multiple first location history data with the multiple second location history data based on the distance between the first feature of each of the multiple first users acquired by the feature acquisition unit 133 and the second feature of each of the multiple second users.

[0041] For example, the matching unit 134 identifies the second feature quantity that is closest to the first feature quantity of each of the multiple first users acquired by the feature quantity acquisition unit 133. Then, the matching unit 134 determines that the first user of the first location history data corresponding to the first feature quantity is the same as the second user of the second location history data corresponding to the second feature quantity, on the condition that the distance between the first feature quantity and the second feature quantity identified for the first feature quantity is equal to or less than a predetermined threshold. As a result of matching, multiple second users may correspond to one first user. In this case, the matching unit 134 may solve this problem by associating the features so that the transportation distance is minimized.

[0042] The output unit 135 associates the first position history data and related data of the first user determined to be the same user based on the result of the matching unit 134 with the second position history data and related data of the second user, and outputs the data as integrated data. For example, the output unit 135 outputs the integrated data to a user terminal used by the user who integrates the first data group and the second data group.

[0043] [Operation flow] Next, a description will be given of the flow of processing related to the information processing device 1. Fig. 4 is a flowchart showing the flow of processing in the information processing device 1. First, the position history data acquisition unit 131 acquires a first data group from the first device 2, and acquires a second data group from the second device 3 (S1).

[0044] Next, the generation unit 132 accepts, from the user terminal of the user using the information processing device 1, designation of the first location history data of the anchor user and the second location history data of the anchor user as positive example data, and also accepts designation of the first location history data of the anchor user and the second location history data of a user other than the anchor user as negative example data (S2).

[0045] Next, the generating unit 132 performs a first learning based on the positive example data and the negative example data, and also performs a second learning to generate a learning model (S3). Note that, in this flowchart, the processes of S2 and S3 are executed, but this is not limited thereto. If a learning model has already been generated, the information processing device 1 does not need to execute the processes related to S2 and S3.

[0046] Next, the feature acquisition unit 133 acquires first features that are features of each of the multiple first users by inputting the first location history data of each of the multiple first users into the generated learning model, and acquires second features that are features of each of the multiple second users by inputting the second location history data of each of the multiple second users into the learning model (S4).

[0047] Next, the matching unit 134 performs matching between the multiple first position history data and the multiple second position history data based on the distance between the first feature amount of each of the multiple first users and the second feature amount of each of the multiple second users (S5).

[0048] Next, the output unit 135 outputs the integrated data based on the result of the matching by the matching unit 134 (S6). Specifically, the output unit 135 generates integrated data by associating the first position history data of the first user and the second position history data of the second user that are determined to be identical, and outputs the integrated data to the user terminal.

[0049] [Effects of information processing device 1] As described above, in the information processing system S according to the present embodiment, the information processing device 1 inputs the first location history data of each of the multiple first users to a learning model trained so that the positive example distance, which is the distance between the feature amount output for the input of the first location history data corresponding to the anchor user and the feature amount output for the input of the second location history data of the same user as the anchor user, is smaller than the negative example distance, which is the distance between the feature amount output for the input of the first location history data corresponding to the anchor user and the feature amount output for the input of the second location history data of a user different from the anchor user, to obtain the first feature amount that is the feature amount of each of the multiple first users, and inputs the second location history data of each of the multiple second users to obtain the second feature amount that is the feature amount of each of the multiple second users. Then, the information processing device 1 performs matching between the multiple first location history data and the multiple second location history data based on the distance between the first feature amount of each of the multiple first users acquired and the second feature amount of each of the multiple second users. In this way, the information processing device 1 can accurately match the first data and the second data of the same user.

[0050] The present invention can contribute to Goal 9 of the United Nations' Sustainable Development Goals (SDGs), "Build resilient infrastructure, promote inclusive and sustainable industrialization, and promote innovation and infrastructure." All or part of the device can be functionally or physically distributed or integrated in any unit. New embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effect of the new embodiment resulting from the combination combines the effects of the original embodiments. [Explanation of symbols]

[0051] 1. Information processing device 2 1st device 3 Second device 11 Communications Department 12 Storage section 13 Control section 131 Location history data acquisition unit 132 Generation part 133 Feature Acquisition Unit 134 Matching Section 135 Output section

Claims

1. a location history data acquisition unit that acquires a first data group including a plurality of first location history data associating first user identification information for identifying each of a plurality of first users collected by a first business operator with the location history of each of the first users, and a second data group including a plurality of second location history data associating second user identification information for identifying each of a plurality of second users collected by a second business operator with the location history of each of the second users; a feature acquisition unit that inputs the first location history data of each of the first users to the learning model, the feature acquisition unit acquiring a first feature that is the feature of each of the first users, and inputs the second location history data of each of the second users to the learning model, the feature acquisition unit acquiring a second feature that is the feature of each of the second users, by inputting the first location history data of each of the first users to the learning model, the feature acquisition unit acquiring a first feature that is the feature of each of the first users, and inputs the second location history data of each of the second users to the learning model, the feature acquisition unit acquiring a second feature that is the feature of each of the second users, by inputting the first location history data of each of the second users to the learning model, the feature acquisition unit acquiring a positive example distance that is a distance between a feature output for an input of the first location history data corresponding to an anchor user who is a reference user and a feature output for an input of the second location history data of the same user as the anchor user, which is smaller than a negative example distance that is a distance between a feature output for an input of the first location history data corresponding to the anchor user and a feature output for an input of the second location history data of a user different from the anchor user; a matching unit that performs matching between the plurality of first location history data and the plurality of second location history data based on a distance between the first feature amount of each of the plurality of first users acquired by the feature amount acquisition unit and the second feature amount of each of the plurality of second users; An information processing device having the above configuration.

2. The learning model is a model trained using a triplet loss that maximizes the distance between the positive example distance and the negative example distance as a loss function. The information processing device according to claim 1 .

3. The location history data included in the first data group or the second data group is associated with associated data related to the location history, the learning model is a learning model that has been further trained so that a similarity between two feature values ​​output for input of two position history data corresponding to two associated data belonging to different clusters is smaller than a similarity indicating a degree of similarity between two feature values ​​output for input of two position history data corresponding to two associated data included in the same cluster when the first data group or the second data group is clustered based on the associated data. The information processing device according to claim 1 .

4. The user different from the anchor user is a user whose associated data belongs to a cluster different from a cluster to which the associated data corresponding to the anchor user belongs. The information processing device according to claim 3 .

5. The computer executes acquiring a first data group including a plurality of first location history data associating first user identification information for identifying each of a plurality of first users collected by a first operator with the location history of each of the first users, and a second data group including a plurality of second location history data associating second user identification information for identifying each of a plurality of second users collected by a second operator with the location history of each of the second users; a learning model that outputs, in response to input of position history data, a feature amount indicating a feature of the position history indicated by the position history data, wherein the learning model is trained so that a positive example distance, which is a distance between a feature amount output in response to an input of the first position history data corresponding to an anchor user who is a reference user, and a feature amount output in response to an input of the second position history data of the same user as the anchor user, is smaller than a negative example distance, which is a distance between a feature amount output in response to an input of the first position history data corresponding to the anchor user and a feature amount output in response to an input of the second position history data of a user different from the anchor user; matching the plurality of first location history data with the plurality of second location history data based on a distance between the first feature amount of each of the plurality of first users and the second feature amount of each of the plurality of second users; An information processing method comprising the steps of:

Images