JP7742435B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

Methods for analyzing people flow data have been proposed in the past.

Patent No. 7027605

However, the above-mentioned conventional technology leaves room for improvement in terms of providing people flow data that meets the user's needs.

For example, the above-mentioned conventional technology acquires location information from user devices and arranges the location information in chronological order based on time information indicating when the location information was acquired, thereby generating the user device's travel route. Based on the trip data obtained from the cleansed travel route, the number of samples (the number of user devices that acquired location information) is expanded to the population of the estimated area, thereby estimating the flow of people within the population of the target area.

For these reasons, the above-mentioned conventional technology can be said to efficiently analyze people flow data over a wide area. However, the above-mentioned conventional technology cannot, for example, generate highly accurate people flow data that controls some of the features contained in the people flow data according to the user's purpose while reflecting other features as they are. In other words, the above-mentioned conventional technology cannot necessarily provide people flow data that meets the user's purpose.

The present invention therefore proposes an information processing device, information processing method, and information processing program that can provide people flow data tailored to the user's purpose.

To solve the above problem, one form of information processing device according to the present invention includes a conversion unit that converts time-series people flow data, which is people flow data of a population within a predetermined range, into function data; an extraction unit that extracts multiple people flow components by performing principal component analysis on the function data; a people flow control unit that controls a predetermined people flow component from among the multiple people flow components to be changed in accordance with the user's purpose; and a generation unit that generates people flow data to be provided to the user based on the multiple people flow components, including the changed people flow component.

This invention makes it possible to provide people flow data tailored to the user's needs.

FIG. 1 is a diagram (1) showing an example of people flow data according to the embodiment. FIG. 2 is a diagram (2) showing an example of people flow data according to the embodiment. FIG. 3 is an explanatory diagram illustrating the prior art. FIG. 4 is an explanatory diagram illustrating the proposed technology. FIG. 5 is a diagram illustrating an example of a system according to an embodiment. FIG. 6 is a diagram illustrating an example of the configuration of an information processing apparatus according to the embodiment. FIG. 7 is a diagram in which the people flow data shown in FIG. 1 is plotted by variable. FIG. 8 is a diagram showing an example of principal component people flow data. FIG. 9 is a flowchart showing the procedure of information processing according to the embodiment. FIG. 10 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the information processing device.

Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. Note that in this specification and drawings, components having substantially the same functional configuration will be assigned the same reference numerals, and redundant explanations will be omitted.

One or more embodiments (including examples, modifications, and application examples) described below can be implemented independently. However, at least a portion of the multiple embodiments described below may be implemented in appropriate combination with at least a portion of another embodiment. These multiple embodiments may include novel features that are different from each other. Therefore, these multiple embodiments may contribute to solving different purposes or problems and may achieve different effects from each other.

(Embodiment)
1. Introduction
Currently, time series data is being used for a variety of services. For example, time series data from sensors is sometimes used to detect various anomalies. For example, predictive services such as the detection of disasters and accidents, or the detection of anomalies in train operation status are being realized based on time series sensor data showing human movement.

There are also services that use time-series sales data to predict how much sales will change depending on temperature and foot traffic.

On the other hand, there are cases where people flow data itself needs to be corrected rather than simply used for prediction. It is generally believed that people flow trends will be similar between the past and present. For this reason, there is a need to appropriately correct people flow data depending on the purpose.

For example, there is a need to control certain features contained in this year's people flow data that are thought not to occur under normal circumstances based on past cases. To give one example, there is a need to remove from this year's people flow data features that correspond to unusual population increases or decreases that are not expected to occur this year (for example, a sharp decrease in population in response to movement restrictions due to a state of emergency declaration, or a sudden increase in population after movement restrictions are lifted).

Furthermore, for example, if we assume that the flow of people has increased or decreased, it may be necessary to estimate a scenario based on the prediction results, rather than simply predicting what the flow of people will be in the future.

This invention proposes a method for generating highly accurate people flow data by correcting the features of people flow data so that it meets the user's needs for people flow data. For example, changing the features contained in people flow data to suit the user's purpose may result in the loss of other important features, making the people flow data obtained less accurate and making it difficult to estimate scenarios. Therefore, this invention aims to generate highly accurate people flow data that changes some of the features contained in the people flow data to suit the user's purpose, while reflecting the other features as they are without controlling them.

[2. People Flow Data]
First, we will explain the people flow data used in the information processing according to the proposed technology of the present invention (information processing according to the embodiment). The people flow data here refers to a transition matrix of the number of people who moved from one range to another range between a specific timing (e.g., date and time) and the next timing. For example, by applying principal component analysis (PCA) to preprocessed multidimensional people flow data, converting it to two dimensions, and plotting it, it is possible to visualize features corresponding to the principal components.

An example of people flow data will be explained using Figures 1 and 2. The people flow data shown in Figures 1 and 2 is multidimensional (multivariate) data containing multiple variables regarding population observations for a specified period of time within a specified range, and also has a time series concept. In other words, the people flow data shown in Figures 1 and 2 is time series people flow data containing multiple variables.

First, let us explain Figure 1. Figure 1 is a diagram (1) showing an example of people flow data according to an embodiment. The people flow data shown in Figure 1 is multidimensional data with a relatively small number of variables.

Figure 1(a) shows an example of observations for obtaining people flow data. As shown in Figure 1(a), people flow data is observed in trip class units, which classify areas according to the distance traveled by users (trips). A trip class is an example of a predetermined range. Figure 1(a) shows a set of trip classes: "Trip Class 1," "Trip Class 2," "Trip Class 3," and "Trip Class 4" (written as "Trip Class 1" ... "Trip Class 4").

As shown in Figure 1(a), people flow data is time series data obtained by statistically processing the number of people (population) observed sequentially (e.g., observed every second) during each specified time period. Figure 1(a) shows examples of specified time periods: "03:00-10:59," "11:00-14:59," "15:00-18:59," and "19:00-26:59." Furthermore, according to the example in Figure 1(a), the people flow data is, more specifically, a compilation of time series data on population obtained during each time period for each day over the 274-day period from July 1, 2020 to March 31, 2021.

As shown in Figure 1(a), the people flow data includes multiple variables. This will be explained using Figure 1(b). For example, if people flow data is created by compiling time-series population data obtained for four time periods for each of four trip classes by day over a 274-day period, the people flow data will include 4 x 4 = 16 variables. Specifically, there are 16 variable combinations between the four class variables ("Trip Class 1," "Trip Class 2," "Trip Class 3," and "Trip Class 4") and the four time period variables ("03:00-10:59," "11:00-14:59," "15:00-18:59," and "19:00-26:59").

In other words, as shown in Figure 1(b), 274 days' worth of time-series people flow data exists for each of the 16 variables. Figure 1(b) shows "People Flow Data DA11" as time-series people flow data corresponding to the class variable "Trip Class 1" and the time period variable "11:00-14:59." Explanation of the people flow data corresponding to each of the other 15 variable sets will be omitted.

In this way, if the people flow data is compiled by compiling the time series population data obtained for each of the four trip classes and four time periods for each day over a 274-day period, the people flow data can be considered multidimensional data with few variables because it contains a relatively small number of variables, 16.

It should be noted that how people flow data is organized is up to you, and it can also be obtained as multidimensional data containing a large number of variables. An example of this is shown in Figure 2. Figure 2 is a diagram (2) showing an example of people flow data according to an embodiment.

Figure 2(a) shows an example of observations for obtaining people flow data. As shown in Figure 2(a), people flow data is observed in units of meshes, which are areas on a map divided into a grid based on latitude and longitude. A mesh is an example of a predetermined range. Figure 2(a) shows 2,385 meshes from "mesh0001" to "mesh2385" (represented as "mesh0001", "mesh0002", "mesh0003", "mesh0004", ... "mesh2382", "mesh2383", "mesh2384", "mesh2385").

Furthermore, as in the example of Figure 1(a), the people flow data is time series data obtained by statistically processing the number of people (population) observed sequentially (for example, observed every second) during each specified time period.

Here, according to Figure 2(a), the people flow data includes multiple variables. This point will be explained using Figure 2(b). For example, if the people flow data is a compilation of time-series population data obtained for each of 2,385 types of meshes, organized by day over a 274-day period, the people flow data will include 2,385 variables.

In other words, as shown in Figure 2(b), 274 days' worth of time-series people flow data exists for each of the 2,385 mesh variables. Figure 2(b) shows "People Flow Data X1" as the time-series people flow data corresponding to mesh variable "mesh0001." Explanation of the people flow data corresponding to the other 2,384 mesh variables will be omitted.

In this way, if the population time series data obtained for each of the 2,385 types of meshes is compiled by day over a 274-day period and used as people flow data, this people flow data can be considered multivariate, multidimensional data because it contains a large number of variables (2,385).

3. Prior Art
Prior to describing the proposed technology of the present invention, a comparative prior art will be described. FIG. 3 is an explanatory diagram illustrating the prior art. FIG. 3(a) shows a conceptual diagram of a prior art for correcting people flow data. According to the example of FIG. 3(a), in the prior art, people flow data for each variable is acquired, and features contained in the people flow data are corrected according to the user's purpose through interpolation processing (linear interpolation, curve interpolation).

Here, Figure 3(b) will specifically explain the conventional technology using an example of correcting 274 days of people flow data DA11 described in Figure 1. For example, within the period "July 1, 2020 to March 31, 2021" included in people flow data DA11, the population is rapidly decreasing during "period T1." Furthermore, within the period "July 1, 2020 to March 31, 2021" included in people flow data DA11, the population is also significantly decreasing during "period T2."

If a user predicts that such a population decline will not occur this year, they may request people flow data DA11 in which the characteristics of the population decline in the "T1 period" and the population decline in the "T2 period" have been deleted. In such cases, conventional technology uses an interpolation process to delete the characteristics of the "T1 period" and the "T2 period," and generates people flow data DA111 that has been corrected to smooth out the data in the deleted areas.

With this type of conventional technology, not only the population decline feature but also other important features may be deleted from the features contained in the people flow data DA11 for the "T1 period." Similarly, not only the population decline feature but also other important features may be deleted from the people flow data DA11 for the "T2 period." Therefore, it is difficult to say that the people flow data DA111 reflects the needs of users, and there is room for improvement in providing people flow data that meets the user's objectives.

[4. Proposed technology]
The proposed technology of the present invention is an idea devised to improve the above-mentioned problems of the conventional technology. Figure 4 is an explanatory diagram explaining the proposed technology. Figure 4(a) shows a conceptual diagram of the proposed technology for correcting people flow data. According to the example of Figure 4(a), the proposed technology acquires people flow data for each variable, and converts each acquired discrete people flow data into functional data that can be expressed as a curve.

Furthermore, multiple people flow components are extracted by performing principal component analysis on each person flow data converted into functional data. Furthermore, with the proposed technology, only the people flow component that corresponds to the user's purpose is controlled according to this purpose.

For example, a user may infer that the first principal component represents a trend based on the contribution rate and principal component loadings used when extracting multiple people flow components, and then wish to remove features from a specific period contained in the first principal component. In such cases, the proposed technology performs interpolation (linear interpolation, curve interpolation) based on the principal component scores used when extracting the first principal component. The proposed technology then reconstructs (restores) people flow data from the multiple people flow components, including the interpolated principal component.

Here, Figure 4(b) also explains the proposed technology in detail, using as an example a situation in which 274 days of people flow data DA11 described in Figure 1 is corrected. As described in Figure 3, within the period "July 1, 2020 to March 31, 2021" included in people flow data DA11, there is a sharp decrease in population during "period T1." Furthermore, within the period "July 1, 2020 to March 31, 2021" included in people flow data DA11, there is also a significant decrease in population during "period T2."

For this reason, the user considers removing the population decline feature for the "T1 period" and the population decline feature for the "T2 period" from the first principal component, which expresses these trends. In such a case, with the proposed technology, of the features included in the people flow data DA11 for the "T1 period," only the population decline feature is removed from the first principal component, and of the features included in the people flow data DA11 for the "T2 period," only the population decline feature is removed from the first principal component.

As a result, a first principal component is obtained in which the feature corresponding to the user's purpose has been removed. Therefore, with the proposed technology, people flow data is reconstructed (restored) based on the first principal component, which has been controlled to remove the feature of population decline, and the other principal components, which remain uncontrolled. Figure 4(b) shows an example of people flow data DA112 generated by reconstruction.

According to people flow data DA112, the population decline characteristics that people flow data DA11 has in the "T1 period" have been removed, while the characteristics corresponding to principal components other than the first principal component remain. Furthermore, according to people flow data DA112, the population decline characteristics that people flow data DA11 has in the "T2 period" have also been removed, while the characteristics corresponding to principal components other than the first principal component remain.

In this way, the proposed technology controls only the features that correspond to the user's purpose, while leaving other features uncontrolled, making it possible to generate people flow data that is corrected with a higher degree of accuracy than conventional technology.

5. System Configuration
Fig. 5 is a diagram showing an example of a system according to an embodiment. Fig. 5 shows a system 1 as an example of a system according to an embodiment. Information processing according to the embodiment (i.e., the proposed technology of the present invention) is realized in the system 1.

As shown in FIG. 1, system 1 includes a user device 10 and an information processing device 100. Furthermore, user device 10 and information processing device 100 are connected via network N so as to be able to communicate with each other via wired or wireless communication. Furthermore, information processing device 100 operates according to a program according to an embodiment.

The user device 10 is an information processing terminal used by a user and corresponds to an edge computer. For example, the user device 10 may be a smartphone, a wearable device, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, or a PDA (Personal Digital Assistant).

An environment (application) that allows access to the information processing device 100 may be installed on the user device 10. The application here may be a general-purpose application such as a browser, or a dedicated application for accessing the information processing device 100.

The term "user" here refers to a person who wishes to edit the features contained in people flow data according to their purpose, and may be an administrator of the information processing device 100 or a service user who requests the provision of people flow data.

The information processing device 100 is a server device that performs information processing related to the proposed technology of the present invention as described in Figure 4, and corresponds to a cloud computer. A detailed configuration example of the information processing device 100 will be described later.

6. Configuration of Information Processing Device
An information processing device 100 according to an embodiment will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the configuration of the information processing device 100 according to an embodiment. According to the example of Fig. 6, the information processing device 100 includes an external interface unit 111, a management interface unit 112, and an input/output interface unit 113.

The information processing device 100 also includes an input storage unit 120, a memory unit 130, and an output storage unit 140. According to the example in Figure 6, the input storage unit 120 includes a people flow data storage unit 120a and a setting value storage unit 120b. The memory unit 130 includes a group-specific people flow storage unit 130a and a parameter storage unit 130b. The output storage unit 140 includes a principal component people flow storage unit 140a.

The information processing device 100 also includes a preprocessing unit 151, an operation unit 152, a clustering unit 153, a learning unit 154, an inference unit 155, an editing unit 156, and a generation unit 157. According to the example of FIG. 6, the learning unit 154 includes a function data conversion unit 154a and a PCA learning unit 154b. The inference unit 155 includes a principal component score calculation unit 155a and a principal component people flow transformation unit 155b. The editing unit 156 includes a score modification unit 156a and a loading amount modification unit 156b. The generation unit 157 includes a principal component people flow correction unit 157a and a reconstruction unit 157b.

(External interface unit 111)
The external interface unit 111 acquires time-series number of people data from an external device (e.g., the user device 10). The time-series number of people data is raw data on people flow that is not summarized using variables such as those described in FIGS. 1 and 2.

(Management Interface Unit 112)
The management interface unit 112 receives input information corresponding to an input operation when the administrator performs an input operation using the user device 10. For example, the management interface unit 112 receives information for editing the characteristics of people flow data from the administrator.

(Input/output interface unit 113)
The input/output interface unit 113 receives input information corresponding to an input operation when the service user performs an input operation using the user device 10. For example, the input/output interface unit 113 receives information for editing the characteristics of people flow data from the service user.

In addition, the input/output interface unit 113 outputs people flow data reconstructed based on the principal components edited in accordance with the user's input information to the user device 10.

(People flow data storage unit 120a)
The people flow data storage unit 120a stores people flow data obtained by preprocessing the time-series number of people data. For example, the people flow data storage unit 120a stores people flow data that has been reorganized by dividing the time-series number of people data using a statistical method (e.g., kshape) or a qualitative criterion (e.g., a predetermined range such as a mesh).

(Setting value storage unit 120b)
The setting value storage unit 120b stores setting values for functional data analysis, i.e., for converting people flow data into functional data. The setting values may be input by an administrator via the management interface unit 112, for example, and may be initial values used in the functional data conversion that converts people flow data into functional data.

(Group people flow storage unit 130a)
The group-specific people flow storage unit 130a stores group-specific people flow data obtained by clustering people flow data. Using the example of FIG. 1(b), there would be 16 variables x 274 pieces of people flow data (4,384 pieces in total). Even more data could exist depending on how the variables are grouped. In this way, when the number of people flow data is enormous, principal component analysis takes a long time, and there is a risk that the features will become scattered, making it impossible to perform the principal component analysis properly. For this reason, in this embodiment, the people flow data is clustered in predetermined units. For example, in this embodiment, the people flow data is clustered by meshes containing people flow data with similar (similar) features. Note that clustering processing is not necessarily required.

(Parameter storage unit 130b)
The parameter storage unit 130b stores parameters used when learning a series of algorithms using principal component analysis. The principal component analysis algorithm is a machine learning algorithm for solving a maximization problem that maximizes the variance of data. More specifically, the algorithm for performing principal component analysis (functional principal component analysis) on function data is a machine learning algorithm for solving a maximization problem that maximizes the variance of a composite variable (principal component score) expressed as the inner product of the function data and a weighting function.

(Principal component people flow storage section 140a)
The principal component people flow storage unit 140a stores people flow data (principal component people flow data) for each principal component (people flow component) extracted by principal component analysis of people flow data. For example, the principal component people flow storage unit 140a stores data obtained by breaking down people flow data for each principal component. Here, extracting principal components means extracting new variables (principal components) from the original variables in principal component analysis. Principal components represent axes transformed in the direction of higher importance of information while retaining the information of the original variables. In other words, principal components (people flow components) summarize people flow data and are expressed in a coordinate system. Therefore, the principal component people flow data referred to here is the restored original function data (unit: people (population)) for each of multiple principal component functions obtained by summarizing using principal component analysis.

(Preprocessing unit 151)
The preprocessing unit 151 performs preprocessing by dividing and reorganizing the time-series number of people data using a statistical method (e.g., kshape) or a qualitative criterion (e.g., a predetermined range such as a mesh). The preprocessing unit 151 also registers the people flow data obtained by the preprocessing in the people flow data storage unit 120a.

(Operation unit 152)
When input information for clustering people flow data is received by the management interface, the operation unit 152 operates in accordance with the input information. For example, the operation unit 152 acquires people flow data from the people flow data storage unit 120a and operates the clustering unit 153 to cluster the acquired people flow data in accordance with the input information.

(Clustering unit 153)
The clustering unit 153 executes the clustering process described above. For example, the clustering unit 153 clusters the people flow data into meshes containing people flow data with similar (similar) characteristics, based on input information transmitted from the operation unit 152. The clustering unit 153 also registers the people flow data for each class (people flow data by group) in the people flow by group storage unit 130a.

(Function data conversion unit 154a)
The function data conversion unit 154a executes a function data conversion process to convert discrete people flow data into function data. For example, the function data conversion unit 154a acquires setting values from the setting value storage unit 120b and group-specific people flow data from the group-specific people flow storage unit 130a, and converts each group-specific people flow data for each class into function data based on the setting values.

Figure 7 shows the people flow data shown in Figure 1 plotted side-by-side for each variable. Figure 7 shows graphs of 16 people flow data items corresponding to 16 variables.

For example, suppose that preprocessing generates 16 variables x 274 items (4,384 items in total) of people flow data, as shown in Figure 1(b). In such a case, the clustering unit 153 may classify the people flow data into mesh groups corresponding to the 16 variables through clustering processing based on the relationship between the characteristics of the people flow data and the meshes. The function data conversion unit 154a converts each piece of people flow data into function data based on the plot shown in Figure 7.

(PCA learning unit 154b)
Returning to FIG. 6 , the PCA learning unit 154b extracts multiple people flow components by performing principal component analysis on the function data. For example, the PCA learning unit 154b calculates eigenvectors and eigenvalues and registers them in the parameter storage unit 130b. As such, the PCA learning unit 154b is a processing unit equivalent to an extraction unit. The PCA learning unit 154b also executes a learning process to maximize the variance of data in the function data. The PCA learning unit 154b also registers parameters that can maximize the variance in the parameter storage unit 130b. Note that the machine learning algorithm used by the PCA learning unit 154b searches for weights that optimize a loss function when learning the axes that can best express data from the training data.

(Inference Unit 155)
The inference unit 155 restores the principal component-scored data to the original data using eigenvectors. For example, the inference unit 155 acquires the principal component-scored data and eigenvectors from the parameter storage unit 130b. For example, if the editing unit 156 modifies the eigenvectors, the inference unit 155 calculates the principal component scores after replacing the acquired eigenvectors with the modified eigenvectors. Note that if the editing unit 156 modifies the principal component scores, the inference unit 155 replaces the calculated principal component scores with the modified principal component scores. The inference unit 155 then restores the original data based on the current principal component scores and eigenvectors. Furthermore, the inference unit 155 generates people flow data for each principal component using multiple people flow components extracted by principal component analysis of the function data. For this reason, the inference unit 155 is a processing unit equivalent to an extraction unit. Note that the inference unit 155 may calculate principal component loadings based on the eigenvalues and eigenvectors calculated by the principal component score calculation unit 155a.

(Principal component score calculation unit 155a)
The principal component score calculation unit 155a calculates principal component scores based on the people flow data. For example, the principal component score calculation unit 155a calculates principal component scores for each piece of function data (16 pieces of function data in the example of FIG. 7 ) in which people flow data belonging to each mesh group is transformed into a function. The principal component scores are used to calculate principal component people flow data. For example, the principal component score calculation unit 155a generates a covariance matrix by calculating the covariance between variables included in the function data. Then, the principal component score calculation unit 155a calculates the eigenvalues and eigenvectors of the covariance matrix. For example, the principal component score calculation unit 155a can acquire parameters (e.g., weights and principal component vectors) from the parameter storage unit 130b and calculate principal component scores based on the acquired parameters. Calculating the principal component scores involves calculating positions on axes (principal component scores), which enables the generation of principal component people flow data tailored to the purpose.

(Principal component people flow conversion unit 155b)
The principal component people flow transform unit 155b creates new data (principal component scores) by transforming the data using eigenvectors. For example, the principal component people flow transform unit 155b creates new data (principal component scores) by projecting the original data onto the eigenvectors. This allows the principal component people flow transform unit 155b to transform the features of the original people flow data, which has multiple variables, into a two-dimensional graph.

The principal component people flow transformation unit 155b also registers the generated people flow data for each principal component in the principal component people flow storage unit 140a. For example, the original people flow data can be obtained by adding together all of the principal component people flow data.

Figure 8 shows an example of principal component people flow data. Figure 8 shows the scene in which the 274 days' worth of people flow data DA11 described in Figure 1 has been converted into function data, and the four principal components, the first through fourth, have been extracted, and each principal component has been converted into a two-dimensional graph. Specifically, Figure 8(a) shows a two-dimensional graph corresponding to the first principal component, and Figure 8(b) shows a two-dimensional graph corresponding to the second principal component. Furthermore, Figure 8(c) shows a two-dimensional graph corresponding to the third principal component, and Figure 8(d) shows a two-dimensional graph corresponding to the fourth principal component.

Based on the principal component loadings, the user can infer, for example, how much of each variable is used in the calculation of each principal component. For example, the first principal component shown in Figure 8(a) shows a sudden population drop in "period T1" from July 1, 2020 to March 31, 2021, and also shows a significant population decline in "period T2" from July 1, 2020 to March 31, 2021. In this example, based on the principal component loadings of the first principal component, the user can determine that the first principal component is a principal component that strongly reflects the variables corresponding to "period T1" and "period T2." Based on this determination, the user can then consider, for example, how to edit the principal component people flow data corresponding to the first principal component.

(Editing unit 156)
The editing unit 156 controls a predetermined people flow component among the multiple people flow components extracted by the inference unit 155 in accordance with the user's purpose (editing operation).

(Score change unit 156a)
When the user inputs information for changing the principal component score corresponding to a predetermined people flow component (principal component) among the plurality of people flow components, the score change unit 156a generates information for changing the principal component score based on the input information. For example, when the user inputs information for changing the principal component score corresponding to the first principal component, the score change unit 156a generates information for changing the current principal component score of the first principal component in accordance with the input information.

The score modification unit 156a also transmits the generated information to the inference unit 155. As a result, for example, the principal component people flow transformation unit 155b generates virtual people flow data based on the modified principal component scores while maintaining the characteristics of the people flow data indicated by a specified people flow component. For example, the principal component people flow transformation unit 155b generates virtual people flow data based on the modified principal component scores corresponding to the first principal component while maintaining the characteristics of the people flow data indicated by the first principal component. The principal component people flow transformation unit 155b registers the virtual people flow data in the principal component people flow storage unit 140a.

For example, suppose that 16 variables x 274 items (4,384 items in total) of people flow data are generated, and the clustering unit 153 classifies the people flow data into 16 types of mesh groups. Then, suppose that the conversion process performed by the function data conversion unit 154a results in function data i (i = 1, ..., 16) as shown in Figure 7.

In this situation, if the user wishes to edit the jth principal component score (j = 1, ..., p), which is the principal component score of the jth principal component of function data i, according to their own expectations, they input information indicating the edit content to the information processing device 100. In this case, the principal component people flow conversion unit 155b generates principal component people flow data ij based on the jth principal component score specified by the user among the principal components of function data i specified by the user and the kth principal component score (j ≠ k) of other function data i. For example, since the contribution of the people flow component whose principal component score has been changed to function data i changes while the contribution of the other flow components remains unchanged, the people flow of function data i changes by the amount of change in the people flow in the people flow component whose principal component score has been changed. In other words, by changing the jth principal component score of function data i, new function data i is generated that is similar to function data i but differs by the contribution of the jth principal component score.

(Load amount change unit 156b)
When the user inputs information for changing the principal component loadings corresponding to a specific one of the plurality of people flow components (principal components), the loading amount changing unit 156b generates information for changing the principal component loadings based on the input information. For example, when the user inputs information for changing the principal component loadings corresponding to the first principal component, the loading amount changing unit 156b generates information for changing the current principal component loadings of the first principal component in accordance with the input information.

The loading amount modification unit 156b also transmits the generated information to the inference unit 155. As a result, for example, the principal component people flow transformation unit 155b modifies the features of the people flow data indicated by a specified people flow component based on the modified principal component loading amount. For example, the principal component people flow transformation unit 155b modifies the features of the people flow data indicated by the first principal component based on the modified principal component loading amount. The principal component people flow transformation unit 155b registers the people flow data with the modified features in the principal component people flow storage unit 140a.

(Generation unit 157)
The generation unit 157 generates people flow data to be provided to a user based on a plurality of people flow components including the modified people flow component.

(Principal component crowd flow correction unit 157a)
When the user inputs modification information that modifies the shape of a graph indicated by a predetermined people flow component (principal component) among the plurality of people flow components, the principal component people flow correction unit 157a corrects the characteristics of the data indicated by the predetermined people flow component among the plurality of people flow components based on the modification information. This point will be explained using the example of Figure 4(b).

For example, suppose a user wishes to delete the feature of population decline in the "T1 period" and the feature of the trend of population decline in the "T2 period" from the first principal component that expresses these features. In such a case, the user can refer to the graph (first principal component people flow data) indicated by the first principal component extracted from people flow data DA11, and perform an operation on the information processing device 100 to delete the features of the "T1 period" and the "T2 period" from the features of the first principal component.

When change information that changes part of the first principal component people flow data is input, the principal component people flow correction unit 157a performs a correction process to delete characteristic parts of the features included in the first principal component people flow data that correspond to the input change information. For example, the principal component people flow correction unit 157a deletes characteristic parts of the features included in the first principal component people flow data that correspond to the input change information, and performs a process to compensate for the deleted characteristic parts using linear interpolation or curve interpolation.

(Reconstruction unit 157b)
The reconstructing unit 157b reconstructs the people flow data to be provided to the user based on the plurality of people flow components including the principal components after the feature control. For example, when the principal component people flow correcting unit 157a corrects some of the features included in the first principal component people flow data, the reconstructing unit 157b adds up the people flow data of the first principal component after the feature correction and the people flow data of the other principal components (e.g., the second principal component to the fourth principal component) to reconstruct the people flow data to be provided to the user.

Of the processing units described so far, the editing unit 156 and generation unit 157 are processing units that correspond to the people flow control unit.

7. Example of operation procedure of information processing device
Next, an example of the operation procedure of the information processing device 100 will be described with reference to Fig. 9. Fig. 9 is a flowchart showing the procedure of information processing according to the embodiment.

When the preprocessing unit 151 acquires raw people flow data (time-series number of people data), it performs preprocessing by dividing and reorganizing the acquired data using statistical methods (e.g., kshape) or qualitative criteria (e.g., a specified range such as a mesh) (step S901).

The function data conversion unit 154a converts the preprocessed people flow data into function data using basis functions. For example, the function data conversion unit 154a calculates the number of bases m for the function data based on the initial value for the function data conversion (step S902).

Here, among the preprocessed people flow data, n pieces of t-period time series data Y are taken as the realizations of the function u(t) expressed in m bases (step S903). In this case, the relationship x_i(t) = u_i(t) + ε (ε is a constant) holds between the i-th data x_i(t) of variable x among the n pieces of t-period time series data Y and the i-th data u_i(t) of variable u among the n pieces of t-period time series data Y.

There are n samples of sets of real data Y represented by m variables, and the i-th data Y is in the form of a matrix (n x 1) called data SM, and the (n x m) weighting coefficients are called transformation matrix W. Transformation matrix W can also be written as coefficient matrix W or weighting matrix W.

The PCA learning unit 154b obtains a threshold for the cumulative contribution rate and calculates the number p of principal components from the sample data SM whose cumulative contribution rate is equal to or greater than the threshold (step S904).

Then, the PCA learning unit 154b finds p (p<m) principal components from the transformation matrix W (step S905). Specifically, the PCA learning unit 154b finds, through unsupervised learning, a transformation matrix W that maximizes the variance of the data in the data SM. For example, the PCA learning unit 154b can find the transformation matrix W as the eigenvector B corresponding to the largest eigenvalue. The PCA learning unit 154b finds, through unsupervised learning, the weight W that maximizes the variance of the inner product of the function data X and the weight W. For example, the PCA learning unit 154b can find the weight W as an eigenvalue problem of the covariance function for the function data X. The eigenvector B can be expressed as a (p x m) matrix. Furthermore, the principal component score calculation unit 155a can calculate the principal component score S, which is expressed as an (n x p) matrix.

In this state, the inference unit 155 determines whether to change the inference input (step S906). For example, the inference unit 155 can determine to change the inference input when the score change unit 156a receives user information that changes the principal component scores, or when the loading change unit 156b receives user information that changes the principal component loadings. On the other hand, the inference unit 155 may determine not to change the inference input when such user information is not received.

If it is determined that the inference input should not be changed (step S906; Yes), the principal component people flow transformation unit 155b reconstructs the graph indicated by each principal component, i.e., the function D(q), based on the people flow data for each principal component (step S907).

On the other hand, if it is determined that the inference input should be changed (step S906; No), the principal component people flow transformation unit 155b controls the features of the corresponding principal component people flow data based on the input user information (step S908).

For example, when user information that modifies the principal component scores is received, the principal component people flow transform unit 155b generates virtual people flow data based on the principal component scores modified in accordance with the user information. For example, the principal component people flow transform unit 155b generates virtual people flow data based on the modified principal component scores corresponding to the principal components specified in the user information, while retaining the characteristics of the people flow data indicated by the principal components.

As another example, when user information that modifies the principal component load is received, the principal component people flow transformation unit 155b modifies the characteristics of the people flow data indicated by the principal component specified in the user information based on the principal component load modified in accordance with the user information.

Also, if processing proceeds to step S908, it proceeds to step S907. Specifically, the principal component people flow transform unit 155b reconstructs principal component people flow data controlled in accordance with user information (user's purpose) and uncontrolled principal component people flow data that is not subject to the user information (outside the user's purpose).

The generation unit 157 then determines whether to correct the shape of the graph indicated by a specific principal component among the multiple people flow components (principal components) in accordance with the user information (step S909). For example, the generation unit 157 can determine to perform correction when modification information that modifies the shape of the graph indicated by a specific principal component among the multiple people flow components is input. More specifically, the generation unit 157 modifies the shape of the graph of the principal component people flow data that the user has specified to be modified, among the principal component people flow data obtained by decomposing the original people flow data into each principal component. On the other hand, the generation unit 157 may determine not to perform correction when modification information that modifies the shape of the graph indicated by a specific principal component among the multiple people flow components is not input.

If it is determined that no correction is required (step S909; Yes), the reconstruction unit 157b sums up the uncorrected people flow data for each principal component to reconstruct one piece of people flow data (step S910).

On the other hand, if it is determined that a correction is to be made (step S909; No), the principal component people flow correction unit 157a determines, based on the user information, whether the correction will not leave the shape of the graph represented by one of the multiple principal components (e.g., the first principal component) (i.e., whether the correction will be for the entire graph or only a part of the graph) (step S911). The principal component people flow correction unit 157a directly edits the principal component people flow data (in people) generated based on the principal component functions and principal component scores. Therefore, the principal component people flow data generated from other principal component functions remains unchanged, and as a result, other features remain.

For example, if the principal component people flow correction unit 157a determines that the correction will not retain the shape of the graph represented by the first principal component (step S911; Yes), it performs a correction process to delete characteristic portions of the features included in the first principal component people flow data that correspond to the input change information. For example, the principal component people flow correction unit 157a deletes characteristic portions of the features included in the first principal component people flow data that correspond to the input user information, and performs a process to compensate for the deleted characteristic portions using linear interpolation or curve interpolation (step S912).

On the other hand, if the principal component people flow correction unit 157a determines that the correction should preserve the shape of the graph (step S911; No), it expands the control points (e.g., holidays, quantiles, etc.) to the same periods as the control points on the principal component people flow data, and expands periods other than the control points by determining the expansion rate using linear interpolation from the control point rate, etc. (step S913).

Also, if processing proceeds to step S912 or S913, it proceeds to step S910. Specifically, the reconstructor 157b reconstructs one piece of people flow data by adding together the principal component people flow data corrected by linear interpolation or the like and the uncorrected principal component people flow data (step S910).

Up to this point, we have explained an example of the operating procedure of the information processing device 100. According to the example in Figure 9, steps S901 to S903 are preprocessing of people flow data. Furthermore, steps S904 to S905 are principal component analysis of the people flow data. Furthermore, steps S906 to S908 are processing for editing the features of people flow data in response to input from the user (e.g., administrator). Meanwhile, steps S909 to S913 are processing for editing the features of people flow data in response to input from the user (e.g., service user).

Here, we will use the example in Figure 9 to show a specific example of a method for generating principal component people flow data ij. First, time series data is treated as a function and approximated with m functions g_1 to g_m. x_i(t) = u_i(t) + ε (ε is a constant). The estimated coefficients x_i_1 to x_i_m are compiled to define an n x m design matrix D. When design matrix D is subjected to principal component analysis, the eigenvector B = (b_1, ..., b_p) is obtained.

The jth principal component function is obtained by taking the dot product of eigenvector b_j and function vector (g_1, g_2, ..., g_m)'. Therefore, m principal component functions (h_1, h_2, ..., h_m) are obtained by taking the transpose of eigenvector B and the dot product of the function vector. Furthermore, the principal component score vector (s_i_1, ..., s_i_m) is obtained by taking the dot product of coefficient vector (x_i_1, ..., x_i_m) and eigenvector B.

Time series_i can be expressed as a linear combination of m principal component functions (h_1, h_2, ..., h_m), and the coefficients in this case are the principal component scores. Specifically, in Time series_i = s_i_1 x h_1 + ... + s_i_m x h_m + ξ, s_i_j x h_j is a function that is simply a constant multiplication of the function h_j, and is not time series data. Substituting x = 0, ..., 273 returns it to time series data, which becomes the principal component people flow data ij.

8. Hardware Configuration
The information processing device 100 according to the above embodiment is realized by a computer 1000 having a configuration as shown in Fig. 10. Fig. 10 is a hardware configuration diagram showing an example of the computer 1000 that realizes the functions of the information processing device 100. The computer 1000 has a CPU 1100, a RAM 1200, a ROM 1300, a HDD 1400, a communication interface (I/F) 1500, an input/output interface (I/F) 1600, and a media interface (I/F) 1700.

The CPU 1100 operates based on programs stored in the ROM 1300 or HDD 1400, and controls each component. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 starts up, as well as programs that depend on the computer 1000's hardware.

HDD 1400 stores programs executed by CPU 1100 and data used by such programs. Communication interface 1500 receives data from other devices via communication network 50 and sends it to CPU 1100, and transmits data generated by CPU 1100 to other devices via communication network 50.

The CPU 1100 controls output devices such as displays and printers, and input devices such as keyboards and mice, via the input/output interface 1600. The CPU 1100 acquires data from input devices via the input/output interface 1600. The CPU 1100 also outputs generated data to output devices via the input/output interface 1600.

Media interface 1700 reads programs or data stored on recording medium 1800 and provides them to CPU 1100 via RAM 1200. CPU 1100 loads the programs from recording medium 1800 onto RAM 1200 via media interface 1700 and executes the loaded programs. Recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase Change Rewritable Disc), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

For example, when the computer 1000 functions as the information processing device 100 according to the embodiment, the CPU 1100 of the computer 1000 executes a program (e.g., an information processing program according to the embodiment) loaded onto the RAM 1200, thereby realizing the functions of each processing unit. In addition, the HDD 1400 stores data in the storage unit. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, the CPU 1100 may obtain these programs from another device via the communication network 50.

[9. Other]
Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

10. Summary
The above describes in detail the embodiments of the present application based on several drawings, but these are merely examples, and the present invention can be implemented in other forms that include the embodiments described in the Disclosure of the Invention section and that have been modified and improved in various ways based on the knowledge of those skilled in the art.

REFERENCE SIGNS LIST 1 System 10 User device 100 Information processing device 111 External interface unit 112 Management interface unit 113 Input/output interface unit 120 Input storage unit 130 Memory unit 140 Output storage unit 151 Preprocessing unit 152 Operation unit 153 Clustering unit 154 Learning unit 155 Inference unit 156 Editing unit 157 Generation unit

Claims (10)

a conversion unit that converts time-series people flow data of a population within a predetermined range into function data;
an extraction unit that extracts a plurality of people flow components by performing principal component analysis on the function data;
A people flow control unit that controls a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
a generation unit that generates people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Equipped with
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion unit converts each of the people flow data that exists according to the number of variables into function data,
the extraction unit calculates, for each of the people flow components, principal component scores according to the plurality of variables when extracting the plurality of people flow components, and principal component loadings indicating relationships with each of the plurality of variables;
When the user inputs information to change the principal component score corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control unit controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component score.
Information processing device. a conversion unit that converts time-series people flow data of a population within a predetermined range into function data;
an extraction unit that extracts a plurality of people flow components by performing principal component analysis on the function data;
A people flow control unit that controls a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
a generation unit that generates people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Equipped with
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion unit converts each of the people flow data that exists according to the number of variables into function data,
the extraction unit calculates, for each of the people flow components, principal component scores according to the plurality of variables when extracting the plurality of people flow components, and principal component loadings indicating relationships with each of the plurality of variables;
When the user inputs information for changing the principal component loading amount corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control unit controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component loading amount.
Information processing device. a conversion unit that converts time-series people flow data of a population within a predetermined range into function data;
an extraction unit that extracts a plurality of people flow components by performing principal component analysis on the function data;
A people flow control unit that controls a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
A generation unit that generates people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Equipped with
The conversion unit generates a graph indicated by each of the plurality of people flow components,
When the user inputs change information for changing the shape of a graph indicated by a predetermined people flow component in the graph, the people flow control unit controls the characteristics of data indicated by the predetermined people flow component based on the change information.
Information processing device. The information processing device according to claim 1 , wherein the generation unit generates people flow data to be provided to the user based on the plurality of people flow components including the predetermined people flow component after the characteristics have been controlled. An information processing method executed by an information processing device,
a conversion step of converting time-series people flow data of a population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control step of controlling a predetermined people flow component among the plurality of people flow components so that the predetermined people flow component is changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Including,
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion step converts each of the people flow data that exists according to the number of variables into function data,
The extraction step calculates, for each of the people flow components, a principal component score corresponding to the plurality of variables when extracting the plurality of people flow components, and a principal component loading indicating a relationship with each of the plurality of variables;
When the user inputs information for changing the principal component score corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control step controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component score.
Information processing methods. An information processing method executed by an information processing device,
a conversion step of converting time-series people flow data of a population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control step of controlling a predetermined people flow component among the plurality of people flow components so that the predetermined people flow component is changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Including,
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion step converts each of the people flow data that exists according to the number of variables into function data,
The extraction step calculates, for each of the people flow components, a principal component score corresponding to the plurality of variables when extracting the plurality of people flow components, and a principal component loading indicating a relationship with each of the plurality of variables;
When the user inputs information for changing the principal component loadings corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control step controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component loadings.
Information processing methods. An information processing method executed by an information processing device,
a conversion step of converting time-series people flow data of a population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control step of controlling a predetermined people flow component among the plurality of people flow components so that the predetermined people flow component is changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
Including,
The converting step generates a graph indicated by each of the plurality of people flow components,
The people flow control step includes, when the user inputs change information for changing the shape of a graph indicated by a predetermined people flow component among the graphs, controlling characteristics of data indicated by the predetermined people flow component based on the change information.
Information processing methods. A conversion procedure for converting time-series people flow data of population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control procedure for controlling a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
on the computer ,
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion procedure converts each of the people flow data that exists according to the number of variables into function data;
The extraction procedure calculates, for each of the people flow components, principal component scores corresponding to the plurality of variables when extracting the plurality of people flow components, and principal component loadings indicating relationships with each of the plurality of variables;
When the user inputs information to change the principal component score corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control procedure controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component score.
Information processing program. A conversion procedure for converting time-series people flow data of population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control procedure for controlling a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
on the computer,
The people flow data is multidimensional data including a plurality of variables regarding observation of population for a predetermined period for each of the predetermined ranges,
The conversion procedure converts each of the people flow data that exists according to the number of variables into function data;
The extraction procedure includes calculating, for each of the people flow components, principal component scores corresponding to the plurality of variables when extracting the plurality of people flow components, and principal component loadings indicating relationships with each of the plurality of variables;
When the user inputs information for changing the principal component loading corresponding to a predetermined people flow component among the plurality of people flow components, the people flow control procedure controls the characteristics of data indicated by the predetermined people flow component based on the changed principal component loading.
Information processing program. A conversion procedure for converting time-series people flow data of population within a predetermined range into function data;
an extraction step of extracting a plurality of people flow components by performing principal component analysis on the function data;
a people flow control procedure for controlling a predetermined people flow component among the plurality of people flow components to be changed according to a user's purpose;
a generation step of generating people flow data to be provided to the user based on the plurality of people flow components including the changed people flow component;
on the computer,
The conversion step includes generating, for each of the plurality of people flow components, a graph indicated by the people flow component;
The people flow control procedure includes, when the user inputs change information for changing the shape of a graph indicated by a predetermined people flow component among the graphs, controlling characteristics of data indicated by the predetermined people flow component based on the change information.
Information processing program.