JP2020024542A - Data analysis device and data analysis method - Google Patents

Abstract

An object of the present invention is to automatically obtain a feature amount from a large amount of data and to display a feature amount having a high degree of importance, so that it is possible to assist in obtaining new knowledge.
A data analysis device can generate a diagram showing a relationship between a first feature amount selected from feature amounts having higher importance levels in accordance with a user input and an objective variable. . The data analysis device calculates a feature amount that contributes to the prediction of the objective variable in a cluster selected from a plurality of clusters shown in the basic diagram, and a feature amount that contributes to the prediction of the objective variable in all clusters or clusters not selected. A comparison diagram showing the comparison with can be generated.
[Selection diagram] FIG.

Description

  The present invention relates to a data analysis device and a data analysis method.

  2. Description of the Related Art Conventionally, so-called data mining, which attempts to obtain useful knowledge that has been unknown so far from a large amount of information, has been known. In a general data mining method, steps of a preprocessing step, a feature extraction step, a model learning step, and a post-processing step are sequentially executed. In the preprocessing step, data necessary for the analysis is collected, and in order to enhance the effect of the analysis, work such as removing unnecessary noise from the data and filling in missing items is performed. This operation is called ETL: Extract / Transform / Load. In the feature extraction step, a feature amount existing in the original data is processed, and a new feature amount that is more useful for analysis is generated. In the model learning step, the data prepared in the preprocessing step and the feature extraction step are input to a machine learning algorithm to obtain an analysis result. In the post-processing step, processing such as outputting the analysis result obtained in the model learning step to an external device or creating a report based on the analysis result is performed.

  By the way, the design of the optimal feature amount requires trial and error by a skilled analyst, which causes a long time for data analysis. On the other hand, for example, as disclosed in Non-Patent Document 1, a technique for automatically generating a new feature amount from data is known. In the techniques disclosed in these documents, a series of predefined operators (+, −, ×, ÷, etc.) are comprehensively applied to an original feature amount, and a large amount of new feature amounts are obtained. Can be generated automatically.

"A Randomized Exhaustive Propositionalization Approach for Molecule Classification" May 26, 2010

  However, according to the technique of the above-mentioned document, the number of new features obtained as a result of a combination of a plurality of operators is enormous. Since the huge number of feature values include a large amount of feature values that are not effective for analysis, there is a problem that it takes time for the subsequent model learning step.

  In addition, even if a new feature that is useful for improving the accuracy of analysis is generated, the relationship between the generated feature and the target variable and the relationship with other features are not intuitively understood. In order to obtain knowledge from analysis results, additional analysis work was required for each feature amount. That is, even if the technique described in the above-mentioned document is used, the manual analysis by the analyst is eventually required in order to obtain useful knowledge, and the problem that data analysis takes a long time remains.

  The present invention has been made in view of such a point, and an object of the present invention is to obtain a feature amount from a large amount of data, to display a feature amount having a high importance, and to assist in obtaining new knowledge. Is to be able to do it.

  In order to achieve the above object, a first invention is a data analysis device for analyzing data to be analyzed, a data input unit for receiving an input of data to be analyzed including a plurality of feature amounts and a target variable; A prediction model generation unit that generates a prediction model for predicting the objective variable from a feature value; an importance calculation unit that calculates importance for prediction by the prediction model for each of the plurality of feature values; A display unit that displays a feature amount having a higher importance based on the importance calculated by the degree calculation unit, and a feature selected in response to a user input from the feature amounts displayed on the display unit. A basic diagram generation unit that divides the analysis target data into a plurality of clusters based on the value of the amount, and generates a basic diagram showing a relationship between each cluster and a representative value of an objective variable of each cluster. A cluster selecting unit that receives selection of any one of the clusters from the plurality of clusters illustrated in the basic diagram; a feature amount that contributes to the prediction of the objective variable in the cluster selected by the cluster selecting unit; In a cluster or a non-selected cluster not selected by the cluster selection unit, a comparison diagram generation unit that generates a comparison diagram showing a comparison with a feature amount contributing to the prediction of the objective variable, the display unit includes: The basic diagram generated by the basic diagram generation unit and the comparison diagram generated by the comparison diagram generation unit can be displayed.

  According to this configuration, when the analysis target data is input, a prediction model for predicting a target variable from a plurality of feature amounts included in the analysis target data is generated. The feature amount may be included in the analysis target data or may be newly generated. Then, the importance for the prediction by the prediction model is calculated for each of the plurality of feature amounts, and among the calculated plurality of feature amounts, the feature amount having the higher importance is displayed on the display unit. At this time, a feature value having lower importance may be displayed. In any case, by viewing the display unit, the user can distinguish and grasp a feature amount having a high importance from a feature amount having a low importance. When the user selects a desired feature amount from the feature amounts displayed on the display unit, the analysis target data is divided into a plurality of clusters based on the value of the selected feature amount.

  In addition, a basic diagram showing the relationship between each cluster and the representative value of the objective variable of each cluster is generated, and when an arbitrary one of the clusters shown in the basic diagram is selected, the selection is made. A comparison diagram is generated that shows a comparison between the feature amount that contributes to the prediction of the objective variable in the cluster that has been made and the feature amount that contributes to the prediction of the objective variable in all clusters or unselected cluster stars. Then, the basic diagram and the comparison diagram are displayed on the display unit.

  By looking at the basic diagram, the user can know the relationship between the clusters obtained by dividing the data to be analyzed and the representative values of the objective variables, and can obtain new knowledge based on this. it can.

  In addition, when there is a cluster that the user wants to pay attention to, when the user selects the cluster, not only the feature amount contributing to the prediction of the objective variable in the cluster but also the feature amount contributing to the prediction of the objective variable in all the clusters is obtained. Since comparison and comparison with a feature value that contributes to the prediction of the objective variable in the non-selected class star can be performed, new knowledge can be obtained based on the comparison.

  The basic diagram and the comparison diagram may be simultaneously displayed on the display unit, or may be displayed at different timings. Further, the “figure” includes a graph, a table, and characters.

  A second invention is characterized in that the display unit is configured to display a plurality of feature values having higher importance levels in order of higher importance.

  According to this configuration, a plurality of feature values having higher importance are displayed on the display unit in a ranking format, so that the user can relatively compare the importance of the plurality of feature values. The direction in which the degrees of importance are arranged in descending order may be, for example, a vertical direction or a horizontal direction as viewed from the user.

  A third invention is characterized in that the display unit is configured to simultaneously display a feature value having the highest importance and a plurality of feature values less than the highest importance.

  According to this configuration, it is possible to easily grasp which feature amount has the highest importance, and a difference between the highest importance and the next highest importance.

  A fourth invention is characterized in that the comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount whose contribution degree is significantly high only in the cluster selected by the cluster selection unit. I do.

  According to this configuration, a feature amount highly relevant to the selected cluster is automatically indicated, so that new knowledge can be obtained based on a plurality of feature amounts.

  According to a fifth aspect, the comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount not selected by a user among the feature amounts having the higher importance. Features.

  According to this configuration, although not selected by the user, it is conceivable that the feature having the higher importance has a large effect on the objective variable, so the feature having the higher importance is also used as a comparison diagram. By doing so, new knowledge can be obtained based on a plurality of feature amounts.

  In a sixth aspect, the comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount contributing to the prediction of the objective variable in a missing cluster in which the value of the analysis target data does not exist. It is characterized by having.

  According to this configuration, when the analysis target data group having no value is set as a missing cluster, the missing cluster may also contribute to the prediction of the objective variable. New knowledge can be obtained by indicating the feature values that contribute.

  A seventh invention is a data analysis method for analyzing data to be analyzed, a data input step of receiving input of data to be analyzed including a plurality of feature amounts and an objective variable, and predicting the objective variable from the plurality of feature amounts. Predictive model generating step of generating a predictive model for calculating the importance of each of the plurality of feature amounts with respect to prediction by the predictive model; and calculating the importance calculated by the importance calculating step. A feature value displaying step of displaying a feature value having a higher importance based on the degree, and a feature value selected according to a user input from the feature values displayed in the feature value displaying step. The analysis target data is divided into a plurality of clusters on the basis of the above, and a basic diagram showing a relationship between each cluster and a representative value of an objective variable of each cluster is shown. Generating a basic diagram to be formed, a cluster selecting step of receiving selection of any one cluster from a plurality of clusters shown in the basic diagram, and predicting the objective variable in the cluster selected in the cluster selecting step. And a comparison diagram generation step of generating a comparison diagram showing a comparison between feature amounts contributing to the prediction of the objective variable in all clusters or unselected clusters not selected by the cluster selection step. And a comparison diagram display step capable of displaying the basic diagram generated in the basic diagram generation step and the comparison diagram generated in the comparison diagram generation step.

  ADVANTAGE OF THE INVENTION According to this invention, the feature value with high importance contained in a large amount of data can be displayed on a display part. In addition, it is possible to know the relationship between the clusters obtained by dividing the data to be analyzed and the representative value of the objective variable, and the feature amount contributing to the prediction of the objective variable in the selected cluster, and all clusters or unselected clusters. Can be compared with the feature amount that contributes to the prediction of the objective variable in the cluster, so that it is possible to help to obtain new knowledge.

It is a figure showing the schematic structure of the data analysis device concerning the embodiment of the present invention. It is a block diagram of a data analyzer. 9 is a flowchart illustrating a procedure of data analysis. 6 is a flowchart illustrating an operation of the data analyzer. FIG. 3 is a diagram illustrating a data input user interface. FIG. 4 is a diagram showing analysis target data stored in two files in a table format. It is a figure showing a data flow display user interface. It is a figure showing the data flow display user interface which performed the preview display of the analysis object data. It is a figure showing a data processing user interface. It is a figure showing the data flow display user interface which displayed the data after processing with an icon. FIG. 4 is a diagram illustrating a feature amount automatic generation user interface. FIG. 7 is a diagram illustrating a method of calculating a ratio in which there is a correspondence between two pieces of analysis target data. It is a figure explaining the case where the total of the price column is calculated and added to "master". FIG. 11 is a diagram illustrating a case where a row to be aggregated is determined based on a combination with a row indicating a date. It is a figure showing a feature-quantity display user interface. It is a figure which shows the feature-value display user interface at the time of selecting "total number of visits in the last 30 days". FIG. 11 is a diagram illustrating a feature amount display user interface when “compare average values” is selected from a detailed menu. FIG. 11 is a diagram illustrating a feature amount display user interface on which a comparison table is displayed. FIG. 11 is a diagram illustrating a feature amount display user interface on which a prediction accuracy comparison graph is displayed. It is a figure which shows the feature-value display user interface when "the number of days passed since the last store visit" is selected. It is a figure showing the feature-quantity display user interface on which the influence comparison graph was displayed. It is a figure showing the feature-value display user interface at the time of selecting “comparison of feature-value contribution” from the detailed menu. FIG. 11 is a diagram illustrating a feature amount display user interface on which a contribution degree display graph is displayed. It is a figure explaining the calculation method of a contribution. It is a figure which shows the feature-value display user interface displayed when "sex" of a feature-value is selected. It is a figure showing the amount-of-features display user interface on which the explanation graph of the difference was displayed. FIG. 11 is a diagram illustrating a feature amount display user interface on which a graph related to a change in the contribution of the feature amount is displayed.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. It should be noted that the following description of the preferred embodiments is merely an example in nature, and is not intended to limit the present invention, its application, or its use.

FIG. 1 is a diagram showing a schematic configuration of a data analysis device 1 according to an embodiment of the present invention, and FIG. 2 is a block diagram of the data analysis device 1. The data analysis device 1 is a device for analyzing data to be analyzed. By using the data analysis device 1, the data analysis method according to the present invention can be executed. The data analysis device 1 and the data analysis method can be used, for example, when performing so-called data mining, which attempts to obtain useful knowledge that has been unknown so far from a large amount of information.
(Data analysis procedure)
A general data analysis procedure will be described based on the flowchart shown in FIG. After the start, in step SA1, a preprocessing step is executed. In the pre-processing step, data necessary for the analysis, that is, data to be analyzed is collected, and in order to enhance the effect of the analysis, an operation of removing unnecessary noise from the data and filling in missing items is performed. This operation is called ETL: Extract / Transform / Load. In step SA2, a feature extraction step is performed. In the feature extraction step, a feature amount existing in the original data is processed to generate a new feature amount that is more useful for analysis. At step SA3, a model learning step is executed. In the model learning step, the data prepared in the preprocessing step and the feature extraction step are input to a machine learning algorithm to obtain an analysis result. At step SA4, a post-processing step is executed. In the post-processing step, processing is performed such as outputting the analysis result obtained in the model learning step to an external device or creating a report based on the analysis result.

(Overall configuration of data analyzer 1)
As shown in FIGS. 1 and 2, the data analysis device 1 includes a device main body 2, a monitor 3, a keyboard 4, and a mouse 5. It is connected to the. For example, a data analysis device 1 can be obtained by installing a program for executing control contents to be described later in a general-purpose personal computer, and the data analysis device 1 is configured by dedicated hardware on which the program is installed. You can also. The device body 2 and the monitor 3 may be integrated, or a part of the device body 2 may be built in the monitor 3.

  The data analyzer 1 has a built-in communication module (not shown) and is configured to be able to communicate with the outside. Thus, data can be downloaded from an external server via the Internet line.

  In addition, the keyboard 4 and the mouse 5 are operation means for operating the data analysis device 1, as well as input means for inputting various information, selection means for performing a selection operation, and the like. In addition to or in place of the keyboard 4 and the mouse 5, a touch panel type input device, a voice input device, a pen type input device, or the like can be used.

(Configuration of monitor 3)
The monitor 3 shown in FIG. 1 is composed of, for example, an organic EL display, a liquid crystal display, or the like, and can be independently referred to as a display unit. The monitor 3 and the display control unit 3a shown in FIG. It can also be called a display unit. The display control unit 3a may be built in the monitor 3 or may be built in the device body 10. The display control unit 3a includes a display DSP for displaying an image on the monitor 3, and the like. The display control unit 3a may include a video memory such as a VRAM for temporarily storing image data when displaying an image. The display control unit 3a transmits a control signal for displaying a predetermined image on the monitor 3 based on a display command (display command) sent from a CPU 11a of the main control unit 11, which will be described later. For example, a control signal for causing the monitor 3 to display various user interfaces, icons, user's operation contents using the keyboard 4 and the mouse 5 is also transmitted. In addition, a pointer and the like that can be operated with the mouse 5 can be displayed on the monitor 3.

  When the monitor 3 is a touch operation panel type monitor, the monitor 3 can have a function of inputting various information.

(Overall configuration of device main body 2)
The device main body 2 shown in FIG. 1 includes a control unit 10 and a storage unit 30. The storage unit 30 includes a hard disk drive, a solid state drive (SSD), and the like. The storage unit 30 is connected to the control unit 10 and is controlled by the control unit 10 so that various data can be stored and the stored data can be read.

(Control unit 10)
Although not specifically shown, the control unit 10 can be configured by an MPU, a system LSI, a DSP, dedicated hardware, or the like. The control unit 10 has various functions as described later, but these may be realized by a logic circuit or may be realized by executing software.

  As shown in FIG. 2, the control unit 10 includes a main control unit 11, a data input unit 12, a prediction model generation unit 13, an importance calculation unit 14, a cluster selection unit 15, a display diagram generation unit 16, , A description automatic generation unit 18. Although the respective units of the control unit 10 are described separately as described above, the same unit may be configured to execute a plurality of types of processing, or may be further subdivided and linked into one processing. May be executed.

  Each of the above-described hardware is connected to be capable of two-way communication or one-way communication via an electrical communication path (wiring) such as a bus.

  The main control unit 11 performs numerical calculations and information processing based on various programs, and controls each unit of the hardware. The main control unit 11 stores a CPU 11a that functions as a central processing unit, a work memory 11b such as a RAM that functions as a work area when the main control unit 11 executes various programs, and a startup program and an initialization program. And a program memory 11c such as a read-only ROM, a flash ROM, or an EEPROM.

  The data input unit 12 is a unit that receives input of analysis target data including a plurality of feature amounts and objective variables. The data input unit 12 displays the data input user interface 50 shown in FIG. The user's operation performed on 50 is accepted. The user's operation includes an operation of the keyboard 4 and an operation of the mouse 5 (including button click, drag and drop, rotation of a wheel, and the like).

  Here, the analysis target data is data including a plurality of feature amounts and objective variables. The plurality of feature amounts originally exist in the analysis target data (existing feature amounts), and the user has noticed. Instead, any data may be used as long as the data includes any one or more of data that is implicitly included in the analysis target data (potential feature value) and data that is newly generated (new feature value). A part of the analysis target data may be missing, and in that case, work to fill in the missing item may be performed as described later.

  The data input user interface 50 shown in FIG. 5 is provided with a file selection button 50a, a database selection button 50b, a URL designation button 50c, a database display area 50d, and a read start button 50e. The arrangement of each button can be set freely.

  For example, if the file storing the analysis target data is stored in the external storage device or the storage unit 30 and is on the desktop or in an open folder, the user drags the file to the database display area 50d. & Drop operation. Thereby, the file name storing the analysis target data is displayed in the database display area 50d. Thereafter, when the reading start button 50e is pressed, the file displayed in the database display area 50d is read and stored in a predetermined area of the storage unit 30.

  When the data to be analyzed is in the database, the user presses the database selection button 50b. When the database selection button 50b is pressed, a setting screen (not shown) for connecting to the database is displayed, prompting the user to input a table name and, if necessary, a password. After that, when the reading start button 50e is pressed, the data to be analyzed is read and stored in a predetermined area of the storage unit 30 in a predetermined file format, and the file name storing the data to be analyzed is displayed in the database display area 50d. You.

  When the data to be analyzed is on the Internet or a server, the user presses the URL designation button 50c. When the URL designation button 50c is pressed, a URL input screen (not shown) is displayed to prompt the user to enter a URL. Then, when the reading start button 50e is pressed, the data to be analyzed is downloaded, read and stored in a predetermined area of the storage unit 30 in a predetermined file format, and the file name storing the data to be analyzed is displayed in the database display area 50d. Will be displayed.

  The number of files storing the analysis target data may be only one, or may be plural. FIG. 5 shows a case where the analysis target data stored in the two files “transaction” and “master” are read. The file may be read by a method other than the three methods described above. The file format is the csv format, but may be another format.

  FIG. 6 shows analysis target data stored in “master” and “transaction”, respectively. In this example, “master” is the customer information, and the age and gender are recorded in association with the customer ID. In addition, in association with the customer ID, whether or not the customer is a customer who has left is “departure”. Recorded in the column. "Leaving" means that the customer visited the store once, but then stopped visiting the store. A customer recorded as “1” in the “Leave” column is a customer who has departed, and a customer recorded as “0” is a customer who has not departed.

  In “transaction”, individual purchase histories are recorded, and in the “customer ID” column, the correspondence between “master” and each row is linked. In “transaction”, a purchase date, a category, and a price are recorded in association with the customer ID. The data described above is an example prepared for explaining the present invention, and does not limit the scope of the present invention. The present invention can analyze various data to be analyzed other than the purchase history data.

  When the reading of the analysis target data is completed, the data flow display user interface 51 shown in FIG. 7 is displayed on the monitor 3 and the user operation performed on the data flow display user interface 51 is received. The data flow display user interface 51 includes a data flow display area 51a, a lower display area 51b, an automatic feature amount generation button 51c, a data processing button 51d, and a data output button 51e.

  In the data flow display area 51a, the read analysis target data is displayed as icons. In the lower display area 51b, a relationship diagram (generally called an ER diagram) between the read analysis target data is displayed. The data input unit 12 shown in FIG. 2 detects whether or not a common column exists in the plurality of read analysis target data. Whether or not a common column exists can be determined, for example, based on the column name. In this example, as shown in FIG. 6, since a column named “customer ID” exists in both “master” and “transaction”, the data input unit 12 detects this. Then, the data input unit 12 determines that the “customer ID” of “master” is related to the “customer ID” of “transaction”, and is displayed in the lower display area 51b as shown in FIG. In the relation diagram, the “customer ID” of “master” and the “customer ID” of “transaction” are connected by a line and displayed.

  The above process is a process for receiving an input of data to be analyzed, and is performed by the data input unit 12. The data input unit 12 may include an input device such as the keyboard 4 and the mouse 5. Further, the above processing corresponds to the data input step of step SB1 in the flowchart shown in FIG.

  In step SB2 following step SB1, the analysis target data input in the data input step is processed. First, data to be processed is selected from the analysis target data displayed in the data flow display area 51a of the data flow display user interface 51 shown in FIG. The data selection operation includes, for example, an operation of clicking an icon corresponding to data to be processed. When the icon is clicked, the relationship diagram in the lower display area 51b is deleted, and the selected analysis target data is displayed in the lower display area 51b as shown in FIG. At this time, only a part of the selected analysis target data can be preview-displayed in the lower display area 51b.

  When "transaction" is preview-displayed, the words "daily goods" and "daily goods A" are mixed in the "category" column. Since “daily goods” and “daily goods A” have different values from each other, they are treated as different categories at the time of model learning, which will be described later. Higher accuracy can be obtained by unifying the notation fluctuation of. This is called data processing.

  When performing data processing, a data processing button 51d of the data flow display user interface 51 shown in FIG. 8 is pressed. When the data processing button 51d is pressed, a data processing user interface 52 shown in FIG. 9 is displayed on the monitor 3, and an operation performed by the user on the data processing user interface 52 is received. In the data processing user interface 52, a processing target data display area 52a, a missing value removal / fill-in button 52b, a data replacement button 52c, a column deletion button 52d, a column addition button 52e, a formula input button 52f, a column An information display button 52g and an end button 52h are provided. In the processing target data display area 52a, the selected analysis target data is displayed in a table format. If the displayed table is large, it can be scrolled.

  The missing value removal / fill-in button 52b is a button operated when removing missing values from the analysis target data and filling in the analysis target data. By operating the missing value removal / fill-in button 52b, an arbitrary missing value can be removed or a missing item can be added. The data replacement button 52c is a button operated when replacing an existing value with another value. For example, by selecting the cell of “daily item A” in the “category” column, pressing the data replacement button 52c, and entering “daily item” as a replacement character (value), the “daily item” in the same column is displayed. The value "A" is automatically replaced with the value "daily delivery". The column delete button 52d is a button operated when deleting an arbitrary column. After selecting an arbitrary row, pressing the row delete button 52d deletes the selected row. The column addition button 52e is a button operated when adding an arbitrary column. After selecting an arbitrary column in the table, by pressing a column addition button 52e, a column is added before or after the selected column. The formula input button 52f is a button for inputting a formula into a cell. The value is calculated by the mathematical expression input by operating the mathematical expression input button 52f. The column information display button 52g is a button for adding predetermined information to an arbitrary column.

  The analysis target data can be processed by operating the missing value removal / fill-in button 52b, data replacement button 52c, column deletion button 52d, column addition button 52e, formula input button 52f, and column information display button 52g. Therefore, the data analyzer 1 can perform the ETL processing, and by performing the ETL processing, the data to be analyzed can be transformed into a form more suitable for analysis. Thereby, the accuracy of the analysis result can be further improved. When a series of operations is completed, an end button 52h of the data processing user interface 52 shown in FIG. 9 is pressed.

  When the end button 52h is pressed, the data flow display user interface 51 is displayed again as shown in FIG. A new icon "transaction-1" is displayed in the data flow display area 51a of the data flow display user interface 51. The “transaction-1” is the data to be analyzed after processing the “transaction”, and the “transaction-1” and the “transaction” are connected by a related line, and the relationship is graphically displayed. . Therefore, the user can intuitively understand that “transaction-1” is the processed data of “transaction” and that “transaction-1” is the data that has been processed. The above series of ETL processing is the data processing processing in step SB2 of the flowchart shown in FIG.

  When a series of ETL processing is completed, the process proceeds to a step of inputting a setting relating to feature amount generation in step SB3 of the flowchart shown in FIG. When a feature amount automatic generation button 51c of the data flow display user interface 51 shown in FIG. 10 is pressed, the feature amount automatic generation user interface 53 shown in FIG. 11 is displayed on the monitor 3 and is performed on the feature amount automatic generation user interface 53. Accepts user operations.

  The feature amount automatic generation user interface 53 is provided with a data flow display region 53a, a lower display region 53b, a relationship designation region 53c, a target variable designation region 53d, a detail setting region 53e, and an execution button 53f. ing. The data flow display area 53a displays a data flow, for example, in the same manner as the data flow display area 51a of the data flow display user interface 51 shown in FIG. In the lower display area 53b, similarly to the lower display area 51b of the data flow display user interface 51 shown in FIG. 7, a relationship diagram between the analysis target data is displayed.

  The relationship designation area 53c shown in FIG. 11 is an area for designating the correspondence between a plurality of analysis target data. Here, the correspondence between “transaction-1” and “master” is specified. As an example, as described above, since the “customer ID” of “transaction-1” and the “customer ID” of “master” have a corresponding relationship, the “customer ID” of “transaction-1” and the “customer ID” of “master” Although the "customer ID" is specified as the correspondence, the present invention is not limited to this, and it is also possible to specify that any column of "transaction-1" and any column of "master" have a correspondence. , The names do not need to match.

  The relationship may be specified by selecting from a list of feature values, may be specified by dragging a line on the relationship diagram, or the name may be directly input. May be specified. If there is only one data, the correspondence between the analysis target data is not specified.

  Below the relationship designation area 53c, there are numerical values displayed in percentage. This numerical value indicates what percentage of the rows correspond to the two data to be analyzed by the designated column. For example, as shown in FIG. 12, one “1”, “2”, “3”, and “4” exist in the “ID” column of the first analysis target data, and the second analysis target data Assuming that there are three “1”, one “3”, and one “5” in the “ID” column of the data, “1” of the first data to be analyzed becomes “1” of the second data to be analyzed. Each corresponds to "1", "3" of the first analysis target data corresponds to "3" of the second analysis target data, and the other IDs do not correspond. In this case, 50% (2/4 rows) correspond to the first analysis target data and 80% (4/5 rows) correspond to the second analysis target data in the “ID” column, and the total is 6%. / 9 rows, that is, about 67% correspond. Such a calculation is performed, and the calculation result is displayed below the relationship designation area 53c.

  The target variable specifying area 53d is an area for specifying a target variable to be analyzed. By specifying the target variable, the table of the aggregation destination can be determined. Here, in order to analyze whether or not the customer has left, the “turn off” column of “master” is designated as the objective variable. Therefore, the feature amounts of “transaction-1” can be aggregated and added as a new feature amount of “master”. The target variable may be specified by selecting from a list of feature values, may be specified by dragging from the list, or may be specified by directly inputting a name. Is also good. When the target variable is specified, the data analyzer 1 sets a flag indicating that the target variable is specified as the target variable. By setting this flag, a feature amount described later is not generated from the target variable itself.

  The series of processes described above is a process of inputting the settings related to the generation of the feature amount in step SB3 of the flowchart shown in FIG.

  In a succeeding step SB4, a process of generating a feature amount is executed. This process is started by pressing an execution button 53f of the feature amount automatic generation user interface 53.

  That is, when a plurality of data to be analyzed is specified in steps SB1 and SB2 and a plurality of data relationships to be analyzed are specified in step SB3, for example, a method of generating a feature amount includes, for example, AJ Knobbe, "Propositionalisation and Aggregates", The techniques described in 2001 can be used. The example shown in FIG. 13 is a case where the sum is calculated for the column of “price” of “transaction” and is added as a feature amount of “master”. The purchase history in which the "customer ID" of "transaction" is "1" is extracted, and the value of "price" is totaled to obtain the total value of the purchase amount. For “master”, a column of “total purchase price” is generated, and the total value is automatically input to the “total purchase price” column of “customer ID” of “1”. This is possible because the correspondence between the rows between “master” and “transaction” is linked by the “customer ID” column.

  The rows to be aggregated may be determined based on, for example, a combination with a row representing a date. For example, as shown in FIG. 14, the purchase history in which the “customer ID” of “transaction” is “1” is extracted, and the values of “price” within the last 10 days starting from February 1, 2018 are summed up. . For “master”, a column of “purchase amount within the last 10 days” is generated, and “customer ID” is “1”, and a column of “purchase amount within the last 10 days” is displayed in the column of “price within the last 10 days”. Is automatically entered. Similar processing can be performed for all customer IDs.

  When the analysis target data is time-series data, the aggregation period can be switched to, for example, 10, 20, or 30 days. By switching the aggregation period, various types of feature amounts can be generated even with the same aggregation function. The unit of the period may be any of seconds, minutes, hours, months, and years. For example, it is also possible to adopt a configuration in which the total period of the analysis target data is acquired, and the acquired total period is automatically divided into a plurality of total periods. When the total period of the analysis target data is 30 days, the data can be divided into 0 to 10 days, 11 to 20 days, and 21 to 30 days. This division of the counting period can be configured to be performed when a later-described derived diagram generation unit 16b generates a derived diagram.

  After generating the feature, the importance of the prediction of the objective variable is evaluated for each feature. The prediction of the objective variable is executed by the prediction model generation unit 13 shown in FIG. This is the prediction model generation step. The prediction model generation unit 13 generates a prediction model for predicting a target variable from a plurality of feature amounts. The prediction model can be created by, for example, multiple regression analysis or a decision tree. There are various generation methods, and in each case, a conventionally known method can be used.

  After generating the prediction model, the importance for prediction by the prediction model is calculated for each of the plurality of feature amounts. This is a process executed by the importance calculation section 14 shown in FIG. 2, and is a importance calculation step. The importance of the feature is known as Feature Importance, and various calculation methods have been proposed, and any of the calculation methods may be used. For example, in the linear multiple regression model represented by the following equation (1), coefficients a1, a2,... For each feature x1, x2,... May be obtained, and the absolute values thereof may be used as importance values.

y = a1x1 + a2x2 +... + anxn (1)
Alternatively, instead of using the coefficient as it is, a normalized coefficient corrected so that the magnitudes of the feature amounts x1, x2,.

  After performing such an analysis of the importance, the control unit 10 extracts the one with the higher importance. At this time, with respect to the feature amounts having different aggregation periods, even if a plurality of feature amounts are ranked higher in importance, the others except for the most important one are deleted. This is because, for example, if the “purchase amount for the last 30 days”, “the purchase amount for the last 20 days”, and the “purchase amount for the last 10 days” are at the top of the feature amount at the same time, if it is presented to the user as it is, Since feature values having substantially the same meaning are displayed repeatedly, it is difficult for the user to understand. In addition, if a feature having the same meaning occupies the upper position, other effective features are expelled from the upper position, and it is difficult to obtain effective knowledge.

  The number in the “higher” may be, for example, 2 or more, 5 or more, 10 or more, or 20 or more. The number in the “upper rank” may be set automatically by the control unit 10 or may be set to an arbitrary number by the user.

  The above-described series of processes is the feature amount generation process in step SB4 of the flowchart shown in FIG. When the generation of the feature amount is completed, the process proceeds to Step SB5. Step SB5 is a step of outputting information of the feature amount generated in step SB4.

  When the generation of the feature amount is completed and the upper-level feature amount is extracted, the feature amount display user interface 60 shown in FIG. 15 is displayed on the monitor 3 and the user operation performed on the feature amount display user interface 60 is performed. Accept.

  The feature amount display user interface 60 is provided with a feature amount display area 60a and a detailed information display area 60b. The feature amount display area 60a and the detailed information display area 60b can be arranged so as to be arranged on the left and right sides of the monitor 3. In the feature amount display area 60a, higher-order feature amounts are displayed. That is, the monitor 5 displays a feature amount having a higher importance based on the importance calculated by the importance calculator 14. This step is a feature amount display step.

  The monitor 5 is configured to arrange and display a plurality of feature values having higher importance levels in descending order of importance, that is, to perform a ranking display. In this example, the feature value having the highest importance and a plurality of feature values less than the highest importance are simultaneously displayed, and the feature value having the highest importance is displayed at the top. , And a plurality of feature amounts less than the highest importance are arranged in descending order according to the magnitude of the importance. The feature amounts may be arranged in ascending order or may be arranged so as to be arranged in the left-right direction. The feature amount displayed in the feature amount display area 60a may be a predetermined number, and the number is not particularly limited.

  The feature amount display area 60a is provided with a name display unit 60c for displaying the name of the feature amount and an importance display unit 60d for displaying the importance. The name of the feature amount existing in “transaction” and “master” is displayed on the name display unit 60c. The names of the feature amounts that existed in “transaction” and “master” can be displayed as they are on the name display unit 60c, but may be changed by the user.

  The importance level is displayed in the importance level display section 60d so as to be arranged next to each feature amount in a bar graph format. In the case of displaying in a bar graph format, the longer the value, the higher the importance. The importance degree display unit 60d may display the weight in numerical values other than the various graph formats, or may display the numerical values and the graph at the same time.

  A feature amount addition button 60f is provided in the feature amount display area 60a. When a feature amount addition button 60f is pressed, feature amounts other than those displayed in the feature amount display area 60a are displayed, and the user selects a feature amount desired to be displayed in the feature amount display area 60a from the displayed feature amounts. You can do it. The selected feature amount is displayed in the feature amount display area 60a. It should be noted that it is also possible to delete or hide a feature that is considered unnecessary in the feature displayed in the feature display area 60a.

(Detailed operation of control unit 10)
The feature displayed on the feature display area 60a of the feature display user interface 60 shown in FIG. 15 can be selected by the user. When the user performs an input operation to select which feature amount using the mouse 5 or the like, the first basic diagram 70 is displayed in the detailed information display area 60b of the feature amount display user interface 60 as shown in FIG. Is displayed. The first basic diagram 70 is generated by the basic diagram generator 16a of the display diagram generator 16 shown in FIG. This process is a basic diagram generation step. In other words, the basic diagram generation unit 16a generates a basic diagram indicating the relationship between the first characteristic amount selected from the characteristic amounts displayed on the monitor 3 in accordance with the user's input and the objective variable. It is. The first feature amount is a feature amount selected by the user, and in the example illustrated in FIG. 16, is the “total number of visiting days in the last 30 days”. The feature quantity selected by the user is highlighted so that it can be easily distinguished from other feature quantities. Also, the first basic diagram 70 divides the analysis target data into a plurality of clusters based on the value of the feature amount selected according to the input of the user, and represents each cluster and a representative of the objective variable of each cluster. It is a diagram showing a relationship with a value.

  Hereinafter, the first basic diagram 70 will be described in detail. The first basic diagram 70 can be divided into three regions: an upper region 71, an intermediate region 72, and a lower region 73. In the upper area 71, a line graph is displayed in which the vertical axis represents the rate of separation, and the horizontal axis represents the total number of visiting days of the last 30 days. The ratio of abandonment is the ratio of customers who have visited the store once, but have stopped visiting the store after that. On the horizontal axis, days 0 to 30 are divided into six periods (sections) and displayed. In the line graph, the relationship between the selected feature amount and the objective variable (separation) is displayed.

  In the intermediate area 72, a bar graph is displayed in which the vertical axis represents the number of data and the horizontal axis represents the total number of days of visits to the store for the last 30 days. The horizontal axis of this bar graph is shared with the horizontal axis of the line graph displayed in the upper area 71. In this bar graph, the number of data present in each section is expressed, and a frequency distribution table is provided. That is, the control unit 10 divides the analysis target data into a plurality of periods based on the value of the first feature amount to form clusters, and generates a data number display graph indicating the number of analysis target data belonging to each cluster. It is configured to be. The monitor 3 is configured to be able to display the generated data number display graph.

  From the line graph in the upper area 71 and the bar graph in the intermediate area 72, "the more the total number of days of visits in the last 30 days, the lower the ratio of data of churn = 1", "the number of days of total visits from 21 to 25 days" In this case, the ratio of separation = 1 is the minimum, but the number of applicable data is small. "

  Further, the control unit 10 calculates an average value of the ratio of separation = 1. The average value can be the average value of all data, and is 21% in this example. Further, in a section where it is determined that there is a significant difference in the average value of the objective variable compared to the overall average value (21%), the series is highlighted, and the explanatory note is displayed at the bottom of the graph, that is, in the lower area 73. Is added to

  If there is no significant difference, no highlight is displayed. "No significant difference" means that the representative value of the section does not have a sufficient difference from the representative value of the section to be compared, taking into account the data variation and the number of data. Refers to. The highlighted display is a display that can be distinguished from other characters, for example, by making the character thicker, changing the character color from another character color, or adding a color to the background of the character.

  The explanatory sentence is a sentence that explains the relationship between the selected first feature amount and the objective variable, and is automatically generated by the automatic explanatory sentence generating unit 18 shown in FIG. The explanatory sentence automatic generation unit 18 selects the first feature amount based on, for example, the name of the analysis target data received by the data input unit 12, the name of the feature amount, the relative difference between the feature amounts, the number of data, and the like. Can be generated to explain the degree of influence of the. This makes it easier for the user to grasp the relationship between the feature value and the target variable, and saves the user from having to create a comment.

  Here, the determination of the significant difference will be described. An index such as a p-value or a t-test commonly used in statistics can be used to determine a significant difference. For example, the t test is calculated by the following equation 1. x and s are the average and standard deviation of the target values in the series of interest, μ is the overall average of the target values, and n is the number of data in the series. The greater this value is, the more significant the difference can be determined.

  By using these indices, it is possible to not only simply look at the degree of deviation from the overall average, but also take into account the amount of data and the variability of individual data. It can be distinguished whether a certain difference occurs.

  By highlighting using this statistical test, the user can pay attention to a truly significant place from the data, and can quickly obtain useful knowledge from the automatically generated feature amount.

  When an arbitrary section is selected from the graph displayed in the upper area 71, a detailed menu 74 is displayed as shown in FIG. The detailed menu 74 is a menu for comparing data included in the selected section with the overall tendency, and can display more detailed information according to the menu selection. The detailed menu 74 includes three menus: “compare the average value”, “compare the contributions of the feature amounts”, and “calculate the description of the difference”.

  As shown in FIG. 17, for example, when the menu “Compare the average value” is selected for the section “Total number of visits to the last 30 days to 5 days”, as shown in FIG. A comparison table 75 in which the average value of each feature amount is compared for each of the data of the section and all the data is displayed.

  Specifically, when the user performs an operation (click operation or the like) of selecting one of the plurality of clusters shown in the first basic diagram 70 shown in FIG. Accepted by the selector 15. This process is a cluster selection step. After the selection is received by the cluster selection unit 15, the comparison diagram generation unit 16 c illustrated in FIG. 2 selects the feature amount that contributes to the prediction of the objective variable in the selected cluster, and selects all the clusters or the cluster selection unit 15. A comparison table 75 (shown in FIG. 18) is generated, which shows a comparison with a feature value that contributes to the prediction of the objective variable in the unselected clusters that have not been selected. This process is a comparison diagram generation step.

  Although the details will be described later, the comparison diagram generation unit 16c may be configured to generate a comparison diagram indicating a feature amount whose contribution degree is significantly high only in the cluster selected by the cluster selection unit 15. Further, the comparison diagram generation unit 16c may be configured to generate a comparison diagram indicating a feature amount that is not selected by the user among feature amounts having higher importance. Further, the comparison diagram generation unit 16c may be configured to generate a comparison diagram indicating a feature amount contributing to the prediction of the objective variable in the missing cluster in which the value of the analysis target data does not exist.

  The comparison table 75 is a comparison diagram, and may be a comparison diagram displaying a graph in addition to the table format. The monitor 3 is configured to be able to display the first basic diagram 70 and the comparison table 75 simultaneously or separately, and to display the first basic diagram 70 and the comparison table 75 simultaneously or separately. Can be performed.

  In the comparison table 75, only the feature amount having a significantly large difference from the average value is displayed from among all the feature amounts including the ones having low importance. The p-value and the t-test can be used for the determination of the significance here as well.

  The feature amount displayed in the comparison table 75 may be configured to be selectable by the user. In the lower part of the comparison table 75, a feature amount addition button 75a is provided. When the feature amount addition button 75a is pressed, an arbitrary feature amount can be added as a comparison axis. The feature amount displayed in the comparison table 75 may be configured so that the user can delete it.

  In this example, the average value is displayed in the comparison table 75. However, for example, a statistic other than the average value such as a variance, a median value, and a minimum value may be displayed.

  As a result, when an interesting data group (in this example, a group of customers with a high regression rate) is found using a certain feature as a starting point, it is possible to dig into the details of the unique properties of the data group, and as a result And new knowledge can be obtained.

  Furthermore, since the feature quantity serving as the axis of comparison and analysis is automatically generated, for example, a skill such as "customers who have a small number of total visits in the last 30 days have a particularly small purchase price of vegetables and daily items" It is possible to easily obtain deep knowledge that is difficult to notice unless the analyst is an analyst.

  Further, the graph displayed when the feature value is selected may include one or more other graphs in addition to the relationship with the objective variable. For example, when “Total number of days visited in the last 30 days” is specified, as shown in FIG. 19, the prediction accuracy comparison regarding “Comparison of accuracy with the case of totaling in periods other than 30 days” as a second graph A graph 76 can be displayed. The prediction accuracy comparison graph 76 is located below the first basic diagram 70 shown in FIG. 18, and therefore, from the state where the first basic diagram 70 is displayed in the detailed information display area 60b, the vertical scroll bar 60e is operated. By scrolling down, the prediction accuracy comparison graph 76 can be displayed. The first basic diagram 70 and the prediction accuracy comparison graph 76 may be displayed simultaneously. As described above, it is possible to execute the derived figure display step of simultaneously or separately displaying the first basic diagram 70 and the prediction accuracy comparison graph 76.

  The prediction accuracy comparison graph 76 shown in FIG. 19 can also be called a third derivative diagram. The third derivative figure shows the accuracy of the prediction model generated based on the first feature amount (in this example, “the total number of days visited in the last 30 days”), belongs to the same category as the first feature amount, and FIG. 9 is a diagram illustrating a comparison between the first feature amount and the accuracy of a prediction model generated based on another feature amount having a different aggregation period. The fact that "the aggregation periods are different" is shown on the horizontal axis of the prediction accuracy comparison graph 76, and is specifically displayed as "10 days", "20 days", .... This third derivative drawing is generated by the derivative drawing generator 16b shown in FIG. This process is a derived diagram generation step. The total number of visiting days of the last 10 days, the total number of visiting days of the last 20 days, and the total number of visiting days of the last 30 days belong to the same category of “total number of visiting days”.

  When the derivative drawing generation unit 16b generates the third derivative drawing, the accuracy of the prediction model predicted based on another feature amount belonging to the same category as the first feature amount and having a different aggregation period is calculated. 1 can be compared with the accuracy of the prediction model generated based on the feature amount. This makes it possible for the user to easily determine which prediction model should be used to improve the accuracy after understanding the relative difference in accuracy between the prediction models.

  Further, when generating the third derived diagram, the derived diagram generating unit 16b calculates the accuracy of each prediction model generated based on a plurality of different feature amounts and the prediction generated based on the first feature amount. Generate a diagram that expresses the accuracy of the model in a comparable manner. Further, when generating the third derived diagram, the derived diagram generating unit 16b determines, as another feature, the accuracy of the prediction model generated based on the feature having a predetermined importance or more, and the first feature. Generating a diagram that allows comparison with the accuracy of the prediction model generated based on. Specifically, it is a graph, but may be a diagram that can be compared with other numerical values.

  For example, as described above, a plurality of patterns of feature amounts having different count periods are generated for feature amounts that are totaled by time or number of days, and an optimal one is automatically selected from among them. . In FIG. 19, by presenting the difference in accuracy from the unselected tally period as a relative value, it is possible to indicate "what would have happened if the total number of days visited in another tally period was used as the feature amount".

  As the accuracy, an F value, AUC, or the like generally used in data analysis is used. The accuracy may be an absolute value, or may be an index combining viewpoints other than accuracy, such as processing time and data amount. As a result, the user who thinks, "Why was the total number of visiting days of 30 days, not 20 days or 40 days, output?" Further, in response to a request that “if the accuracy does not change much, it is desirable to reduce the amount of data to be aggregated as much as possible”, a determination criterion can be provided by presenting the degree of influence of the amount of data.

  FIG. 20 shows a case where “the number of days elapsed since the last visit to the store” is selected from the feature amounts displayed in the feature amount display area 60 a of the feature amount display user interface 60. A second basic diagram 77 is displayed in the detailed information display area 60b of the feature amount display user interface 60. The second basic diagram 77 is generated by the basic diagram generator 16a of the display diagram generator 16 shown in FIG. 2, and this process is a basic diagram generation step. That is, the basic diagram generation unit 16a generates a basic diagram showing the relationship between the first characteristic amount selected from the characteristic amounts displayed on the monitor 3 according to the user's input and the objective variable. In the example shown in FIG. 20, the first feature amount is “the number of days elapsed since the last visit to the store”.

  The second basic diagram 77 can be divided into three regions: an upper region 77a, an intermediate region 77b, and a lower region 77c. In the upper area 77a, a line graph is displayed in which the vertical axis represents the separation rate and the horizontal axis represents the number of days elapsed since the last visit to the store. On the horizontal axis, days 0 to 30 are divided into six periods (sections) and displayed.

  In the intermediate area 77b, a bar graph is displayed in which the vertical axis represents the number of data and the horizontal axis represents the number of days elapsed since the last visit to the store. The horizontal axis of this bar graph is shared with the horizontal axis of the line graph displayed in the upper area 77a. In the lower area 77c, a description automatically generated by the description automatic generation unit 18 is displayed.

  In the example shown in FIG. 20, the relationship between the selected feature amount and the objective variable is also displayed. However, the series to be emphasized and the contents of the description sentence described in the lower area 77c are different from the characteristics of the feature amount. Accordingly, this is different from the example shown in FIG.

  Further, the second graph for detailed description of the feature amount is an impact comparison graph 78 showing a composite effect with another feature amount, as shown in FIG. Such an influence comparison graph 78 can also be generated, and this graph 78 can be called a simultaneous display graph that simultaneously displays the first feature amount and the second feature amount. The generated simultaneous display graph can be displayed on the monitor 3. The influence comparison graph 78 may be displayed simultaneously with the second basic diagram 77 or may be displayed separately. The step of displaying the influence comparison graph 78 and the second basic diagram 77 is a derived figure display step.

  The influence comparison graph 78 can also be called a first derivative drawing. The first derivative diagram divides the analysis target data into a plurality of clusters based on the value of the first feature amount (in this example, “the number of days elapsed since the last visit to the store”), and represents the representative value of the objective variable of each cluster. Is calculated, and in at least one of the clusters, a second feature value having a representative value of the objective variable determined to have a significant difference from the representative value of the objective variable of the first feature value (this example FIG. 7 is a diagram illustrating the relationship between the second feature amount and the objective variable by extracting “the total number of visiting days in the last 30 days”. This first derivative drawing is generated by the derivative drawing generator 16b shown in FIG. This process is a derived diagram generation step.

  That is, when the derivative diagram generation unit 16b illustrated in FIG. 2 generates the impact comparison graph 78 (shown in FIG. 21) as the first derivative diagram, for example, the first feature amount divided into a plurality of periods is used. And calculate an average value, a median value, or the like as a representative value of the objective variable in each period. The objective variable having a significant difference from the representative value of the objective variable of the first feature amount in an arbitrary period. , And a relationship between the second feature value and the target variable can be shown. Thereby, it is possible to obtain knowledge that could not be obtained only by the first feature amount. The condition for distributing the analysis target data can be referred to as a section, and the section may be not only a period but also, for example, gender, location, and the like.

  Further, the derived diagram generation unit 16b divides the analysis target data into a plurality of clusters based on the value of the first feature amount, and generates a data number display graph 78a indicating the number of analysis target data belonging to each cluster. Is configured. When dividing the analysis target data into a plurality of clusters, the derived diagram generation unit 16b acquires all the aggregation periods of the analysis target data, and automatically divides the acquired total aggregation periods into a plurality of aggregation periods. As a result, it is possible to save the user's labor of the division work.

  The data number display graph 78a is a bar graph displayed below the first derivative diagram (impact degree comparison graph 78). The monitor 3 is configured to be able to display the data number display graph 78a generated by the derived diagram generation unit 16b. According to the data number display graph 78a, the number of data to be analyzed can be grasped.

  Further, the configuration may be such that the user is notified that there is a cluster in which the number of data to be analyzed is equal to or less than the first predetermined number. For example, if there is a cluster having a number of data of less than a few percent of the total number of data, the user is notified that the number of data constituting the cluster is less than a few percent of the total number of data, so that the user can recognize the characteristic of the cluster. It is a material for judging the reliability of quantity.

  At the time of generating the first derived drawing, the derived drawing generating unit 16b regards the feature quantity in which the number of data to be analyzed is equal to or smaller than the second predetermined number to the representative value of the objective variable of the first feature quantity. It is configured to determine that there is no difference. The second predetermined number may be the same as the first predetermined number, or one may be smaller than the other. For example, if there is a feature amount having only a few percent or less of the total number of data, it can be considered that the reliability of the objective variable is poor. In such a case, it is determined that there is no significant difference. Thus, it can be prevented from being used for analysis.

  The influence comparison graph 78 shows that even if the data is in the same section when viewed only with the selected feature value, further dividing the data by another feature value causes a large difference in the relationship with the objective variable. ing. In the line graph displayed in the upper area 77a shown in FIG. 20, the fact that "the smaller the number of days elapsed since the last visit to the store, the lower the ratio of churn = 1" could be read. From the impact comparison graph 78 shown, it is possible to further obtain a new finding that "even if the number of days elapsed since the last visit to the store is small, the customer who has visited the store for less than 10 days has a high turnover rate."

  In the influence degree comparison graph 78, it is tested whether or not the difference in the section due to another difference in the feature amount is significant, and only the section determined to be significant is highlighted. Here, the feature amount “the total number of days visited in the last 30 days” is combined with the selected feature amount. However, when a graph of the combination of all other feature amounts is displayed, the number displayed is enormous.

  Therefore, after calculating the value of the significance for each combination, only those having a particularly high significance are selected and displayed. In the example shown in FIG. 16, the influence comparison graph 78 is not described, because there is no significant difference in the series in any combination. That is, when a significant difference occurs in the series, the influence comparison graph 78 can be automatically generated and displayed.

  At this time, the feature amount to be combined may be a feature amount with a high importance or a feature amount with a low importance, and can be determined regardless of the level of the importance.

  In addition, as to which section (5 days in FIG. 21) the feature quantity to be combined is divided into, the significance is verified by each of various division methods, and a value having the highest significance is selected. be able to. For example, the significance can be verified in each of the case where the data is divided in 5 days and in the case where the data is divided in 10 days, and the control unit 10 may automatically perform such a method of dividing the section. Alternatively, the user may manually perform the setting.

  In the feature amount display user interface 60 shown in FIG. 22, a detailed menu 74 is displayed by selecting an arbitrary section from the graph displayed in the upper area 71 thereof. , “Comparison of contribution of feature amount” is selected. Then, the contribution degree display graph 79 shown in FIG. 23 can be displayed on the feature amount display user interface 60. In the contribution degree display graph 79, the contribution degree of each feature amount is compared between the selected data group and the whole, and is equivalent to a comparison diagram. Therefore, the contribution degree display graph 79 is generated by the comparison diagram generation unit 16c. This step is a comparison diagram generation step. Further, the contribution degree display graph 79 and the second basic diagram 77 can be displayed on the monitor 3 simultaneously or separately. The process of displaying the contribution degree display graph 79 and the second basic diagram 77 simultaneously or separately is a comparative diagram display step.

  Here, a method of calculating the contribution of the feature amount will be described in detail. In an analysis technique called linear multiple regression analysis, a prediction formula (2) expressed in the following format is used.

y = a1x1 + a2x2 + a3x3 +... + anxn + b (2)
.. represent the predicted values, x1, x2,... represent the values (explanatory variables) of the respective feature values, and a1, a2,. B is a constant term. In the multiple regression analysis, the coefficient a and the constant term b are learned so that the value of y approaches the target variable for each data.

  From the above equation (2), the predicted value y is composed of the sum of terms (a1x1, a2x2,...) Relating to each feature and a constant term. It can be interpreted that the larger the absolute value of the term, the greater the feature value has on the prediction.

  Thus, the product axii of the coefficient and the value of the feature amount is defined as a contribution degree regarding the feature amount i. By averaging this contribution for predictions for a plurality of data, it is possible to calculate the average contribution to the prediction of the data group for an arbitrary data group.

  For example, as shown in FIG. 24, when y = 3x1-2x2 + 1, when averaging four data, feature x1 has a slightly positive contribution to the predicted value, and feature x2 has a slightly positive contribution to the predicted value. Has a large negative contribution.

  In this example, the procedure for calculating the degree of contribution by the linear multiple regression analysis of the above equation was explained.However, even with a non-linear algorithm, similar calculations can be performed by approximating each prediction with a linear model. Yes (eg, Scott M. Lundberg, “Consistent Individualized Feature Attribution for Tree Ensembles” (2018)).

  The contribution degree display graph 79 shown in FIG. 23 shows that the contribution degree defined above is calculated for the data being selected and also for all the data, and the two are compared. According to the contribution degree display graph 79, for example, tendencies such as "the minimum purchase amount does not significantly affect the separation, but has a large effect on the data group of interest", "However, the influence of the entire purchase amount is still larger" Can be grasped, and this becomes new knowledge.

  The feature values displayed in this example are narrowed down to two features, a feature value having a higher importance and a feature value (minimum purchase price) whose contribution is significantly higher only in the selected data group. In addition, only those having a significantly higher difference in contribution may be displayed. The feature quantity to be displayed may be deleted by the user.

  When "sex" is selected from the feature amounts displayed in the feature amount display area 60a of the feature amount display user interface 60 shown in FIG. 15, the third basic diagram 80 shown in FIG. Will be displayed. The third basic diagram 80 is generated by the basic diagram generation unit 16a of the display diagram generation unit 16 shown in FIG. 2, and this process is a basic diagram generation step. In the third basic diagram 80, unlike the example shown in FIG. 16 and the example shown in FIG. 20, the feature amount is not a numerical value, so the expression of the third basic diagram 80 is changed to a bar graph. The item “missing” indicates a case where no value exists in the original analysis target data. In the example shown in FIG. 25, no significant difference was found in the value of the objective variable due to the difference in gender, and therefore, none of the series is displayed in an emphasized state. However, if a significant difference is found, it can be highlighted.

  Here, after selecting the series “missing” shown in FIG. 25 and displaying the detailed menu 74 as shown in FIG. 17, selecting “calculate the explanation of the difference” from the detailed menu 74 results in the processing shown in FIG. An explanation graph 81 of the difference can be displayed on the feature amount display user interface 60. The difference description graph 81 shows which feature value mainly explains the difference in the value of the objective variable when the selected data group is compared with the entire data group, and corresponds to a comparison diagram. Therefore, the difference description graph 81 shown in FIG. 26 is generated by the comparison diagram generation unit 16c. This step is a comparison diagram generation step. Further, the difference explanatory graph 81 and the third basic diagram 80 can be displayed on the monitor 3 simultaneously or separately. The process of displaying the difference explanatory graph 81 and the third basic diagram 80 simultaneously or separately is a comparative diagram display step. Note that the above-described value of the contribution is used for the calculation for the explanation.

  Here, a method of calculating the difference will be described. First, for all data and each selected data group, the above-described contribution is calculated for each feature amount.

y_all = B1all + B2all + B3all + ... + Bnall + b ... (3)
y_sel = B1sel + B2sel + B3sel + ... + Bnsel + b ... (4)
y_all and y_sel represent average predicted values for all data / selected data. Biall and Bisel represent the contribution of all data / selected data for the i-th feature amount.

When (4) − (3) is calculated,
y_sel = y_all + (B1sel-B1all) + (B2sel-B2all) +… + (Bnsel-Bnall)
... (5)
Becomes The second and subsequent terms on the right side represent the difference between the contributions of the selected data and all data (this is referred to as Di). Further, y_sel and y_all are considered to include prediction errors with respect to the actual average values y_sel_true and y_all_true of the objective variables.

y_sel_true = y_sel + δ_sel (6)
y_all_true = y_all + δ_all (7)
Substituting (6) and (7) into (5) gives
y_sel_true = y_all_true + D1 + D2 +… Dn + (δ_all-δ_sel) ・ ・ ・ (8)
Becomes From equation (8), the average value of the objective variable of the selected data (y_sel_true) is the sum of the three elements of the average value of the objective variable of all data, the difference in the contribution of each feature, and the prediction error. Explain.

  In the above-mentioned graph, among the differences D1, D2,... Of the contributions, those having a particularly large value are individually displayed, and the remaining elements and the prediction errors are collectively displayed in the item of “other / prediction error”. I have.

  As described above, by displaying the graph 81 explaining the difference between the objective variables based on the difference between the contribution degrees, the question “Why is the gender = the departure rate of the user group with a deficiency as high as 54%?” "The high regression rate for gender-deficient users is partially affected by the" total purchase price "", "However, about 15% of the difference is explained in the main feature difference Can't do it ".

  However, as the gender = deficiency series was not originally highlighted, there is not enough information to say that gender = deficiency has a large departure, and the above insights may be incorrect. The wording to call attention is also displayed together with the graph 81.

  Further, as another graph for explaining the gender, as shown in FIG. 27, a graph 82 relating to a change in the degree of contribution of the feature amount can be displayed. In this example, the same type of graph displayed by clicking the graph in the example shown in FIG. 20 is displayed for gender.

  The graph 82 shown in FIG. 27 can also be called a second derivative diagram. The second derivative diagram is a diagram illustrating feature amounts that contribute to the prediction of the objective variable in a predetermined cluster among a plurality of clusters. The cluster is the same as that of the first derivative figure. In this example, gender is “male”, “female”, and “missing”, but may be a period. The second derivative drawing is generated by the derivative drawing generator 16b shown in FIG. This process is a derived diagram generation step. After the derived diagram generation step, a derived diagram display step of displaying the basic diagram and the second derived diagram simultaneously or separately can be performed.

  Further, in this example, the feature amount contributing to the prediction of the objective variable in the predetermined cluster and the first feature amount of all the clusters can be simultaneously displayed.

  When the derivative diagram generation unit 16b generates the second derivative diagram, when a cluster is divided into a man and a woman as described above, the degree of contribution to the prediction of the objective variable may greatly change depending on the gender, New knowledge can be obtained by indicating such feature amounts that contribute to the prediction of the objective variable.

  In addition, based on the graph shown in FIG. 27, “Gender alone does not affect the average value of the objective variable, but the degree of contribution of other feature values greatly changes depending on the gender. It is a feature amount. "

(Operation and effect of the embodiment)
As described above, according to the data analysis device 1 and the data analysis method according to this embodiment, it is possible to display, on the monitor 3, a highly important feature amount included in a large amount of data. Then, a basic diagram showing the relationship between the feature and the objective variable, a derived diagram showing the relationship between a plurality of features and the objective variable, and a derived diagram showing the feature that contributes to the prediction of the objective variable in a predetermined cluster. In addition, it is possible to display a derivative drawing or the like showing the accuracy comparison of the prediction models generated based on the feature amounts having different aggregation periods on the monitor 3.

  In addition, for each cluster obtained by dividing the data to be analyzed, the relationship between the cluster and the representative value of the objective variable can be known. Can be compared with feature values that contribute to the prediction of the objective variable in the cluster. Therefore, it can help to obtain new knowledge.

(Other embodiments)
The embodiments described above are merely examples in all respects and should not be construed as limiting. Furthermore, all modifications and changes belonging to the equivalent scope of the claims are within the scope of the present invention.

  In the above embodiment, a line graph or a bar graph is used to show the relationship between the objective variable and the feature value, but another expression method such as a scatter diagram may be used. In the case of a scatter diagram, instead of clicking the series of the graph, a part of the data can be selected by dragging an area in the graph, and a detailed graph can be further displayed.

  The first derivative diagram (the influence comparison graph 78 shown in FIG. 21), the second derivative diagram (the graph 82 regarding the change in the contribution of the feature amount shown in FIG. 27), and the third derivative diagram (the prediction shown in FIG. 19) Any one of the accuracy comparison graphs 76) and the basic diagrams 70, 77, and 80 may be simultaneously displayed on the monitor 3, or may be displayed at different timings. The derivative drawing generation unit 16b may generate all of the first derivative drawing, the second derivative drawing, and the third derivative drawing, and the generation order is not particularly limited. Further, all of the first derivative drawing, the second derivative drawing and the third derivative drawing may be displayed on the monitor 3.

  Although only three basic diagrams 70, 77, and 80 are shown in the present embodiment, the basic diagram generating unit 16a generates the basic diagrams 70, 77, and 80 according to the feature amount selected by the user. When switching between the basic diagrams 70, 77, and 80 and the first, second, and third derived diagrams for display, operation means such as a screen switching button can be used.

  As described above, the data analysis device and the data analysis method according to the present invention can be used when attempting to obtain useful information that has been unknown so far from a large amount of information.

1 data analyzer 3 monitor (display unit)
3a display control unit 11 main control unit 12 data input unit 13 prediction model generation unit 14 importance calculation unit 15 cluster selection unit 16a basic diagram generation unit 16b derived diagram generation unit 16c comparison diagram generation unit 17 automatic name generation unit 18 automatic description text Generation unit 30 Storage units 70, 77, 80 Basic diagram 76 Prediction accuracy comparison graph (third derived diagram)
75 Comparison table (comparison chart)
78 Impact comparison graph (first derivative)
79 Contribution display graph (comparison chart)
81 Difference graph (comparison chart)
83 Graph on the change in the contribution of the feature value (second derived figure)

Claims (7)

In a data analyzer that analyzes data to be analyzed,
A data input unit that receives input of analysis target data including a plurality of feature amounts and objective variables,
A prediction model generation unit that generates a prediction model for predicting the target variable from the plurality of feature amounts,
An importance calculation unit that calculates the importance for prediction by the prediction model for each of the plurality of feature amounts,
Based on the importance calculated by the importance calculation unit, a display unit that displays a feature amount having a higher importance,
The analysis target data is divided into a plurality of clusters based on a value of a feature amount selected according to a user input from among the feature amounts displayed on the display unit, and each cluster and an objective variable of each cluster are divided. A basic diagram generation unit that generates a basic diagram indicating a relationship with a representative value of
A cluster selection unit that receives selection of any one cluster from the plurality of clusters illustrated in the basic diagram;
In the cluster selected by the cluster selection unit, the feature amount contributing to the prediction of the objective variable, and in the entire cluster or in the unselected cluster not selected by the cluster selection unit, the characteristic amount contributing to the prediction of the objective variable A comparison diagram generation unit that generates a comparison diagram indicating a comparison with the amount.
The data analysis device, wherein the display unit is configured to be able to display a basic diagram generated by the basic diagram generation unit and a comparison diagram generated by the comparison diagram generation unit. The data analyzer according to claim 1,
The data analysis apparatus according to claim 1, wherein the display unit is configured to display a plurality of feature values having higher importance levels in order of higher importance. The data analyzer according to claim 2,
The data analysis device, wherein the display unit is configured to simultaneously display a feature having the highest importance and a plurality of features having a lower importance than the highest. The data analyzer according to any one of claims 1 to 3,
The data analysis device, wherein the comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount having a significantly higher contribution only in the cluster selected by the cluster selection unit. The data analyzer according to any one of claims 1 to 4,
The data analysis device, wherein the comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount not selected by a user among the feature amounts having the higher importance. . The data analyzer according to any one of claims 1 to 5,
The comparison diagram generation unit is configured to generate the comparison diagram indicating a feature amount contributing to the prediction of the objective variable in a missing cluster in which the value of the analysis target data does not exist, Data analyzer. In a data analysis method for analyzing data to be analyzed,
A data input step of receiving input of analysis target data including a plurality of feature amounts and objective variables,
A prediction model generation step of generating a prediction model for predicting the objective variable from the plurality of feature amounts,
An importance calculation step of calculating importance for prediction by the prediction model for each of the plurality of feature amounts;
A feature value display step of displaying a feature value having a higher importance based on the importance calculated by the importance calculation step;
The analysis target data is divided into a plurality of clusters based on the value of the feature amount selected in response to the user's input from the feature amounts displayed in the feature amount display step, and each cluster is divided into a plurality of clusters. A basic diagram generation step of generating a basic diagram showing a relationship between the objective variable and a representative value,
A cluster selection step of receiving a selection of any one cluster from the plurality of clusters shown in the basic diagram;
In the cluster selected in the cluster selection step, a feature amount contributing to the prediction of the objective variable, and in a cluster not selected in the entire cluster or the cluster selection step, a feature contributing to the prediction of the objective variable A comparison diagram generation step of generating a comparison diagram indicating the comparison with the amount;
A data analysis method, comprising: a comparison diagram display step capable of displaying a basic diagram generated in the basic diagram generation step and a comparison diagram generated in the comparison diagram generation step.