Information processing device and information processing method

Abstract

This invention provides an information processing device and an information processing method that process the original data while considering the relationship between the original data to be anonymized and other data. [Solution] The information processing device 1 includes an acquisition unit 131 that acquires user data relating user identification information for each of multiple users, first data, and second data related to the first data but with a simpler data structure than the first data, and a generation unit 132 that inputs each of the two first data corresponding to the two second data belonging to the same cluster when the second data is clustered into a learning model that outputs feature quantities that show the characteristics of the first data in response to the input of the first data, and compared the similarity of the two feature quantities output from the learning model with the similarity of the two first data corresponding to the two second data belonging to different clusters into the learning model, and generates a learning model that has been trained so that the similarity of the two feature quantities output from the learning model is small.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus and an information processing method.

Background Art

[0002] Conventionally, user information, which is information about users, has been collected and data analysis has been performed. In this case, in order to protect the privacy of users, at least a part of the user information is processed such as K-anonymization to anonymize the user information (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Many anonymization techniques such as K-anonymization perform anonymization by processing the original data. On the other hand, when the original data to be processed is high-dimensional or has a complex structure, applying the anonymization technique causes a significant loss of information in the original data.

[0005] When avoiding a significant loss of information in the original data, it is conceivable to convert the original data so that the meaning of the original data is not lost. However, if the original data is converted without considering the relationship between the original data and other data related to the original data, the relationship between the original data and other data is lost, which may have an adverse effect on data analysis using the converted data and other data.

[0006] Therefore, the present invention has been made in view of these points, and aims to enable processing of the original data while considering the relationship between the original data to be anonymized and other data. [Means for solving the problem]

[0007] An information processing device according to a first aspect of the present invention includes: an acquisition unit that acquires user data relating user identification information for identifying each of a plurality of users collected by a predetermined business operator, first data and second data related to the first data, the second data having a simpler data structure than the first data; and a generation unit that generates a learning model that outputs feature quantities indicating the characteristics of the first data in response to the input of the first data, inputting each of the two first data corresponding to two second data belonging to the same cluster when the second data is clustered into the learning model, and inputting each of the first data corresponding to two second data belonging to different clusters into the learning model, thereby generating a learning model that is trained so that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model.

[0008] The information processing device may further include a replacement unit that inputs each of the multiple first data contained in the user data into the learning model, acquires a feature corresponding to each of the multiple first data, and replaces each of the multiple first data contained in the user data with the acquired feature.

[0009] The generation unit may perform K-anonymization on the second data, input each of the two first data corresponding to the two second data belonging to the same cluster when the K-anonymized second data is clustered into the learning model, and input each of the two first data corresponding to the two second data belonging to different clusters into the learning model, thereby generating a learning model that is trained to have a smaller similarity between the two feature quantities output from the learning model.

[0010] The generation unit may extract the user identification information and the second data as partial data from the user data acquired by the acquisition unit, perform clustering of the partial data, and generate a learning model that is trained such that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model by inputting each of the two first data items corresponding to two user identification information items belonging to the same cluster into the learning model.

[0011] The first data may be location history data indicating the user's location history, and the second data may be related location history data created based on the location history data.

[0012] A second aspect of the present invention relates to an information processing method which includes the steps of: obtaining user data, which is performed by a computer, that associates user identification information for identifying each of a plurality of users collected by a predetermined business operator, with first data and second data related to the first data, the second data having a simpler data structure than the first data; and a learning model that outputs feature quantities that represent the characteristics of the first data in response to the input of the first data, wherein each of the two first data corresponding to two second data belonging to the same cluster when the second data is clustered is input into the learning model, and each of the first data corresponding to two second data belonging to different clusters is input into the learning model, and the learning model is trained to produce a learning model in which the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model. [Effects of the Invention]

[0013] According to the present invention, the original data to be anonymized can be processed while taking into account the relationship between the original data and other data. [Brief explanation of the drawing]

[0014] [Figure 1] This is a diagram illustrating the overview of an information processing device. [Figure 2] This diagram shows the functional configuration of an information processing device. [Figure 3] This figure shows examples of the first and second data sets. [Figure 4] This is a flowchart showing the processing flow in an information processing device. [Figure 5] This figure shows an example of how the first data point included in user data is converted into features. [Figure 6] This is a flowchart showing the processing flow in an information processing device. [Modes for carrying out the invention]

[0015] [Overview of Information Processing Apparatus 1] FIG. 1 is a diagram for explaining the overview of the information processing apparatus 1. The information processing apparatus 1 is a computer that aggregates user data collected by each of a plurality of operators. The information processing apparatus 1 is operated by, for example, an aggregation operator that provides a service for aggregating data and providing the aggregated data, and is communicably connected to an external device such as an operator terminal 2 via a communication network (not shown) such as the Internet or a mobile phone line.

[0016] The operator terminal 2 is, for example, a computer operated by a predetermined operator. The operator terminal 2 manages user data in which a user ID (Identification), which is user identification information for identifying each of a plurality of users collected by a predetermined operator, is associated with first data and second data related to the first data. The second data is, for example, data having a simpler data structure than the first data.

[0017] The information processing apparatus 1 acquires user data from the operator terminal 2 ((1) in FIG. 1). Although only one operator terminal is shown in FIG. 1, actually, there are a plurality of operator terminals corresponding to each of a plurality of operators, and the information processing apparatus 1 is assumed to collect data from each of the plurality of operator terminals.

[0018] The information processing apparatus 1 generates a learning model that outputs a feature amount indicating the feature of the first data in response to the input of the first data ((2) in FIG. 1). For example, the information processing apparatus 1 inputs each of two pieces of first data corresponding to two pieces of second data belonging to the same cluster when the second data is clustered into the learning model, and compares the degree of similarity indicating the degree of similarity between the two feature amounts output from the learning model, and inputs each of the two pieces of first data corresponding to two pieces of second data belonging to different clusters into the learning model, and generates a learning model learned so that the degree of similarity between the two feature amounts output from the learning model becomes small.

[0019] The information processing apparatus 1 inputs each of a plurality of first data included in user data into the generated learning model, and acquires feature amounts corresponding to each of the plurality of first data from the learning model (step (3) in FIG. 1). Then, the information processing apparatus 1 converts each of the plurality of first data into the acquired feature amounts (step (4) in FIG. 1).

[0020] By doing so, compared with the similarity of two feature amounts corresponding to two first data corresponding to two second data belonging to the same cluster when the second data output from the learning model is clustered, the similarity of two feature amounts corresponding to two first data corresponding to two second data belonging to different clusters becomes smaller, and these feature amounts maintain the similarity and dissimilarity relationship of the second data. Therefore, the information processing apparatus 1 can perform processing on the first data while considering the relationship with the second data by replacing the first data with the feature amounts and performing anonymization.

[0021] [Functional Configuration of Information Processing Apparatus 1] Next, the functional configuration of the information processing apparatus 1 will be described. FIG. 2 is a diagram showing the functional configuration of the information processing apparatus 1.

[0022] As shown in FIG. 2, the information processing apparatus 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 operator terminal 2 and the like via a communication network.

[0023] The storage unit 12 is a storage medium that stores various types of data, and includes a ROM (Read Only Memory), a RAM (Random Access Memory), a hard disk, an SSD (Solid State Drive), a flash memory, and the like. 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 an acquisition unit 131, a generation unit 132, a replacement unit 133, and an output unit 134.

[0024] The control unit 13 is, for example, a CPU (Central Processing Unit). The control unit 13 functions as an acquisition unit 131, a generation unit 132, a replacement unit 133, and an output unit 134 by executing a program stored in the storage unit 12.

[0025] The acquisition unit 131 acquires user data that associates the user ID of each of several users collected by a predetermined business operator with first data and second data related to the first data, the second data having a simpler data structure than the first data. The first data is, for example, location history data that shows the user's location history, and the second data is location history related data created based on the location history data.

[0026] Figure 3 shows an example of user data. As shown in Figure 3, it can be seen that the user ID is associated with location history data showing the user's location history, and related location history data such as the number of visits to location A and the number of visits to location B. For simplicity of explanation, the example shown in Figure 3 assumes that there are 4 user data entries.

[0027] The generation unit 132 generates a learning model that outputs feature quantities that represent the characteristics of the first data when the first data is input. Specifically, the generation unit 132 inputs each of the two first data that correspond to the two second data that belong to the same cluster when the second data is clustered into the learning model, and compared the similarity of the two feature quantities output from the learning model, which indicates the degree of similarity, the generation unit 132 inputs each of the first data that correspond to the two second data that belong to different clusters into the learning model, and generates a learning model that is trained so that the similarity of the two feature quantities output from the learning model is small. Here, the greater the similarity of the two feature quantities, the more similar the two feature quantities are considered to be.

[0028] Specifically, first, the generation unit 132 extracts the user ID and second data as partial data from the user data acquired by the acquisition unit 131, and performs clustering based on the second data included in the partial data. Here, the generation unit 132 may perform K-anonymization on the second data included in the user data before extracting the partial data, and then perform clustering based on the second data after K-anonymization.

[0029] Figure 4 shows an example of clustering the secondary data included in the user data shown in Figure 3. As shown in Figure 4, it can be confirmed that the secondary data included in the four user data entries are classified into two clusters.

[0030] Next, the generation unit 132 generates a learning model in which the similarity of two features output from the learning model is smaller than the similarity of two features output from the learning model when two first data points corresponding to two user IDs belonging to different clusters are input to the learning model when two first data points corresponding to two user IDs belonging to the same cluster in the partial data is input to the learning model.

[0031] Specifically, P(i) is the set of indices of users in the same cluster as the i-th user, A(i) is the set of indices of all users except the i-th user, and z i , z p , z a Let be the feature quantities for the i-th, p-th, and a-th users, respectively, and let τ be the temperature parameter.

[0032] The generation unit 132 learns to minimize the control loss output by the loss function L shown in equation (1) below.

number

[0033] The replacement unit 133 inputs each of the multiple first data points included in the user data into the learning model and obtains a feature corresponding to each of the multiple first data points. Then, the replacement unit 133 replaces each of the multiple first data points included in the user data with the obtained feature.

[0034] Figure 5 shows an example where the first data included in the user data has been converted into features. In the example shown in Figure 5, the features are represented as 0 or 1 for simplicity of explanation, but it can be seen that the location history as the first data has been converted into features. The replacement unit 133 may also perform K-anonymization on the entire user data after converting the first data into features to anonymize the user data.

[0035] The output unit 134 outputs the user data after the first data has been converted. For example, the output unit 134 outputs a file showing the user data after the first data has been converted to the storage unit 12.

[0036] [Operation Flow] Next, we will explain the processing flow related to the information processing device 1. Figure 6 is a flowchart showing the processing flow in the information processing device 1. First, the acquisition unit 131 acquires user data from the operator terminal 2 (S1).

[0037] Next, the generation unit 132 extracts the user ID and the second data as partial data from the user data (S2), and clusters the partial data based on the second data (S3). Next, the generation unit 132 extracts two user IDs from each of the clustered partial data and uses the first data associated with the extracted user IDs to train the learning model (S4). Specifically, the generation unit 132 generates the learning model by training it so that the control loss output by the loss function L shown in equation (1) is minimized.

[0038] Next, the replacement unit 133 inputs each of the multiple first data points included in the user data into the learning model and obtains the corresponding feature for each of the multiple first data points (S5). Then, the replacement unit 133 replaces each of the multiple first data points included in the user data with the obtained feature (S6). The output unit 134 outputs the user data after the first data points have been converted into feature (S7).

[0039] [Effects of Information Processing Device 1] As described above, in the information processing system S according to this embodiment, the information processing device 1 acquires user data which is obtained by associating user identification information for identifying each of multiple users, first data, and second data related to the first data, the second data having a simpler data structure than the first data. The learning model outputs feature quantities that show the characteristics of the first data in response to the input of the first data. The learning model is trained so that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model which is obtained by inputting each of the two first data items corresponding to the two second data items belonging to the same cluster when the second data items are clustered.

[0040] Furthermore, this invention makes it possible to contribute to Goal 9 of the United Nations-led Sustainable Development Goals (SDGs), "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation." In addition, all or part of the device can be configured by functionally or physically distributing and integrating it in any unit. Moreover, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of this invention. The effects of the new embodiments resulting from the combinations are combined with the effects of the original embodiments. [Explanation of Symbols]

[0041] 1. Information Processing Device 2 1st device 3 Second device 11 Communications Department 12 Storage section 13 Control Unit 131 Acquisition Department 132 Generation part 133 Substitution part 134 Output section

Claims

1. An acquisition unit that acquires user data by associating user identification information for identifying each of multiple users collected by a designated business operator, first data, and second data related to the first data, the second data having a simpler data structure than the first data, A learning model that outputs feature quantities representing the characteristics of the first data in response to the input of the first data, comprising: a generation unit that inputs two first data corresponding to two second data belonging to the same cluster when the second data is clustered into the learning model, and generates a learning model that is trained such that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model, by inputting two first data corresponding to two second data belonging to different clusters into the learning model; An information processing device having

2. The system further includes a replacement unit that inputs each of the multiple first data points included in the user data into the learning model, obtains a feature corresponding to each of the multiple first data points, and replaces each of the multiple first data points included in the user data with the obtained feature. The information processing apparatus according to claim 1.

3. The generation unit performs K-anonymization on the second data, inputs two first data sets corresponding to two second data sets belonging to the same cluster when the K-anonymized second data sets are clustered into the learning model, inputs two first data sets corresponding to two second data sets belonging to different clusters into the learning model, and generates a learning model that is trained so that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model. The information processing apparatus according to claim 1.

4. The generation unit extracts the user identification information and the second data as partial data from the user data acquired by the acquisition unit, performs clustering of the partial data, and generates a learning model that is trained such that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model by inputting each of the two first data quantities corresponding to two user identification information belonging to different clusters into the learning model, by inputting each of the two first data quantities corresponding to two user identification information belonging to different clusters into the learning model. The information processing apparatus according to claim 1.

5. The first data is location history data indicating the user's location history, and the second data is related location history data created based on the location history data. The information processing apparatus according to claim 1.

6. A computer executes A step of obtaining user data by associating user identification information for identifying each of several users collected by a designated business operator, first data, and second data related to the first data, the second data having a simpler data structure than the first data; A learning model that outputs feature quantities representing the characteristics of the first data in response to the input of the first data, comprising the steps of: inputting two sets of first data corresponding to two sets of second data belonging to the same cluster when the second data is clustered into the learning model, and inputting two sets of first data corresponding to two sets of second data belonging to different clusters into the learning model, thereby generating a learning model that is trained so that the similarity of the two feature quantities output from the learning model is smaller than the similarity of the two feature quantities output from the learning model, An information processing method having

Images