Information processing program, information processing device, and information processing method

Abstract

An information processing device (1): generates binary feature amount data (22) from feature amount data (21) obtained by, on a sample-by-sample basis, accumulating values of a plurality of feature amounts held by each sample, the binary feature amount data (22) being obtained by, on a sample-by-sample basis, binarizing the values of the plurality of feature amounts held by each sample on the basis of a preset item; uses the binary feature amount data (22) to list a set of items for which all sample sets indicate a true value; calculates, for each set of items, a correlation between a plurality of feature amounts in the feature amount data (21) of the set of samples associated with each set of items; and selects, as a condition to be used in a cause-and-effect search, a set of items for which a correlation is determined to exist. As a result of the foregoing, a condition for which a correlation appears can be accurately selected.

Description

[Technical field]

[0001] The present invention relates to an information processing program and the like. [Background technology]

[0002] In recent years, research has been conducted on efficiently narrowing down the number of conditions for causal exploration by extracting correlated conditions. Figure 8 is a reference diagram showing condition extraction for statistical causal exploration. As shown in Figure 8, this technology extracts all candidate conditions for correlated data from past data for learning by AI. Then, this technology extracts all conditions for causal data from the extracted candidate conditions for data. However, this technology searches for causal relationships for all candidate conditions for data, which is unrealistic in terms of the amount of calculation required.

[0003] Therefore, a technique has been disclosed that efficiently narrows down the number of conditions for which causal search should be performed by relaxing the search target of conditions from causal relationships to correlation relationships (for example, Non-Patent Document). FIG. 9 is a reference diagram showing a technique for discovering individual characteristic causal relationships. As shown in FIG. 9, this technique first uses an emergent pattern discovery technique to comprehensively find pairs of important factor candidates that have a strong correlation with the objective variable under specific conditions and the conditions at that time from a past sample set. Note that the past sample set is used after being binarized based on a threshold value.

[0004] Then, for each of the conditions found, causal search technology is used to determine whether the candidate important factors under those conditions are correct. For example, assume that there is a case where "x1 ∧ x3 ∧ x4 → y" (if x1 = x3 = x4 = 1, then y = 1). In such a case, one variable selected from the left side is the "candidate important factor" and the rest are the "conditions". Here, x4 is the "candidate important factor" and the remaining "x1 ∧ x3" is the "condition". With this technology, if there is a high correlation between the "candidate important factor" and y on the right side among the past sample set that satisfies the "condition", that "condition" is adopted. The conditions and important factors found in this way are stored in a DB (DataBase). Then, when applied, for a sample for which causal relationships are to be known, the conditions that the sample satisfies are selected from the DB, and the corresponding important factors are presented. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Yusuke Koyanagi and 4 others, “Development of a technology to discover individual characteristic causal relationships and its application to marketing data”, 18th Business Informatics Study Group of the Japanese Society for Artificial Intelligence, March 2021,<URL:http: / / sig-bi.jp / doc / 18thSIG-BI2021 / 18thSIG-BI2021 paper13.pdf> Summary of the Invention [Problem to be solved by the invention]

[0006] However, when the features included in the past sample set are numerical data, there is a problem that a pair of features that exhibits high correlation under "conditions" may or may not be detected depending on the binarization threshold setting. That is, when the binarization threshold setting is changed, the correlation between the pair of features under "conditions" changes. Here, the problem of the correlation between the pair of features under "conditions" changing when the binarization threshold setting is changed will be described with reference to Figs. 10A and 10B. Figs. 10A and 10B are diagrams for explaining the problem of the correlation between the pair of features changing depending on the binarization threshold setting.

[0007] 10A and 10B, the left diagram shows a graph plotting samples that satisfy the "condition" and samples that do not satisfy the "condition" by focusing on feature amount a and feature amount b. The white circles are samples that satisfy the "condition" and the black circles are samples that do not satisfy the "condition."

[0008] Under these circumstances, in Figure 10A, the threshold for binarizing samples of feature a is set to 3.5, and the threshold for binarizing samples of feature b is set to 6.3. The table data in the right diagram shows each sample binarized with these thresholds. When extracting samples that satisfy the "condition" from the binarization, since the correlation between feature a and feature b is high, a pair of feature a and feature b that shows a high correlation under the "condition" is detected. As a result, this "condition" is adopted as the condition for the correlation to appear.

[0009] In contrast, in Figure 10B, the threshold for binarizing samples of feature a is set to 5, and the threshold for binarizing samples of feature b is set to 4. The table data on the right shows each sample binarized with these thresholds. When extracting samples that satisfy the "condition" from the binarization, the correlation between feature a and feature b appears to be low, so the pair of feature a and feature b that shows a high correlation under the "condition" is not detected. As a result, this "condition" is not adopted as a condition for correlation to appear.

[0010] In this way, when the features contained in the past sample set are numerical data, depending on the binarization threshold setting, pairs of features that show a high correlation under certain "conditions" may or may not be detected.

[0011] In one aspect, the present invention aims to accurately select conditions under which a correlation appears. [Means for solving the problem]

[0012] In one aspect, an information processing program causes a computer to execute the following process: generate second data for each sample by binarizing the values ​​of multiple attributes possessed by each sample based on preset attribute conditions from first data that accumulates the values ​​of multiple attributes possessed by each sample; use the second data to enumerate sets of attribute conditions for which a set of samples all show true values; calculate, for each set of attribute conditions, the correlation between the multiple attributes in the first data for the set of samples associated with each set of attribute conditions; and select, as conditions for causal search, the set of attribute conditions that are determined to be correlated. Effect of the Invention

[0013] According to one embodiment, the conditions under which the correlation appears can be selected with high precision. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a functional block diagram illustrating a configuration of an information processing device according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a generation process according to the embodiment. [Diagram 3] FIG. 3 is a diagram illustrating an example of the enumeration process according to the embodiment. [Figure 4A] FIG. 4A is a diagram illustrating an example of a calculation process and a selection process according to the embodiment. [Figure 4B] FIG. 4B is a diagram illustrating an example of a calculation process and a selection process according to the embodiment. [Figure 4C]FIG. 4C is a diagram illustrating an example of a calculation process and a selection process according to the embodiment. [Diagram 5] FIG. 5 is a diagram illustrating an example of a flowchart of information processing according to the embodiment. [Figure 6] FIG. 6 is a diagram illustrating an effect of information processing according to the embodiment. [Figure 7] FIG. 7 is a diagram illustrating an example of a computer that executes an information processing program. [Figure 8] FIG. 8 is a reference diagram showing condition extraction for statistical causal exploration. [Figure 9] FIG. 9 is a reference diagram showing a technique for discovering individual characteristic causal relationships. [Figure 10A] FIG. 10A is a diagram illustrating a problem in which the correlation between pairs of feature amounts changes depending on the threshold setting for binarization. [Figure 10B] FIG. 10B is a diagram illustrating the problem that the correlation between pairs of feature amounts changes depending on the threshold setting for binarization. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an information processing program, an information processing device, and an information processing method disclosed in the present application will be described in detail with reference to the accompanying drawings. Note that the present invention is not limited to the embodiments. EXAMPLES

[0016] [Functional configuration of information processing device] FIG. 1 is a functional block diagram showing a configuration of an information processing device according to an embodiment. The information processing device 1 shown in FIG. 1 generates second data by binarizing the values ​​of multiple attributes of each sample based on items from first data in which the values ​​of multiple attributes of each sample are accumulated. Then, the information processing device 1 uses the second data to enumerate sets of items in which all sample sets indicate true values. Then, the information processing device 1 calculates correlation coefficients of multiple attribute pairs in the first data within the sets of samples corresponding to each set of items. Then, the information processing device 1 selects sets of items predicted to be correlated as conditions for causal search.

[0017] 1, the information processing device 1 includes a control unit 10 and a storage unit 20. The control unit 10 includes a generation unit 11, an enumeration unit 12, a calculation unit 13, and a selection unit 14.

[0018] The storage unit 20 stores various data. The storage unit 20 has feature amount data 21, binary feature amount data 22, and an item set 23.

[0019] The feature amount data 21 is table data that accumulates values ​​of a plurality of feature amounts (attributes) that each sample has. Each row of the feature amount data 21 corresponds to each sample. Each column of the feature amount data 21 corresponds to each feature amount (attribute).

[0020] The binary feature data 22 is table data in which the values ​​of multiple feature quantities (attributes) of each sample from the feature quantity data 21 are binarized based on predetermined items. The binarization is performed, for example, by comparing the magnitude of the original feature quantity value with a constant. Each row of the binary feature data 22 corresponds to each sample that is the same as the feature quantity data 21. Each column of the binary feature data 22 corresponds to each binary feature. The binary feature corresponds to each item for binarizing each feature quantity. For example, when the feature quantity is a, the item indicating the binary feature quantity indicates the condition for comparing the magnitude of a with a constant for binarizing the feature quantity.

[0021] The item set 23 is a set of items enumerated using the binary feature data 22, and is a set of items corresponding to a sample set for which the same binary feature (item) is true. The set of individual items included in the item set 23 must be a frequent item set. A frequent item set is a set of items with a number of samples equal to or greater than a certain size. The reason that the item set 23 is a frequent item set is that a number of samples equal to or greater than a certain size is required for statistical causal discovery.

[0022] In addition, it is preferable to use a saturated itemset as a representative of each set of items included in the itemset 23. A saturated itemset is a union of sets of items that have the same sample set. For example, if the sample set {1, 2, 3} has an item set {x3} and also has an item set {x2, x3}, the saturated itemset indicating {x2, x3}, which is the union of the item sets, is included as a representative in the itemset 23. The reason that the itemset 23 is a saturated itemset is that it is pointless to associate a plurality of item sets with the same sample set. The itemset 23 is generated by the enumeration unit 12.

[0023] The generating unit 11 generates binary feature data 22 from the feature data 21. For example, the generating unit 11 generates binary feature data 22 by binarizing the value of each feature into true and false based on a binary feature (item) from the feature data 21. Then, the generating unit 11 stores the generated binary feature data 22 in the storage unit 20.

[0024] The enumeration unit 12 uses the binary feature data 22 to enumerate frequent saturated item sets whose sample sets all indicate true values. For example, the enumeration unit 12 uses the binary feature data 22 to extract correspondence information that associates each set of items with a set of samples that all indicate true values. Each set of items is a frequent item set with a certain number of samples or more. In addition, each set of items is a saturated item set that is the union of sets of items that have the same sample set. In other words, each set of items is a frequent saturated item set. Then, the enumeration unit 12 stores the group of frequent saturated item sets in the storage unit 20 as an item set 23.

[0025] The enumeration unit 12 also generates a directed acyclic graph starting from an empty set for the enumerated frequent closed itemsets. Items are added to the directed acyclic graph from upstream to downstream. Note that frequent closed itemsets can be enumerated using the algorithm for enumerating frequent closed itemsets described in, for example, "Takeaki Uno, Tatsuya Asai, Yuzo Uchida, Hiroki Arimura, "An Efficient Algorithm for Enumerating Closed Patterns in Transaction Databases", Discovery Science 2004, LNAI 3245, pp.16-31".

[0026] The calculation unit 13 calculates the correlation between a plurality of features in the feature data 21 in a set of samples associated with each frequent saturated item set. For example, the calculation unit 13 selects frequent saturated item sets in order from an empty set in the directed acyclic graph toward downstream. The calculation unit 13 calculates the correlation coefficient of each feature pair in the feature data 21 in a sample set that satisfies the conditions of all items included in the selected frequent saturated item set. As an example, the calculation unit 13 extracts the value of each feature of a set of samples associated with the selected frequent saturated item set from the feature data 21. Then, the calculation unit 13 calculates the correlation coefficient of each feature pair using the value of the extracted feature.

[0027] The selection unit 14 selects a frequent saturated item set that is predicted to be correlated as a condition for causal search. For example, when there are a certain number or more feature pairs with a correlation coefficient equal to or greater than a threshold set for the selected frequent saturated item set, the selection unit 14 determines that there is a correlation and selects the selected frequent saturated item set as a condition. The certain number may be "1" or "2" and may be set in advance.

[0028] The threshold is set for each feature pair. The threshold is updated so that there are feature pairs whose correlation increases by a certain amount or more when an item is added from upstream to downstream. The threshold may also be updated so that a set of feature pairs whose correlation increases by a certain amount or more when an item is added is different from others.

[0029] [An example of generation processing] Here, the generation process performed by the generation unit 11 according to the embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the generation process according to the embodiment. As shown in Fig. 2, in the left diagram, feature amount data Y is represented as feature amount data 21. In the feature amount data Y, a, b, c, ... are set as feature amounts, and a value of the feature amount is set for each sample.

[0030] From such feature data Y, the generating unit 11 generates binary feature data 22. That is, the generating unit 11 generates binary feature data 22 by binarizing the value of each feature to true or false based on a binary feature (item) set in advance from the feature data Y. In the right diagram, binary feature data X is represented as the binary feature data 22. In the binary feature data X, x1, x2, x3,... are set as binary features (items), and "1" and "0" are set for each sample. The item x1 indicates a condition that the feature a is smaller than 5. The item x2 indicates a condition that the feature a is greater than 2. The item x3 indicates a condition that the feature c is 0. For x1, x2, and x3, "1" is set when the condition is satisfied, and "0" is set when the condition is not satisfied.

[0031] As an example, when the sample is "1", for item x1 (a<5), the feature a is "1.3" and is less than "5", so "1" is set. For item x2 (a≧2), the feature a is "1.3" and is less than "2", so "0" is set. For item x3 (c=0), the feature c is "1", and is not "0", so "0" is set.

[0032] [An example of enumeration processing] Next, the enumeration process performed by the enumeration unit 12 according to the embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the enumeration process according to the embodiment. As shown in Fig. 3, binary feature data X is represented as binary feature data 22 in the left diagram.

[0033] Using such binary feature data X, the enumeration unit 12 extracts a set of items whose sample sets all indicate true values. Here, the set of items to be extracted has three or more samples. The enumeration unit 12 extracts the following sets of items: When the set of samples is {1, 2, 3, 4, 5, 6, 7}, an empty set {} is extracted; When the set of samples is {1, 2, 3, 4}, {x1} is extracted; When the set of samples is {2, 3, 4, 5, 6}, {x2} is extracted; When the set of samples is {2, 3, 4, 5, 6}, {x2, x3} is extracted; When the set of samples is {2, 3, 4, 5}, {x1, x2} is extracted; When the set of samples is {2, 3, 4}, {x1, x3} is extracted; When the set of samples is {2, 3, 4}, {x1, x3} is extracted. {x1,x2,x3} is extracted when the sample set is {2,3,4}.

[0034] Then, the enumeration unit 12 enumerates frequent saturated item sets from the extracted set of items. Here, the set of items {x1, x2}, {x1, x3}, and {x1, x2, x3} indicated by the symbol i1 have the same sample set {2, 3, 4}, so {x1, x2, x3} is listed as a frequent saturated item set. Also, the set of items {x3} and {x2, x3} indicated by the symbol i2 have the same sample set {2, 3, 4, 5}, so {x2, x3} is listed as a frequent saturated item set. Note that the empty set {}, {x1}, and {x2} are also listed as frequent saturated item sets.

[0035] Then, the enumeration unit 12 generates a directed acyclic graph starting from the empty set for the enumerated frequent closed itemsets. The directed acyclic graph for these frequent closed itemsets is shown in the right figure.

[0036] [Example of calculation and selection processing] Next, the calculation process and selection process performed by the calculation unit 13 and the selection unit 14 according to the embodiment will be described with reference to Fig. 4A to Fig. 4C. Fig. 4A to Fig. 4C are diagrams showing an example of the calculation process and selection process according to the embodiment. In Fig. 4A to Fig. 4C, the correlation coefficient of each feature pair in the feature data 21 for the features a, b, c, and d is calculated.

[0037] As shown in Fig. 4A, the upper left diagram shows the initial values ​​of the thresholds for positive correlation and negative correlation for each feature pair for feature amounts a, b, c, and d. Here, the initial value of the threshold θ for positive correlation is 0.4. The initial value of the threshold θ for negative correlation is -0.4.

[0038] Under such circumstances, the calculation unit 13 selects a set of items indicating the starting point from the directed acyclic graph for the frequent closed item set. Then, the calculation unit 13 calculates the correlation coefficient of each feature pair in the feature data 21 in a sample set that satisfies the condition of the empty set {}, which is the set of selected items. The sample set that satisfies the condition of the empty set {} here means the entire sample. Here, the correlation coefficient of each feature pair when the set of items is the empty set {} is shown in the right figure. In the correlation coefficients of each feature pair in the right figure, the positive correlation coefficient of the feature a, b pair is 0.7, which is greater than the threshold value of 0.4. Also, the negative correlation coefficient of the feature d, b pair is -0.5, which is less than the threshold value of -0.4.

[0039] Therefore, since there exists a feature pair having a correlation coefficient equal to or greater than the set threshold for positive correlation, the selection unit 14 adopts (selects) an empty set {}, which is a set of selected items, as a condition. Also, since there exists a feature pair having a correlation coefficient equal to or less than the set threshold for negative correlation, the selection unit 14 can adopt (select) an empty set {}, which is a set of selected items, as a condition.

[0040] Then, the selection unit 14 updates the threshold value used downstream of the empty set {}, which is a set of items. That is, the selection unit 14 updates the threshold value in order to find a condition that is better than the set of items already adopted (selected) as a condition. The lower left diagram shows the updated threshold value. Here, the feature pair with high correlation when the set of items is the empty set {} was the feature a, b pair. Therefore, the threshold value of the positive correlation of the feature a, b pair is updated to be increased by a certain amount δ (=0.2). That is, the threshold value of the positive correlation of the feature a, b pair is updated to 0.9. That is, such a threshold value is updated so that there is a feature pair whose correlation increases by a certain amount δ (=0.2) or more due to the addition of an item when moving downstream from the empty set {}. Also, the feature pair with high correlation when the set of items is the empty set {} was the feature d, b pair. Therefore, the threshold value of the negative correlation of the feature d, b pair is updated to be decreased by a certain amount δ (=0.2). That is, the threshold for the negative correlation of the feature d, b pair is updated to -0.7. That is, the threshold is updated so that there is a feature pair whose correlation increases by a certain amount δ (=0.2) or more by adding an item from the empty set {} when moving downstream. Here, the positive correlation coefficient of the feature c, d pair is 0.3, which is lower than the threshold 0.4, but in order to find a better set of items, the threshold for the positive correlation of the feature c, d pair is updated to increase by a certain amount δ (=0.2). That is, the threshold for the positive correlation of the feature c, d pair is updated to 0.5. That is, the threshold is updated so that the set of feature pairs whose correlation increases by a certain amount or more by adding an item from the empty set {} when moving downstream is different from the others.

[0041] As shown in Figure 4B, the thresholds updated when moving downstream from the empty set {} are set in the left diagram. Here, the threshold for the positive correlation of the feature a, b pair is updated to 0.9. The threshold for the negative correlation of the feature d, b pair is updated to -0.7. Also, the threshold for the positive correlation of the feature c, d pair is updated to 0.5.

[0042] Under such circumstances, the calculation unit 13 selects the next set of items from the directed acyclic graph for the frequent closed item set. Here, it is assumed that {x1} is selected as the set of items downstream of the empty set {}. Then, the calculation unit 13 calculates the correlation coefficient of each feature pair in the feature data 21 in the sample set that satisfies the condition of the selected set of items {x1}. For example, the correlation coefficient of each feature pair when the set of items is {x1} is shown in the right figure. In the correlation coefficients of each feature pair in the right figure, the positive correlation coefficients of all feature pairs are lower than the threshold updated when moving downstream from the empty set {}. The negative correlation coefficients of all feature pairs are higher than the threshold updated when moving downstream from the empty set {}.

[0043] Therefore, the selection unit 14 does not adopt (select) the item set {x1} as a condition. In other words, the sample set that satisfies the item set {x1} is a subset of the sample set that satisfies the empty set {}, and is a similar sample set, and there is no feature pair whose correlation increases by a certain amount δ or more. Therefore, the item set {x1} is not adopted (selected) as a condition.

[0044] As shown in Figure 4C, the upper left diagram shows the updated thresholds when moving downstream from the empty set {}. The updated thresholds are the same as those shown in the left diagram of Figure 4B. That is, the threshold for the positive correlation of the feature a, b pair is updated to 0.9. The threshold for the negative correlation of the feature d, b pair is updated to -0.7. Moreover, the threshold for the positive correlation of the feature c, d pair is updated to 0.5.

[0045] Under such circumstances, the calculation unit 13 selects the next set of items from the directed acyclic graph for the frequent closed item set. Here, it is assumed that {x2} is selected as the set of items downstream of the empty set {}. Then, the calculation unit 13 calculates the correlation coefficient of each feature pair in the feature data 21 in the sample set that satisfies the condition of the selected item set {x2}. For example, the correlation coefficient of each feature pair when the item set is {x2} is shown in the right diagram. In the correlation coefficient of each feature pair in the right diagram, the positive correlation coefficient of the feature a, d pair is 0.7, which is greater than the threshold value of 0.4.

[0046] Therefore, since there is a feature pair having a correlation coefficient equal to or greater than a set threshold for positive correlation, the selection unit 14 adopts (selects) the set of selected items {x2} as a condition.

[0047] Then, the selection unit 14 updates the threshold used downstream of the set of items {x2}. That is, the selection unit 14 updates the threshold to find a condition that is better than the set of items already adopted (selected) as a condition. The updated threshold is shown in the lower left diagram. Here, the feature pair with high correlation when the set of items is {x2} was the feature a, d pair. Therefore, the threshold of the positive correlation of the feature a, d pair is updated to increase by a certain amount δ (=0.2). That is, the threshold of the positive correlation of the feature a, d pair is updated to 0.9. That is, such a threshold is updated so that there is a feature pair whose correlation increases by a certain amount or more by adding an item from upstream to downstream. And the threshold of the other feature pairs shows the threshold updated with the empty set {} shown in the upper left diagram. In this example, the negative correlation coefficient of the feature d, b pair was -0.6, which was lower than the threshold of -0.7, but in order to find a better set of items, the threshold of the negative correlation of the feature d, b pair has been updated to be lower by a certain amount δ (=0.2). In other words, the threshold of the negative correlation of the feature d, b pair has been updated to -0.8. In other words, the threshold has been updated so that the set of feature pairs whose correlation increases by a certain amount or more due to the addition of items from upstream to downstream is different from the others.

[0048] [Information processing flow chart] FIG. 5 is a diagram illustrating an example of a flowchart of information processing according to the embodiment.

[0049] As shown in FIG. 5, the generating unit 11 binarizes the feature amount data Y to generate binary feature amount data X (step S11).

[0050] Then, the enumeration unit 12 enumerates frequent closed itemsets in the binary feature data X, and generates a directed acyclic graph with the empty set {} as a starting point for the enumerated frequent closed itemsets (step S12).

[0051] Then, the calculation unit 13 selects a frequent closed item set I from the empty set {} in a depth-first or breadth-first manner (step S13). The calculation unit 13 calculates the correlation coefficient of each feature pair in the feature data Y in the sample set that satisfies the condition of the frequent closed item set I (step S14).

[0052] If there is a feature pair with a correlation coefficient equal to or greater than the threshold set upstream of the frequent saturated item set I, the selection unit 14 selects and outputs the selected frequent saturated item set I as a condition (step S15). Note that if the frequent saturated item set I is an empty set at the starting point, the threshold used in the frequent saturated item set I is set in advance.

[0053] Then, the selection unit 14 sets a threshold to be used downstream of the selected frequent closed item set I (step S16). For example, the selection unit 14 updates the threshold so that the threshold of the feature pair having a correlation equal to or greater than the threshold is increased by a certain amount. The selection unit 14 may also update the threshold so that the threshold of the feature pair different from the feature pair having a correlation equal to or greater than the threshold is increased by a certain amount.

[0054] Then, the calculation unit 13 moves to step S13 to repeat the process until there are no unselected frequent saturated item sets (step S17). When there are no unselected frequent saturated item sets, the calculation unit 13 ends the information processing.

[0055] In this manner, the information processing according to the embodiment can accurately select the condition under which the correlation appears.

[0056] FIG. 6 is a diagram showing the effect of the information processing according to the embodiment. The left diagram shows a graph plotting samples that satisfy the "condition" and samples that do not satisfy the "condition" by focusing on the feature amount a and the feature amount b. The white circles are samples that satisfy the "condition", and the black circles are samples that do not satisfy the "condition". In the information processing according to the embodiment, the feature amount pairs are not evaluated from information binarized by threshold setting, but from information indicating the values ​​of the original feature amount, so that the feature amount pairs that show a high correlation under the "condition" are not overlooked. Therefore, the information processing according to the embodiment can accurately select the "condition" where the correlation appears based on the correlation coefficient between the feature amounts calculated from the values ​​of the original feature amount.

[0057] [Effects of the embodiment] According to the above embodiment, the information processing device 1 generates binary feature data 22 by binarizing the values ​​of multiple feature quantities possessed by each sample based on preset items from feature quantity data 21 in which the values ​​of multiple feature quantities possessed by each sample are accumulated for each sample. The information processing device 1 uses the binary feature data 22 to enumerate sets of items in which the sets of samples all show true values. Then, for each set of items, the information processing device 1 calculates the correlation between multiple feature quantities in the feature quantity data 21 in the sets of samples associated with each set of items. Then, the information processing device 1 selects a set of items determined to have a correlation as a condition for causal search. This allows the information processing device 1 to accurately select a condition in which a correlation appears.

[0058] Furthermore, according to the above embodiment, the information processing device 1 enumerates a set of items that indicates a union of sets of items that have the same set of samples. This allows the information processing device 1 to suppress extraction of a plurality of sets of items that have the same set of samples.

[0059] Furthermore, according to the above embodiment, the information processing device 1 further enumerates sets of items for sample sets that include a predetermined number or more of samples, thereby enabling the information processing device 1 to increase the accuracy of selecting conditions by selecting a sample set that includes a predetermined number or more of samples.

[0060] Furthermore, according to the above embodiment, the information processing device 1 selects a set of items in order, and calculates the correlation coefficient of a plurality of pairs of feature quantities in the feature quantity data 21 for a set of samples associated with the selected set of items. When a pair of feature quantities having a correlation coefficient equal to or greater than a predetermined threshold exists, the information processing device 1 selects the selected set of items as a condition. In this way, the information processing device 1 can avoid overlooking a pair of feature quantities that exhibits a high correlation under a condition by using the original feature quantity values ​​in the set of samples to evaluate the correlation coefficient.

[0061] Also, according to the above embodiment, the information processing device 1 uses the enumerated item sets to generate a directed acyclic graph starting from the empty set of items. The information processing device 1 selects item sets in order from the empty set included in the directed acyclic graph in a depth-first or breadth-first manner, and calculates a correlation coefficient for the selected item sets. Then, when there is a pair of features having a correlation coefficient equal to or greater than a threshold set at a higher level than the selected item set, the information processing device 1 selects the selected item set as a condition. This allows the information processing device 1 to select a better item set as a condition by using the threshold set at a higher level.

[0062] Note that each component of the illustrated information processing device 1 does not necessarily have to be physically configured as illustrated. In other words, the specific form of distribution / integration of the information processing device 1 is not limited to that shown in the figure, and all or part of it can be functionally or physically distributed / integrated in any unit according to various loads, usage conditions, etc. For example, the calculation unit 13 and the selection unit 14 may be integrated. Also, the storage unit 20 may be connected as an external device of the information processing device 1 via a network.

[0063] Moreover, the various processes described in the above embodiments can be realized by executing a prepared program on a computer such as a personal computer or a workstation. Therefore, an example of a computer that executes an information processing program that realizes the same functions as the information processing device 1 shown in Fig. 1 will be described below. Here, an information processing program that realizes the same functions as the information processing device 1 will be described as an example. Fig. 7 is a diagram showing an example of a computer that executes an information processing program.

[0064] 7, the computer 200 includes a CPU 203 that executes various arithmetic processes, an input device 215 that accepts data input from a user, and a display control unit 207 that controls a display device 209. The computer 200 also includes a drive device 213 that reads programs and the like from a storage medium, and a communication control unit 217 that transmits and receives data to and from other computers via a network. The computer 200 also includes a memory 201 that temporarily stores various information, and an HDD (Hard Disk Drive) 205. The memory 201, the CPU 203, the HDD 205, the display control unit 207, the drive device 213, the input device 215, and the communication control unit 217 are connected by a bus 219.

[0065] The drive device 213 is, for example, a device for the removable disk 210. The HDD 205 stores an information processing program 205a and information processing related information 205b.

[0066] The CPU 203 reads out the information processing program 205a, expands it in the memory 201, and executes it as a process. Such a process corresponds to each functional unit of the information processing device 1. The information processing related information 205b corresponds to the feature data 21, the binary feature data 22, and the item set 23. For example, a removable disk 210 stores each piece of information such as the information processing program 205a.

[0067] It should be noted that the information processing program 205a does not necessarily have to be stored in the HDD 205 from the beginning. For example, the program may be stored in a "portable physical medium" such as a flexible disk (FD), a CD-ROM (Compact Disk Read Only Memory), a DVD (Digital Versatile Disk), a magneto-optical disk, or an IC (Integrated Circuit) card that is inserted into the computer 200. Then, the computer 200 may read out the information processing program 205a from these and execute it. [Explanation of symbols]

[0068] 1. Information processing device 10 Control section 11 Generation part 12 Enumerations 13 Calculation section 14 Selection section 20 Memory section 21 Feature Data 22 Binary feature data 23 Item Set

Claims

1. generating second data by binarizing the values ​​of the multiple attributes of each sample based on a preset attribute condition from the first data in which the values ​​of the multiple attributes of each sample are accumulated for each sample; Using the second data, enumerate a set of attribute conditions for which the set of samples all exhibit true values; For each set of attribute conditions, calculate a correlation between a plurality of attributes in the first data in a set of samples associated with each set of attribute conditions; A set of attribute conditions that are determined to be correlated is selected as the conditions for causal exploration. An information processing program that causes a computer to execute processing.

2. The process of enumerating a set of attribute conditions includes enumerating a set of attribute conditions that indicates a union of sets of attribute conditions having the same set of samples.

2. The information processing program according to claim 1,

3. The process of enumerating the set of attribute conditions further includes enumerating the set of attribute conditions for a set of samples that includes a predetermined number or more of samples.

3. The information processing program according to claim 2,

4. The process of calculating the correlation includes: Selecting the set of attribute conditions in order, and calculating correlation coefficients of pairs of attributes in the first data for a set of samples associated with the selected set of attribute conditions; The process of selecting the set of attribute conditions includes selecting the set of attribute conditions as the condition when there is a pair of attributes having a correlation coefficient equal to or greater than a predetermined threshold.

2. The information processing program according to claim 1,

5. The process of enumerating the set of attribute conditions includes generating a directed acyclic graph using the enumerated set of attribute conditions, the graph starting from an empty set of the set of attribute conditions; The process of calculating the correlation includes selecting a set of attribute conditions in a depth-first or breadth-first manner from an empty set included in the directed acyclic graph, calculating a correlation coefficient for the selected set of attribute conditions, The process of selecting the set of attribute conditions includes selecting the set of attribute conditions as the condition when there is a pair of attributes having a correlation coefficient equal to or greater than a threshold value set at a higher level in the selected set of attribute conditions.

5. The information processing program according to claim 4.

6. a generating unit that generates second data by binarizing the values ​​of a plurality of attributes possessed by each sample based on a preset attribute condition from first data in which values ​​of a plurality of attributes possessed by each sample are accumulated for each sample; an enumeration unit that uses the second data to enumerate a set of attribute conditions for which a set of samples all indicate true values; a calculation unit that calculates, for each set of attribute conditions, a correlation between a plurality of attributes in the first data in a set of samples associated with each set of attribute conditions; a selection unit that selects a set of attribute conditions that are determined to be correlated as conditions for which causal search should be performed; 13. An information processing device comprising:

7. generating second data by binarizing the values ​​of the multiple attributes of each sample based on a preset attribute condition from the first data in which the values ​​of the multiple attributes of each sample are accumulated for each sample; Using the second data, enumerate a set of attribute conditions for which the set of samples all exhibit true values; For each set of attribute conditions, calculate a correlation between a plurality of attributes in the first data in a set of samples associated with each set of attribute conditions; A set of attribute conditions that are determined to be correlated is selected as the conditions for causal exploration. An information processing method in which processing is performed by a computer.

Images