Feature extraction device, feature extraction method, and feature extraction program

Abstract

This feature extraction device comprises: a combining unit (11) that generates a plurality of data pairs that are combinations of two sets of time-series data; a data analysis unit (12) that analyzes the degree of similarity between the two sets of time-series data included in each data pair using a plurality of analysis methods; and an occurrence probability calculation unit (13) that calculates, for each analysis method, the probability of occurrence of the degree of similarity for each data pair. The feature extraction device further comprises: a deviation level calculation unit (14) that calculates a deviation level of the probability of occurrence for each data pair calculated for each analysis method; a visualization unit (15) that visualizes and presents each set of time-series data included in a data pair and the deviation level to a user; an input unit (16) that receives similarity or dissimilarity determination input from the user; and a feature extraction unit (17) that extracts a feature of each analysis method on the basis of the determination input.

Description

[Technical field]

[0001] The present invention relates to a feature extraction device, a feature extraction method, and a feature extraction program. [Background technology]

[0002] A data analysis device for analyzing time series data is disclosed in Patent Document 1. Patent Document 1 describes a method for calculating an index value indicating the amount of change over time of each data to be analyzed, and displaying graphs of multiple time series data in an order based on the index value. Patent Document 1 makes it possible to focus on data that changes significantly in a specific period of time, thereby supporting data analysis. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6592411 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-mentioned Patent Document 1 does not mention determining which of multiple analysis methods is suitable for analyzing time series data when there are multiple analysis methods for analyzing the time series data.

[0005] This has resulted in the problem that it is difficult to select an appropriate analysis method when analyzing the target time series data.

[0006] The present invention has been made in consideration of the above-mentioned circumstances, and an object of the present invention is to provide a feature extraction device, a feature extraction method, and a feature extraction program capable of extracting features of an analysis method for analyzing time series data. [Means for solving the problem]

[0007] A feature extraction device according to one embodiment of the present invention includes a combination unit that generates a plurality of data pairs by combining two time series data; an analysis unit that analyzes the similarity of the two time series data included in each data pair using a plurality of analysis methods; an occurrence probability calculation unit that calculates the occurrence probability of the similarity of each data pair for each analysis method based on an analysis result by the analysis unit; a deviation calculation unit that calculates, for each data pair, a deviation of the occurrence probability calculated for each analysis method; a visualization unit that visualizes each time series data included in the data pairs and the deviation and presents them to a user; an input unit that accepts a judgment input of similarity or dissimilarity from the user; and a feature extraction unit that extracts features of the analysis method based on the judgment input and the deviation.

[0008] A feature extraction method according to one embodiment of the present invention includes the steps of: generating a plurality of data pairs by combining two time series data; analyzing the similarity of the two time series data included in each data pair using a plurality of analysis methods; calculating the occurrence probability of the similarity of each data pair for each analysis method based on the analyzed similarity; calculating a deviation of the occurrence probability calculated for each analysis method for each data pair; visualizing each time series data included in the data pair and the deviation and presenting it to a user; accepting a judgment input of similarity or dissimilarity from the user; and extracting features of the analysis method based on the judgment input and the deviation.

[0009] One aspect of the present invention is a feature extraction program for causing a computer to function as the feature extraction device. Effect of the Invention

[0010] According to the present invention, it is possible to extract the characteristics of an analysis method for analyzing time-series data. [Brief description of the drawings]

[0011] [Figure 1]FIG. 1 is a block diagram showing the configuration of a feature extraction device according to the first embodiment. [Diagram 2] FIG. 2 is an explanatory diagram showing an example of time series data and a data pair in which two pieces of time series data are combined. [Diagram 3] FIG. 3 is an explanatory diagram showing the first to fourth analysis methods. [Figure 4A] FIG. 4A is an explanatory diagram showing analytical values ​​calculated for a plurality of data pairs by the first to fourth analysis methods. [Figure 4B] FIG. 4B is an explanatory diagram showing distribution curves of the analytical values ​​shown in FIG. 4A, where (a) shows the distribution curve by the first analytical method, (b) shows the distribution curve by the first analytical method, (c) shows the distribution curve by the third analytical method, and (d) shows the distribution curve by the fourth analytical method. [Diagram 5] FIG. 5 is an explanatory diagram showing a normalized curve obtained by normalizing the distribution curve s1 shown in FIG. 4B. [Figure 6] FIG. 6 is a diagram showing a plurality of data pairs, the occurrence probability of similarity when each data pair is analyzed using four analysis methods, and the deviation of the occurrence probability. [Figure 7A] FIG. 7A is a diagram showing two time-series data constituting a data pair and occurrence probabilities calculated using four analysis methods. [Figure 7B] FIG. 7B is a diagram showing two time-series data constituting a data pair and occurrence probabilities calculated using the four analysis methods. [Figure 8] FIG. 8 is a flowchart showing the processing procedure of the feature extraction device according to the first embodiment. [Figure 9] FIG. 9 is an explanatory diagram showing characteristic patterns of each piece of time-series data acquired by the recording unit 18. As shown in FIG. [Figure 10] FIG. 10 is a block diagram showing the configuration of a feature extraction device according to the second embodiment. [Figure 11A] FIG. 11A is an explanatory diagram showing the analysis results of multiple data pairs and normalized curves of the analysis results. [Figure 11B] FIG. 11B is an explanatory diagram showing the items of the time-series data constituting the data pairs and the degree of deviation of the occurrence probability of each data pair. [Figure 12] FIG. 12 is a block diagram showing the hardware configuration of this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0013] [First embodiment] FIG. 1 is a block diagram showing the configuration of a feature extraction device according to the first embodiment.

[0014] 1, a feature extraction device 1 according to the first embodiment is connected to a database 2 (denoted as "DB" in the figure). The feature extraction device 1 includes a combination unit 11, a data analysis unit 12 (analysis unit), an occurrence probability calculation unit 13, a deviation calculation unit 14, a visualization unit 15, an input unit 16, a feature extraction unit 17, and a recording unit 18.

[0015] The database 2 stores a plurality (m items) of time series data qi (i=1 to m). The time series data qi is, for example, the Consumer Price Index provided by the Statistics Bureau of the Ministry of Internal Affairs and Communications.

[0016] The combination unit 11 generates a data pair by combining two time series data. Specifically, the combination unit 11 selects and combines two of the time series data qi stored in the database 2 to set a data pair aj (j=1 to n). FIG. 2 is an explanatory diagram showing an example of combining two time series data to set a data pair aj. As shown in FIG. 2, the time series data q1 and q2 are combined to generate a data pair a1. The time series data q2 and q3 are combined to generate a data pair a2. The time series data q1 and q4 are combined to generate a data pair a3. The time series data q3 and q4 are combined to generate a data pair a4.

[0017] When there are m pieces of time series data, "m*(m-1) / 2" data pairs are set. That is, "n=m*(m-1) / 2". For example, when there are 380 pieces of time series data, (380*379) / 2=72010 data pairs are set.

[0018] The data analysis unit 12 analyzes the similarity between the two time series data included in each data pair using a plurality of analysis methods. Specifically, the data analysis unit 12 includes a plurality of analysis method calculation programs for analyzing the data pair aj set in the combination unit 11. The data analysis unit 12 includes a first analysis unit 21 that analyzes the data pair using a first analysis method, a second analysis unit 22 that analyzes the data pair using a second analysis method, a third analysis unit 23 that analyzes the data pair using a third analysis method, and a fourth analysis unit 24 that analyzes the data pair using a fourth analysis method.

[0019] The data analysis unit 12 analyzes the similarity between the two time series data constituting the data pair aj using the first to fourth analysis methods, and outputs the analysis result as an analysis value. Note that in this embodiment, an example using four analysis methods is shown, but an analysis method other than these four may also be used. Specific processing of the first to fourth analysis methods will be described below with reference to FIG. 3.

[0020] The first analysis method, as shown in Fig. 3(a), integrates the absolute difference between two time series data. Specifically, the absolute value of the difference between one time series data and the other time series data obtained at a predetermined time interval is calculated, and the absolute value of the difference is integrated within a certain period of time. The integrated value is output as the analysis value. The higher the similarity between the two time series data, the smaller the analysis value will be.

[0021] As shown in FIG. 3(b), the second analysis method calculates the amount of change over time for each piece of time series data, and integrates the difference between the calculated amounts of change. Specifically, the difference between the amount of change between one piece of time series data and the other piece of time series data obtained at a predetermined time interval is calculated, and the difference is integrated within a certain period of time. For example, if the amount of change in one piece is "+1" and the amount of change in the other piece is "-1", the difference is "2". If the amount of change in one piece is "+1" and the amount of change in the other piece is also "+1", the difference is "0". If the amount of change in one piece is "-2" and the amount of change in the other piece is "+1", the difference is "3". The second analysis method integrates these differences, and outputs the integrated value as the analysis value. The higher the similarity between the two pieces of time series data, the smaller the analysis value.

[0022] As shown in FIG. 3(c), the third analysis method calculates the rate of change over time for each piece of time series data, and integrates the difference between the calculated rates of change. Specifically, the difference between the amount of change between one piece of time series data and the other piece of time series data obtained at a predetermined time interval is calculated, and the difference is integrated over a certain period of time. For example, if one piece of time series data is "+3%" and the other piece of time series data is "-1%, the difference is "4". If one piece of time series data is "+1%" and the other piece of time series data is also "+1%, the difference is "0". The third analysis method integrates these differences over a certain period of time, and outputs the integrated value as the analysis value. The higher the similarity between the two pieces of time series data, the smaller the analysis value will be.

[0023] In the fourth analysis method, as shown in Fig. 3(d), the average value is calculated for each time series data at a predetermined time interval. Furthermore, as in the third analysis method described above, the difference between the average values ​​is integrated within a certain period, and the integrated value is output as the analysis value. The higher the similarity between the two time series data, the smaller the analysis value will be.

[0024] The first analysis unit 21 to the fourth analysis unit 24 create a distribution curve by plotting the analysis values ​​of each data pair aj based on the analysis values ​​calculated by the first to fourth analysis methods. Hereinafter, the processing of the data analysis unit 12 will be described in detail with reference to Figs. 4A and 4B.

[0025] 4A is an explanatory diagram showing analytical values ​​calculated for multiple data pairs by the first to fourth analysis methods. The data analysis unit 12 calculates analytical values ​​"bk-j" calculated by the first to fourth analysis methods for each data pair aj (j=1 to n) stored in the database 2 shown in FIG. 1. Note that "k" indicates the number of the analysis method, and "j" indicates the number of the data pair. In other words, "k" is an integer from 1 to 4, and "j" is an integer from 1 to n.

[0026] For example, for data pair a1 of two time series data "Vegetables / Seaweed" and "Sushi (eating out)", the analysis value calculated using the first analysis method is assumed to be "b1-1". In FIG. 4A, the analysis value "b1-1" is set to "0.05". For data pair a1, the analysis value calculated using the second analysis method is assumed to be "b2-1". In FIG. 4A, the analysis value "b2-1" is set to "0.21".

[0027] Similarly, the analytical value calculated for data pair a2 using the third analysis method is set as analytical value "b3-2". In FIG. 4A, the analytical value "b3-2" is set as "0.33". The analytical value calculated for data pair a3 using the fourth analysis method is set as analytical value "b4-3". In FIG. 4A, the analytical value "b4-3" is set as "0.64". Similarly, each analytical value "bk-j" is calculated.

[0028] The data analysis unit 12 generates distribution curves of the analysis values ​​of the data pairs aj calculated by the first to fourth analysis methods. Specifically, for the data pairs aj (j=1 to n), distribution curves s1 to s4 of the analysis values ​​"bk-1 to bk-n" (k=1 to 4) calculated by the first to fourth analysis methods are generated. For example, as shown in Fig. 4B(a) to (d), distribution curves s1 to s4 of "b1-j", "b2-j", "b3-j", and "b4-j" (where j=1 to n) are generated. Fig. 4B(a) to (d) are graphs plotting the analysis values ​​by the first to fourth analysis methods, with the horizontal axis indicating the analysis value and the vertical axis indicating the frequency.

[0029] In FIG. 4B(a), analytical values ​​b1-j obtained by analyzing n data pairs a1 to an by the first analysis method are plotted, and a curve along each analytical value is designated as a distribution curve s1.

[0030] In FIG. 4B(b), analytical values ​​b2-j obtained by analyzing n data pairs a1 to an using the second analysis method are plotted, and a curve along each analytical value is designated as a distribution curve s2.

[0031] In FIG. 4B(c), analytical values ​​b3-j obtained by analyzing n data pairs a1 to an using the third analysis method are plotted, and a curve along each analytical value is designated as a distribution curve s3.

[0032] In FIG. 4B(d), analytical values ​​b4-j obtained by analyzing n data pairs a1 to an using the fourth analysis method are plotted, and a curve along each analytical value is designated as a distribution curve s4.

[0033] Returning to FIG. 1, the occurrence probability calculation unit 13 normalizes the distribution curves s1 to s4 created by a plurality of analysis methods. That is, the distribution curves s1 to s4 shown in FIG. 4B(a) to (d) cannot be directly compared with each other. Therefore, the distribution curves s1 to s4 are normalized. For example, by normalizing the distribution curve s1, the normalized curve s11 shown in FIG. 5 is obtained. That is, the occurrence probability calculation unit 13 normalizes the analysis result of the data analysis unit 12 to calculate the occurrence probability.

[0034] The occurrence probability calculation unit 13 calculates the occurrence probability of the similarity of each data pair for each analysis method based on the analysis result by the data analysis unit 12. Specifically, the occurrence probability calculation unit 13 calculates the occurrence probability of the analysis value (i.e., similarity) by the first to fourth analysis methods for each data pair aj (j=1 to n). The occurrence probability is an index indicating the rank of the analysis value in the range of "0 to 1". The closer the occurrence probability is to "0", the higher the similarity of the two time series data is. The closer the occurrence probability is to "1", the lower the similarity of the two time series data is. The occurrence probability of the analysis value "bk-j" of the data pair aj by the kth analysis method is indicated by "pk-j". The occurrence probability of the analysis value "b1-1" of the data pair a1 by the first analysis method is "p1-1". For example, if the analysis value "b1-1" of the data pair a1 obtained by the first analysis method belongs to the top 30%, the occurrence probability "p1-1" is "0.3".

[0035] In the normalized curve s11 shown in FIG. 5, the higher the occurrence probability of the analysis value of the data pair aj, the higher the similarity between the two time series data included in the data pair aj.

[0036] The deviation calculation unit 14 calculates the deviation of the occurrence probability calculated for each data pair aj for each analysis method. Specifically, the deviation calculation unit 14 calculates the deviation of the occurrence probability for the target data pair aj using four analysis methods. The deviation of the data pair aj using the k-th analysis method is indicated as "dk-j". For example, the deviation of the data pair a1 using the first analysis method is "d1-1".

[0037] The deviation dk-j is defined as shown in the following formulas (1) to (4) using the occurrence probabilities "p1-j" to "p4-j".

[0038] d1-j={(p2-j)+(p3-j)+(p4-j)} / 3-(p1-j)…(1) d2-j={(p1-j)+(p3-j)+(p4-j)} / 3-(p2-j)…(2) d3-j={(p1-j)+(p2-j)+(p4-j)} / 3-(p3-j)…(3) d4-j={(p1-j)+(p2-j)+(p3-j)} / 3-(p4-j)…(4) As can be seen from the above formulas (1) to (4), the deviation of the occurrence probability pk-j by the kth analysis method is a numerical value indicating the difference between the occurrence probability by the kth analysis method and the average of the occurrence probabilities by the three analysis methods other than the kth analysis method. Therefore, the greater the difference between the occurrence probability calculated by one analysis method and the average of the occurrence probabilities calculated by the other three analysis methods, the greater the numerical value of the deviation of the occurrence probability calculated by the one analysis method.

[0039] The deviation calculation unit 14 executes a process of sorting the data pairs aj in descending order of the absolute value of the deviation dk-j calculated by the above formulas (1) to (4) for each analysis method. Fig. 6 is an explanatory diagram showing data in which the deviation d1-j of the occurrence probability p1-j calculated using the first analysis method is sorted in descending order. For example, the data pairs (combination of item 1 and item 2) are sorted in the order of the data pair of time series data of "vegetables / seaweed" and "sushi (eating out)", the data pair of time series data of "instant noodles" and "fresh food", and so on.

[0040] The visualization unit 15 visualizes each time series data included in the data pair aj and the deviation degree and presents it to the user. Specifically, the visualization unit 15 has a display unit (not shown) such as a display, and displays on the display unit a graph of a data pair whose deviation degree dk-j calculated by the deviation degree calculation unit 14 is greater than a certain value (for example, greater than 0.6). For example, the graphs and occurrence probability data shown in Figs. 7A and 7B are displayed on the display unit. That is, the visualization unit 15 visualizes only a predetermined number of analysis results whose deviation degrees calculated by the deviation degree calculation unit 14 are large.

[0041] Fig. 7A(a) shows a graph of data pair a11 of time series data q11 (e.g., household durable goods) and time series data q12 (e.g., furniture and household goods), and Fig. 7A(b) shows the occurrence probability of data pair a11 calculated using the first to fourth analysis methods. Fig. 7B(a) shows a graph of data pair a12 of time series data q13 (e.g., vegetables and seaweed) and time series data q14 (e.g., sushi (eating out)), and Fig. 7B(b) shows the occurrence probability of data pair a12 calculated using the first to fourth analysis methods.

[0042] The visualization unit 15 displays the data shown in Figures 7A and 7B on the display unit. The user can recognize the displayed information by looking at the display unit.

[0043] The input unit 16 accepts a judgment input of similarity or dissimilarity from the user. Specifically, the input unit 16 includes an operating device such as a keyboard, and accepts a judgment input of similarity or dissimilarity for the information displayed on the visualization unit 15. For example, as shown in FIG. 7A, the time series data q11 and q12 are separated, so the user inputs a judgment result of dissimilarity. On the other hand, as shown in FIG. 7B, the time series data q13 and q14 are close to each other, so the user inputs a judgment result of similarity.

[0044] The feature extraction unit 17 extracts the feature of the analysis method based on the judgment input and the deviation. Specifically, the feature extraction unit 17 extracts the feature of each analysis method based on the judgment input input by the input unit 16. For example, the time series data q11 and q12 of the data pair a11 shown in FIG. 7A(a) are graph-dissociated and have low similarity. Therefore, the occurrence probability of the data pair a11 should be a large value. As shown in FIG. 7A(b), the occurrence probability calculated by the third analysis method is a small value. The feature extraction unit 17 extracts the feature that the third analysis method is not suitable for the analysis of the time series data q11 and q12. That is, the feature extraction unit 17 extracts time series data that is not suitable for analysis by the analysis method as a feature of the analysis method.

[0045] In addition, the time series data q13 and q14 of the data pair a12 shown in FIG. 7B(a) are close to each other in graphs and have a high similarity. Therefore, the occurrence probability of the data pair a12 should be a small value. As shown in FIG. 7B(b), the occurrence probabilities calculated by the second, third, and fourth analysis methods are large values. The feature extraction unit 17 extracts a feature that the second, third, and fourth analysis methods are not suitable for analyzing the time series data q11 and q12. The feature extraction unit 17 is provided with a storage device (not shown) and stores the extracted features in the storage device.

[0046] The recording unit 18 records characteristic data of the time series data. For example, since it is recognized in advance that "vegetables and seaweed" are affected by seasonal changes, the recording unit 18 records this characteristic data. Furthermore, since it is recognized in advance that "automobile license fees" change in amount in a stepped manner, the recording unit 18 records this characteristic data. Furthermore, the visualization unit 15 described above may visualize the characteristics of the time series data in addition to each of the time series data constituting the data pair and the deviation degree.

[0047] Next, the operation of the feature extraction device 1 according to the first embodiment will be described with reference to the flowchart shown in Fig. 8. First, in step S11 in Fig. 8, the combination unit 11 generates data pairs aj by combining multiple time series data qi (i = 1 to m) stored in the database 2. When the number of time series data qi is m, "m*(m-1) / 2" data pairs are generated.

[0048] In step S12, the data analysis unit 12 calculates an analysis value by analyzing each data pair aj using a plurality of analysis methods. Specifically, the first analysis unit 21 calculates an analysis value for each data pair aj using the first analysis method. The second analysis unit 22 calculates an analysis value for each data pair aj using the second analysis method. The third analysis unit 23 calculates an analysis value for each data pair aj using the third analysis method. The fourth analysis unit 24 calculates an analysis value for each data pair aj using the fourth analysis method.

[0049] Furthermore, the data analysis unit 12 generates a distribution curve of the analytical values ​​calculated by each analysis method. Specifically, as shown in Fig. 4B(a) to (d), a distribution curve s1 of the analytical values ​​calculated by the first analysis method, a distribution curve s2 of the analytical values ​​calculated by the second analysis method, a distribution curve s3 of the analytical values ​​calculated by the third analysis method, and a distribution curve s4 of the analytical values ​​calculated by the fourth analysis method are generated.

[0050] In step S13, the occurrence probability calculation unit 13 generates a normalized curve by normalizing each of the distribution curves s1 to s4, for example, the normalized curve s11 shown in FIG.

[0051] In step S14, the deviation calculation unit 14 calculates the occurrence probability of each data pair aj based on the normalization curve s11. For example, as shown in Fig. 5, if the target data pair belongs to the top 30% of the total, the occurrence probability is set to "0.3". If the target data pair belongs to the top 70%, the occurrence probability is set to "0.7".

[0052] In step S15, the deviation calculation unit 14 calculates the deviation of each occurrence probability. Specifically, the occurrence probability of the data pair aj of each analysis method is calculated by the above-mentioned formulas (1) to (4). Furthermore, the deviation calculation unit 14 executes a process of sorting the data pairs aj in descending order of deviation. As a result, for example, as shown in FIG. 6, data is obtained in which the deviations d1-j of the occurrence probability p1-j calculated using the first analysis method are sorted in descending order.

[0053] For example, for the data pair "vegetables / seaweed" and "sushi (eating out)", the occurrence probability calculated using the first analysis method is "0.0473", while the occurrence probability calculated using the second to fourth analysis methods is approximately "1.0000". Therefore, the occurrence probability calculated using the first analysis method has a large difference from the occurrence probabilities calculated using the other three analysis methods, and the deviation is a high value of "0.926428".

[0054] In step S16, the visualization unit 15 displays on a display unit (not shown) a graph of the data pairs determined to have a large deviation d1-j (for example, data pairs of 0.6 or more) and the occurrence probability data. That is, the graph of the data pairs and the occurrence probability data are visualized. For example, the information shown in FIG. 7A and FIG. 7B are displayed on the screen.

[0055] By visually checking this screen, the user judges the validity of the analysis results by each analysis method. For example, in the graph shown in FIG. 7A(a), the two time series data q11 and q12 are not similar. Therefore, it is presumed that the occurrence probability will be a large value (a value close to "1"). In the data shown in FIG. 7A(b), the occurrence probability calculated by the first, second, and fourth analysis methods indicates a value close to "1", and the occurrence probability calculated by the third analysis method is a value "0.16" that deviates from the above three analysis methods. In this case, it is presumed that the analysis value using the third analysis method is inappropriate, and the analysis value using the first, second, and fourth analysis methods is appropriate.

[0056] On the other hand, in the graph shown in FIG. 7B(a), the two time series data q13 and q14 are similar. Therefore, it is presumed that the occurrence probability will be a small value (a value close to "0"). In the data shown in FIG. 7B(b), the occurrence probability calculated by the second, third, and fourth analysis methods is close to "1", and the occurrence probability calculated by the first analysis method is "0.05", a value that deviates from the above three analysis methods. In this case, it is presumed that the analysis values ​​obtained by the second, third, and fourth analysis methods are inappropriate, and the analysis value obtained by the first analysis method is appropriate.

[0057] Furthermore, the visualization unit 15 reads the characteristic data of each time series data recorded in the recording unit 18 and displays it on the display unit. For example, if the data pair to be analyzed includes time series data of "vegetables and seaweed", the characteristic data "affected by seasonal changes" is displayed on the display unit. Also, if the data pair to be analyzed includes time series data of "automobile license fees", the characteristic data "amount changes in a stepped manner" is displayed on the display unit. By visually checking this characteristic data, the user can use it as a reference for judging the analysis results.

[0058] In step S17, the input unit 16 accepts a judgment input of similarity or dissimilarity from the user. The user refers to the visualized information and inputs a judgment result of whether the analysis value obtained by each analysis method is appropriate or not. For example, in the example shown in FIG. 7A described above, the input unit 16 inputs a judgment result that the analysis value obtained by the third analysis method is inappropriate and that the analysis values ​​obtained by the first, second, and fourth analysis methods are appropriate. In the example shown in FIG. 7B described above, the input unit 16 inputs a judgment result that the analysis values ​​obtained by the second, third, and fourth analysis methods are inappropriate and that the analysis value obtained by the first analysis method is appropriate.

[0059] In other words, if there is a high degree of discrepancy between the occurrence probability calculated by analyzing the time series data using one analysis method and the occurrence probability calculated by analyzing the time series data using another analysis method, there is a high possibility that the one analysis method or the other analysis method is inappropriate as an analysis method to be used for analyzing the time series data. By acquiring a judgment input by the user, it becomes possible to recognize the characteristics of each analysis method (for example, the first analysis method is not suitable for analyzing the time series data a1) with high accuracy.

[0060] In step S18, the feature extraction unit 17 calculates a score according to the judgment result of appropriateness or inappropriateness based on the judgment input inputted by the input unit 16. Specifically, the score of the analysis method judged to be appropriate is set to "+1", and the score of the analysis method judged to be inappropriate is set to "-1". In the example shown in FIG. 7A, the score of the third analysis method is set to "-1", and the scores of the first, second, and fourth analysis methods are set to "+1". In the example shown in FIG. 7B, the scores of the second, third, and fourth analysis methods are set to "-1", and the score of the first analysis method is set to "+1". The feature extraction unit 17 accumulates the scores for each of the first to fourth analysis methods. Note that the numerical values ​​of the scores are not limited to "+1" and "-1", and may be numerical values ​​such as "+2", "+1", "-1", and "-2" according to the degree of "appropriateness" and "inappropriateness".

[0061] The feature extraction unit 17 extracts features of each analysis method based on the integrated score values ​​described above. For example, the feature extraction unit 17 extracts a feature that the analysis method with the highest score among the four analysis methods is suitable for analyzing the target time-series data. The feature extraction unit 17 records the extracted features in a storage device (not shown). Alternatively, the feature extraction unit 17 modifies the features already recorded in the storage device based on the extracted features.

[0062] In step S19, the data analysis unit 12 determines whether the first to fourth analysis methods need to be modified. For example, as shown in FIG. 7A, it is determined that the third analysis method is not suitable for analyzing the data pair a11, and in this case, it determines whether the third analysis method needs to be modified. If it is determined that modification is necessary (S19; YES), the process proceeds to step S20; otherwise (S19; NO), the process ends.

[0063] In step S20, the data analysis unit 12 corrects the target analysis method or determines it to be inappropriate. Then, the process ends. In this way, the characteristics of the analysis method for analyzing the similarity of time-series data can be extracted.

[0064] Thus, the feature extraction device 1 according to the first embodiment includes a combination unit 11 that generates multiple data pairs by combining two time series data, an analysis unit (data analysis unit 12) that analyzes the similarity between the two time series data included in each data pair using multiple analysis methods, an occurrence probability calculation unit 13 that calculates the occurrence probability of the similarity between each data pair for each analysis method based on the analysis results by the analysis unit, a deviation calculation unit 14 that calculates the deviation of the occurrence probability calculated for each analysis method for each data pair, a visualization unit 15 that visualizes each time series data included in the data pair and the deviation and presents them to a user, an input unit 16 that accepts a judgment input of similarity or dissimilarity from the user, and a feature extraction unit 17 that extracts features of the analysis method based on the judgment input and the deviation.

[0065] The feature extraction device 1 configured as described above can extract features that indicate which type of time series data an analysis method for analyzing time series data is suitable for or not suitable for. Therefore, when a user such as a data scientist analyzes time series data using a data analysis device, it is possible to support the user in selecting an appropriate analysis method from a plurality of analysis methods that the user has in stock.

[0066] Furthermore, the visualization unit 15 visualizes only a predetermined number of analysis results with large deviations calculated by the deviation calculation unit 14. For example, only analysis results with deviations of 0.6 or more are visualized. This makes it possible to omit visualization of analysis results with small deviations. In other words, the fact that all of the deviations of the four analysis methods are small means that the analysis values ​​obtained by the four analysis methods are almost the same numerical value, and it is considered that there is little need for user intervention. By visualizing only a predetermined number of analysis results with large deviations, it is possible to reduce the effort required by the user.

[0067] Furthermore, characteristic data of each time series data that has been recognized in advance is recorded in the recording unit 18, and by displaying this characteristic data on the display unit of the visualization unit 15, the user can use it as a reference when judging the suitability of each analysis method.

[0068] That is, as shown in Fig. 9, for a data pair a1 including time series data of "vegetables and seaweed", feature data that "is influenced by seasonal fluctuations" is recorded in the recording unit 18. Also, for a data pair a10 including time series data of "automobile license fees", feature data that "prices change in a step-like manner" is recorded in the recording unit 18. When analyzing the data pairs a1 and a10, the user can refer to these feature data to determine the features of each analysis method.

[0069] [Second embodiment] Next, a second embodiment will be described. Fig. 10 is a block diagram showing the configuration of a feature extraction device 1a according to the second embodiment and its peripheral devices. The second embodiment differs from the first embodiment in that a selection unit 19 is provided. Therefore, the components other than the selection unit 19 are given the same reference numerals and the description of the configuration will be omitted.

[0070] The selection unit 19 selects one data pair from among a plurality of data pairs when the occurrence probability of the similarity of the one data pair is close to the occurrence probability of the similarity of the other data pair, and the time series data included in the one data pair is identical to or similar to the time series data included in the other data pair.

[0071] That is, the selection unit 19 selects data pairs whose time series data are similar from among the data pairs generated by the combination unit 11. The visualization unit 15 visualizes the occurrence probabilities of the data pairs selected by the selection unit 19, excluding the occurrence probabilities.

[0072] Fig. 11A is a diagram showing a normalized distribution curve of an analysis result obtained by analyzing a plurality of data pairs by the first analysis method, and Fig. 11B is a diagram showing two time series data constituting a data pair and a deviation degree d1-j.

[0073] The data pairs x1, x2, and x3 shown in FIG. 11B all contain time series data of "university tuition fees." In addition, in FIG. 11A, the positions where the data pairs x1, x2, and x3 are plotted are close to each other. Therefore, two of these three data pairs x1, x2, and x3 are considered to be redundant and unnecessary. The selection unit 19 excludes the data pairs x2 and x3 from the analysis target.

[0074] The data pair x4 shown in FIG. 11B contains time series data for "Chinese noodles," and the data pair x5 contains time series data for "soba." In addition, in FIG. 11A, the positions where the data pairs x4 and x5 are plotted are close to each other. Therefore, one of these two data pairs x4 and x5 is considered to be redundant and unnecessary. The selection unit 19 excludes the data pair x5 from the analysis target.

[0075] In this way, in the feature extraction device 1a according to the second embodiment, data analysis is performed by excluding other data pairs that are similar to one data pair from multiple data pairs, thereby reducing the load required for the analysis processing of the data pairs.

[0076] That is, the selection unit 19 selects one data pair from among a plurality of data pairs when the occurrence probability of the similarity of the one data pair is close to the occurrence probability of the other data pair and the time series data constituting the one data pair is identical or similar to the time series data constituting the other data pair. Then, the visualization unit 15 displays the occurrence probability of the data pair selected by the selection unit 19 on the display unit, excluding the occurrence probability of the data pair. This makes it possible to avoid displaying unnecessary data and reduce the computation load.

[0077] The feature extraction device 1 of the present embodiment described above can be, for example, a general-purpose computer system including a CPU (Central Processing Unit, processor) 901, a memory 902, a storage 903 (HDD: Hard Disk Drive, SSD: Solid State Drive), a communication device 904, an input device 905, and an output device 906, as shown in Fig. 12. The memory 902 and the storage 903 are storage devices. In this computer system, the CPU 901 executes a predetermined program loaded on the memory 902, thereby realizing each function of the feature extraction device 1.

[0078] The feature extraction device 1 may be implemented in one computer, or in multiple computers, or may be a virtual machine implemented in a computer.

[0079] The program for the feature extraction device 1 can be stored in a computer-readable recording medium such as a HDD, SSD, USB (Universal Serial Bus) memory, CD (Compact Disc), or DVD (Digital Versatile Disc), or can be distributed via a network.

[0080] The present invention is not limited to the above-described embodiment, and various modifications are possible within the scope of the present invention. [Explanation of symbols]

[0081] 1, 1a Feature Extraction Device 2 Database 11 Combination section 12 Data Analysis Department (Analysis Department) 13. Occurrence Probability Calculation Section 14 Deviation calculation section 15 Visualization section 16 Input section 17 Feature Extraction Unit 18 Recording Section 19 Selection section 21 Analysis Department 1 22 Analysis Section 2 23 Analysis Section 3 24 Analysis Section 4

Claims

1. A combination unit that generates a plurality of data pairs by combining two pieces of time series data; an analysis unit that analyzes the similarity of two pieces of time series data included in each data pair by using a plurality of analysis methods; an occurrence probability calculation unit that calculates the occurrence probability of the similarity of each data pair for each analysis method based on the analysis result by the analysis unit; a deviation calculation unit that calculates a deviation of the occurrence probability calculated for each data pair for each analysis method; a visualization unit that visualizes each of the time series data included in the data pair and the deviation degree and presents the visualized data to a user; an input unit that accepts an input of a similarity / dissimilarity judgment by the user; a feature extraction unit that extracts features of the analysis method based on the judgment input and the deviation degree; A feature extraction device comprising:

2. The feature extraction device according to claim 1 , wherein the occurrence probability calculation unit calculates the occurrence probability by normalizing the analysis result of the analysis unit.

3. The visualization unit visualizes only a predetermined number of analysis results having a large degree of deviation calculated by the deviation calculation unit. The feature extraction device according to claim 1 or 2.

4. a selection unit that selects one of the plurality of data pairs when an occurrence probability of a similarity between the one data pair and an occurrence probability of a similarity between the other data pair are similar and the time series data included in the one data pair and the time series data included in the other data pair are identical or similar to each other, 4. The feature extraction device according to claim 1, wherein the visualization unit performs visualization excluding the occurrence probabilities of other data pairs selected by the selection unit.

5. A recording unit that records characteristic data of the time series data, The visualization unit visualizes characteristic data of the time series data in addition to each of the time series data included in the data pair and the degree of discrepancy. The feature extraction device according to any one of claims 1 to 4.

6. 6. The feature extraction device according to claim 1, wherein the feature extraction section extracts, as a feature of the analysis method, time-series data that is not suitable for analysis by the analysis method.

7. A feature extraction method executed by a feature extraction device including a combination unit, an analysis unit, an occurrence probability calculation unit, a deviation calculation unit, a visualization unit, an input unit, and a feature extraction unit, The combining unit generates a plurality of data pairs by combining two pieces of time series data; The analysis unit analyzes the similarity between two pieces of time series data included in each data pair by using a plurality of analysis methods; the occurrence probability calculation unit calculating an occurrence probability of the similarity of each data pair for each analysis method based on the analyzed similarity; a step of the deviation calculation unit calculating a deviation of the occurrence probability calculated for each data pair for each analysis method; the visualization unit visualizing each time-series data included in the data pair and the deviation degree and presenting the visualized data to a user; a step of receiving an input of a judgment of similarity or dissimilarity from the user by the input unit; the feature extraction unit extracting features of the analysis method based on the judgment input and the deviation degree; The feature extraction method includes:

8. A feature extraction program that causes a computer to function as the feature extraction device according to any one of claims 1 to 6.

Images