Pattern search program, pattern search device, and pattern search method

Abstract

The present invention provides a pattern search program, a pattern search device, and a pattern search method. Priorities of attributes included in multiple attribute patterns are determined based on an inclusion relationship among occurrence sets of the attributes included in multiple attribute patterns in multiple samples, wherein the multiple attribute patterns each represent one or more attributes. Whether each of the multiple attribute patterns is a revealing pattern is determined based on a search order based on the determined priorities. During the determination process, if the frequency of occurrence of a first attribute pattern in multiple samples is greater than the frequency of occurrence of a second attribute pattern in multiple samples, determination regarding the first attribute pattern is not performed, wherein the second attribute pattern includes all attributes included in the first attribute pattern except the attribute with the lowest priority.

Description

Technical Field

[0001] The present invention relates to a pattern search program, a pattern search device and a pattern search method. Background Art

[0002] In recent years, data mining, which aims to discover new insights by analyzing collected data, has been actively promoted. For example, in such data mining, new insights are discovered by extracting data that meets specified conditions from multiple data sets each having attributes and identifying characteristic trends shared by the extracted data.

[0003] Specifically, in data mining such as that described above, new insights are discovered, for example, by extracting attribute patterns (hereinafter also referred to as emerging patterns) that appear more frequently in data that meet the target conditions and appear less frequently in data that do not meet the target conditions (for example, refer to non-patent document 1).

[0004] Non-patent document 1: AMGarcia-Vico et al., "An overview of emerging patternmining in supervised descriptive rule discovery: taxonomy, empirical study, trends, and prospects", WIREs Data Mining and Knowledge Discovery 2018,8:e1231.doi:10.1002 / widm.1231

[0005] In data mining as described above, the time from data collection to analysis may be constrained, depending on the intended use of the analysis results. Furthermore, the amount of data analyzed in data mining is increasing year by year. Therefore, data mining as described above needs to be performed as efficiently as possible.

[0006] However, extracting patterns in data mining is generally computationally expensive, and the process from data collection to analysis can take a long time. Therefore, companies and research institutions that perform data analysis need to reduce the computational costs required to extract patterns. Summary of the Invention

[0007] Therefore, in one aspect, an object of the present invention is to provide a pattern search program, a pattern search device, and a pattern search method that can reduce the computational processing cost required to extract an exposure pattern.

[0008] In one embodiment, a computer is caused to perform the following processing: determining the priority of each attribute included in the multiple attribute patterns based on the inclusion relationship of the occurrence sets of each attribute included in the multiple attribute patterns in multiple samples, wherein the multiple attribute patterns respectively represent more than one attribute; and determining whether each attribute pattern in the multiple attribute patterns is a revealing pattern according to a search order based on the determined priorities. In the process of performing the above-mentioned determination, if the occurrence frequency of the first attribute pattern in the multiple samples is greater than the occurrence frequency of the second attribute pattern in the multiple samples, the above-mentioned determination regarding the first attribute pattern is not performed, wherein the second attribute pattern has all the attributes included in the first attribute pattern except the attribute with the lowest priority.

[0009] According to one aspect, the computational processing cost required to extract the exposure pattern can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a diagram illustrating the configuration of the information processing system 10 .

[0011] Figure 2 This is a diagram illustrating the hardware configuration of the information processing device 1 .

[0012] Figure 3 This is a functional block diagram of the information processing device 1 .

[0013] Figure 4 This is a flowchart illustrating the pattern search process in the first embodiment.

[0014] Figure 5 This is a flowchart illustrating the pattern search process in the first embodiment.

[0015] Figure 6 This is a flowchart illustrating the pattern search process in the first embodiment.

[0016] Figure 7 This is a flowchart illustrating the pattern search process in the first embodiment.

[0017] Figure 8 This is a flowchart illustrating the pattern search process in the first embodiment.

[0018] Figure 9 This is a flowchart illustrating the pattern search process in the first embodiment.

[0019] Figure 10 This is a flowchart illustrating the pattern search process in the first embodiment.

[0020] Figure 11 This is a flowchart illustrating the pattern search process in the first embodiment.

[0021] Figure 12 This is a flowchart illustrating the pattern search process in the first embodiment.

[0022] Figure 13 This is a diagram illustrating a specific example of the data set D.

[0023] Figure 14 4 is a diagram illustrating a specific example of a table including an appearance set and an appearance frequency of attribute patterns in a data set D.

[0024] Figure 15 This is a diagram for explaining a specific example of the processing of S12 and S13.

[0025] Figure 16 Yes Figures 6 to 8 A diagram illustrating a specific example of the mode determination processing described in .

[0026] Figure 17 Yes Figures 6 to 8 A diagram illustrating a specific example of the mode determination processing described in .

[0027] Figure 18 This is a diagram illustrating a specific example of the data set D.

[0028] Figure 19 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in .

[0029] Figure 20 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in .

[0030] Figure 21 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in .

[0031] Figure 22 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in .

[0032] Figure 23 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in .

[0033] Figure 24 Yes Figures 9 to 12 A diagram illustrating a specific example of other mode determination processing described in . DETAILED DESCRIPTION

[0034] [Structure of information processing system]

[0035] First, the configuration of the information processing system 10 will be described. Figure 1 This is a diagram illustrating the configuration of the information processing system 10 .

[0036] like Figure 1 As shown, the information processing system 10 includes, for example, an information processing device 1 and an operation terminal 2. The information processing device 1 is connected to the operation terminal 2 via a network NW such as the Internet.

[0037] exist Figure 1 In the example shown, the information processing device 1 determines whether each of the attribute patterns included in a plurality of data (hereinafter also referred to as a plurality of samples) input via the operation terminal 2 is a revealing pattern.

[0038] Specifically, the information processing device 1 in this embodiment determines the priority of each attribute included in a plurality of attribute patterns based on the inclusion relationship of the occurrence sets of each attribute included in the plurality of attribute patterns in a plurality of samples, wherein the plurality of attribute patterns each represent one or more attributes. Furthermore, the information processing device 1 determines whether each attribute pattern in the plurality of attribute patterns is an emerging pattern based on a search order based on the determined priority. It should be noted that an emerging pattern in this embodiment refers to an attribute pattern that appears more frequently in data that meets the target condition and appears less frequently in data that does not meet the target condition.

[0039] Here, whenever determining whether each attribute pattern is a revealing pattern (hereinafter, simply referred to as determination), information processing device 1 determines whether the frequency of occurrence of the attribute pattern (hereinafter, also referred to as the first attribute pattern) in multiple samples is greater than the frequency of occurrence of an attribute pattern (hereinafter, also referred to as the second attribute pattern) that includes all attributes included in the first attribute pattern except the attribute with the lowest priority in the multiple samples. Furthermore, if the frequency of occurrence of the first attribute pattern is greater than the frequency of occurrence of the second attribute pattern, information processing device 1 does not determine the first attribute pattern.

[0040] That is, if the frequency of occurrence of the first attribute pattern is not lower than the frequency of occurrence of the second attribute pattern, information processing device 1 in this embodiment determines that the first attribute pattern is not a revelation pattern that does not include other revelation patterns (hereinafter also referred to as a minimal revelation pattern). Therefore, if the processing purpose is to extract minimal revelation patterns, information processing device 1 omits the determination of the first attribute pattern based on the detection that the frequency of occurrence of the first attribute pattern is not lower than the frequency of occurrence of the second attribute pattern.

[0041] Furthermore, in order to create a plurality of first attribute patterns that can omit execution of determination as described above, the information processing device 1 in this embodiment pre-ranks the attributes included in the plurality of attribute patterns (determines the priority of each attribute).

[0042] Thus, the information processing device 1 can reduce the computational processing cost required to extract the exposure pattern. Specifically, the information processing device 1 can reduce the computational processing cost required to extract an extremely small exposure pattern.

[0043] [Hardware structure of information processing system]

[0044] Next, the hardware configuration of the information processing system 10 will be described. Figure 2 This is a diagram illustrating the hardware configuration of the information processing device 1 .

[0045] like Figure 2 As shown, the information processing device 1 includes a CPU 101 as a processor, a memory 102 , a communication device 103 , and a storage medium 104 . These components are connected to each other via a bus 105 .

[0046] The storage medium 104 includes, for example, a program storage area (not shown) that stores a program 110 for performing a process for searching for a revealed pattern (hereinafter, also referred to as a pattern search process). Furthermore, the storage medium 104 includes, for example, a storage unit 130 (hereinafter, also referred to as an information storage area 130) that stores information used in the pattern search process. Furthermore, the storage medium 204 may be, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0047] The CPU 101 executes the program 110 loaded from the storage medium 104 to the memory 102 to perform pattern search processing.

[0048] In addition, the communication device 103 communicates with, for example, the network NW.

[0049] [Functions of information processing systems]

[0050] Next, the functions of the information processing system 10 will be described. Figure 3 This is a functional block diagram of the information processing device 1 .

[0051] like Figure 3 As shown, the information processing device 1 realizes various functions including an information management unit 111, an information receiving unit 112, a ranking determination unit 113, a mode determination unit 114 and an information sending unit 115 by organically cooperating with hardware such as a CPU 101 and a memory 102 and a program 110.

[0052] In addition, if Figure 3 As shown, the information processing device 1 stores, for example, a data set D including a plurality of samples, graph information 131 , ranking information 132 , and pattern information 133 in the information storage area 130 .

[0053] The information management unit 111 stores, for example, the data set D input via the operation terminal 2 in the information storage area 130 .

[0054] The information receiving unit 112 receives, for example, an instruction to start a search for content in the reveal mode (hereinafter also referred to as a search instruction). Specifically, the information receiving unit 112 receives, for example, a search instruction input via the operation terminal 2 .

[0055] The ranking determination unit 113 determines the priority of each attribute included in the plurality of attribute patterns based on the inclusion relationship between the occurrence sets of each attribute included in the plurality of samples. Furthermore, the information management unit 111 stores, for example, ranking information 132 indicating the priority of each attribute determined by the ranking determination unit 113 in the information storage area 130.

[0056] Pattern determination unit 114 determines whether each of the multiple attribute patterns is a revealing pattern based on the search order determined by ranking determination unit 113. In this case, pattern determination unit 114 determines whether the frequency of occurrence of the first attribute pattern in the multiple samples is lower than the frequency of occurrence of the second attribute pattern in the multiple samples. If it is determined that the frequency of occurrence of the first attribute pattern is lower than the frequency of occurrence of the second attribute pattern, pattern determination unit 114 does not determine the first attribute pattern.

[0057] Then, the mode determination unit 114 specifies the minimal exposure mode among the attribute modes determined to be the exposure mode, and the information management unit 111 stores the mode information 133 indicating the attribute mode determined to be the minimal exposure mode by the mode determination unit 114 in the information storage area 130 .

[0058] The information transmitting unit 115 transmits, for example, the mode information 133 stored in the information storage area 130 to the operation terminal 2. A specific example of the graph information 131 will be described later.

[0059] [First embodiment]

[0060] Next, a first embodiment will be described. Figures 4 to 12 This is a flowchart illustrating the pattern search process in the first embodiment. Figures 13 to 24 This is a diagram for explaining the details of the pattern search process in the first embodiment.

[0061] [Specific example of data set]

[0062] First, a specific example of the data set D in the first embodiment will be described. Figure 13 This is a diagram illustrating a specific example of the data set D. Specifically, Figure 13 (A) is a diagram illustrating a specific example of a data set DP containing positive samples (samples that satisfy the target condition) in the data set D. Figure 13 (B) is a diagram illustrating a specific example of a data set DN containing negative samples (samples that do not satisfy the target condition) in the data set D.

[0063] Figure 13 The data set D shown has, as items, “ID” for setting identification information of each sample included in the data set D and “attribute” for setting attributes included in each sample.

[0064] Specifically, in Figure 13 In the dataset DP shown in (A), the "Attribute" of the sample with "ID" "1" is set to "Married." Furthermore, the "Attribute" of the sample with "ID" "2" is set to "Male" and "Married." The "Attribute" of the sample with "ID" "3" is set to "Male," "In his 30s," and "Married."

[0065] On the other hand, Figure 13 In the dataset DN shown in (B), the "Attribute" of the sample with "ID" "4" is set to "None," indicating that no attribute is set. Furthermore, the "Attribute" of the sample with "ID" "5" is set to "Male" and "Married," and the "Attribute" of the sample with "ID" "6" is set to "Male" and "In their 30s."

[0066] [Specific examples of occurrence sets and occurrence frequencies of attribute patterns]

[0067] Next, a specific example of the appearance set and appearance frequency of the attribute pattern in the data set D will be described. Figure 14 4 is a diagram illustrating a specific example of a table including an appearance set and an appearance frequency of attribute patterns in a data set D.

[0068] Figure 14 The table shown has "attribute mode" as an item for setting identification information of each attribute mode. Figure 14The table shown has "Occurrence Set (DP)" and "Occurrence Frequency (DP)" as items. The "Occurrence Set (DP)" sets the "ID" of the sample containing each attribute set in the "Attribute Pattern" among the samples included in the data set DP, and the "Occurrence Frequency (DP)" sets the number of attributes set in the "Occurrence Set (DP)". Figure 14 The table shown has "Occurrence Set (DN)" and "Occurrence Frequency (DN)" as items. "Occurrence Set (DN)" specifies the "ID" of each sample included in the dataset DN that contains each attribute specified in the "Attribute Pattern," while "Occurrence Frequency (DN)" specifies the number of attributes specified in the "Occurrence Set (DN)." Furthermore, samples with "None" specified in the "Attribute Pattern" (samples in the first row) correspond to attribute patterns that do not contain any attributes.

[0069] Specifically, in Figure 14 In the table shown, for example, in the sample where the "attribute mode" is "none" (the first row of samples), "1", "2" and "3" are set as the "occurrence set (DP)", "3" is set as the "occurrence frequency (DP)", "4", "5" and "6" are set as the "occurrence set (DN)", and "3" is set as the "occurrence frequency (DN)".

[0070] In addition, Figure 14 In the table shown, for example, in the sample (second row sample) where the "attribute mode" is "male", "2" and "3" are set as the "appearance set (DP)", "2" is set as the "appearance frequency (DP)", "5" and "6" are set as the "appearance set (DN)", and "2" is set as the "appearance frequency (DN)". Figure 14 A description of any additional data included.

[0071] Furthermore, the occurrence set Occ(x, D) of samples having the attribute pattern x among the samples included in the data set D is calculated by the following formula (1).

[0072] [Formula 1]

[0073]

[0074] In addition, the appearance frequency frq(x, D) of the attribute pattern x in each sample included in the data set D is calculated by the following formula (2).

[0075] [Formula 2]

[0076] frq(x,D)=|Occ(x,D)|···(Formula 2)

[0077] [Rank determination process (1)]

[0078] Next, a description will be given of a process of determining the priority of each attribute included in a plurality of attribute patterns (hereinafter also referred to as a ranking determination process) in the pattern search process. Figure 4 This is a flowchart explaining the ranking determination process.

[0079] like Figure 4 As shown, the information receiving unit 112 of the information processing device 1 waits, for example, until receiving a search instruction input via the operation terminal 2 (No in S11).

[0080] Moreover, when a search instruction is received ("Yes" in S11), the ranking determination unit 113 of the information processing device 1 generates a directed edge (S12) from the attribute with a smaller occurrence set to the attribute with a larger occurrence set for each two attributes that have an inclusion relationship in the occurrence set of negative examples in multiple samples.

[0081] Next, the ranking determination unit 113 generates a directed acyclic graph based on each directed edge generated in the process of S12 (S13). Thereafter, the information management unit 111 of the information processing device 1 generates, for example, graph information 131 representing the directed acyclic graph generated by the ranking determination unit 113 and stores it in the information storage area 130. A specific example of the processes of S12 and S13 will be described below.

[0082] [Specific Example of Processing in S12 and S13]

[0083] Figure 15 This is a diagram for explaining a specific example of the processing of S12 and S13.

[0084] exist Figure 14 In the table described in , the attributes included in each of the attribute patterns set in the "attribute pattern" are "male", "over 30 years old" and "married". Figure 14 In the table described in , in the "Occurrence Set (DN)" of the sample whose "attribute mode" is "Male" (the second row of samples), "5" and "6" are set, in the "Occurrence Set (DN)" of the sample whose "attribute mode" is "30s" (the third row of samples), "6" is set, and in the "Occurrence Set (DN)" of the sample whose "attribute mode" is "Married" (the fourth row of samples), "5" is set.

[0085] That is, in Figure 14 The expression in shows that “6”, which is the occurrence set of negative examples corresponding to “over 30”, is included in “5” and “6”, which are the occurrence sets of negative examples corresponding to “male”. Figure 14The expression described in indicates that “5”, which is the occurrence set of negative examples corresponding to “married”, is included in “5” and “6”, which are the occurrence sets of negative examples corresponding to “male”.

[0086] Therefore, in the process of S12, the ranking determination unit 113 generates a directed edge SD1 with the vertex corresponding to "30s" as the output source and the vertex corresponding to "Male" as the output destination. Furthermore, in the process of S12, the ranking determination unit 113 generates a directed edge SD2 with the vertex corresponding to "Married" as the output source and the vertex corresponding to "Male" as the output destination.

[0087] Furthermore, the ranking determination unit 113 performs the following processing in S13: Figure 15 As shown in (A) of FIG, by connecting the attributes "male", "30s", and "married" using the generated directed edges SD1 and SD2, a directed acyclic graph GR1 is generated.

[0088] Return to Figure 4 The ranking determination unit 113 refers to the directed acyclic graph generated in the process of S13 (graph information 131 stored in the information storage area 130) and determines the priority of each attribute included in the plurality of attribute patterns so that the attribute of the output source in each directed edge is ranked higher than the attribute of the output destination (S14). Then, the information management unit 111 stores, for example, ranking information 132 indicating the priority determined in the process of S14 in the information storage area 130.

[0089] Specifically, for example, Figure 15 As shown in (B), the ranking determination unit 113 determines the ranking of each attribute in the order of "over 30 years old", "married", and "male" by sorting each vertex among the output source vertices so as to be higher than the output destination vertex.

[0090] [Rank determination process (2)]

[0091] Next, Figure 4 A description will be given of a ranking determination process having a different content from the ranking determination process described in (hereinafter also referred to as other ranking determination process). Figure 5 This is a flowchart for explaining other ranking determination processing.

[0092] like Figure 5 As shown, the information receiving unit 112 waits, for example, until receiving a search instruction input via the operation terminal 2 (No in S21).

[0093] Furthermore, when a search instruction is received ("Yes" in S21), the ranking determination unit 113 determines the priority of each attribute included in the plurality of attribute patterns so that the priority of the attribute with the lowest frequency of occurrence of negative examples in the plurality of samples is given a higher priority (S22). The information management unit 111 then stores, for example, ranking information 132 indicating the priority determined in the process of S22 in the information storage area 130.

[0094] For example, if, among the attributes included in multiple attribute patterns, the occurrence set of negative examples for a certain attribute includes the occurrence set of negative examples for another attribute, it can be determined that the occurrence frequency of the attribute is greater than the occurrence frequency of the other attributes. Therefore, the rank determination unit 113 can also determine the attribute rank of each attribute based on the magnitude relationship between the occurrence frequencies of the attributes. A specific example of the processing in S22 is described below.

[0095] [Specific example of the process of S22]

[0096] exist Figure 14 In the table described in , the attributes included in each attribute pattern set in "attribute pattern" are "male", "over 30 years old" and "married". Figure 14 In the table described in , the "Occurrence Frequency (DN)" of the sample with the "attribute mode" of "Male" (the second row of samples) is set to "2", the "Occurrence Frequency (DN)" of the sample with the "attribute mode" of "30s" (the third row of samples) is set to "1", and the "Occurrence Frequency (DN)" of the data with the "attribute mode" of "Married" (the fourth row of data) is set to "1".

[0097] That is, in Figure 14 The expression in indicates that the frequency of occurrence of negative examples corresponding to "over 30 years old" (1) is smaller than the frequency of occurrence of negative examples corresponding to "male" (2). Figure 14 The expression described in indicates that the frequency of appearance of “1”, which is a negative example corresponding to “married”, is smaller than the frequency of appearance of “2”, which is a negative example corresponding to “male”.

[0098] Therefore, in the process of S22 , the ranking determination unit 113 determines the ranking of each attribute in the order of, for example, “over 30 years old,” “married,” and “male.”

[0099] [Mode determination processing (1)]

[0100] Next, a description will be given of a process of determining whether or not a plurality of attribute patterns are exposed patterns (hereinafter also referred to as pattern determination processing) in the pattern search process. Figures 6 to 8 This is a flowchart explaining the mode determination process.

[0101] like Figure 6 As shown, the mode determination unit 114 of the information processing device 1 waits, for example, until the mode determination timing is reached ("No" in S31). The mode determination timing may be, for example, the timing after the ranking determination process is completed. Alternatively, the mode determination timing may be, for example, the timing when an instruction to start the mode determination process is input via the operation terminal 2.

[0102] When the pattern determination timing arrives (Yes in S31 ), the pattern determination unit 114 determines the attribute pattern corresponding to the top search order among the plurality of attribute patterns based on the priority of each attribute determined in the process of S14 or S22 ( S32 ).

[0103] Specifically, when the pattern determination unit 114 determines the search order of multiple attribute patterns so that the attributes included in each attribute pattern are in lexicographic order according to the priority determined in the process of S14, for example, it identifies the attribute pattern associated with the top search order.

[0104] Furthermore, when an attribute pattern not including an attribute is included in a plurality of attribute patterns, the pattern determination unit 114 identifies the attribute pattern not including an attribute as the attribute pattern associated with the search order of the front end.

[0105] Specifically, in Figure 14 In the table described in [ ], the data set in the "attribute pattern" includes "None," "Male," "30s," "Married," "Male, 30s," "Male, Married," "30s, Married," and "Male, 30s, Married." Therefore, in the process of S32, pattern determination unit 114 determines "None" as the attribute pattern corresponding to the search order at the front end.

[0106] Next, the mode determination unit 114 determines whether the attribute mode specified in the process of S32 is the exposure mode ( S33 ).

[0107] As a result, if it is determined that the attribute mode determined in the process of S32 is the exposure mode ("Yes" in S34), Figure 7 As shown, the information management unit 111 stores the pattern information 133 indicating the attribute pattern identified in the process of S32 in the information storage area 130 ( S41 ).

[0108] Then, the mode determination unit 114 determines whether all attribute patterns have been determined in the process of S32, the process of S44, or the process of S52 (S42). If the mode determination unit 114 determines in the process of S34 that the attribute pattern determined in the process of S32 is not a revealing pattern ("No" in S34), the process of S42 is similarly performed.

[0109] As a result, when it is determined that not all attribute patterns have been determined in the processing of S32, the processing of S44, or the processing of S52 ("No" in S42), the pattern determination unit 114 determines the attribute pattern corresponding to the next search order among the attribute patterns that have not been determined in the processing of S32, the processing of S44, or the processing of S52 based on the priority of each attribute determined in the processing of S14 or the processing of S22 (S44).

[0110] Specifically, in Figure 14 In the table described in [ ], the data set in the "attribute pattern" includes "none," "male," "30s," "married," "male, 30s," "male, married," "30s, married," and "male, 30s, married." Therefore, if the ranking of the attributes determined in the process of S14 is in the order of "30s," "married," and "male," in the first process of S44, the pattern determination unit 114 determines "30s" as the attribute pattern corresponding to the next search order.

[0111] Then, the pattern determination unit 114 determines whether the appearance frequency of negative examples of the attribute pattern identified in the process of S44 is lower than the appearance frequency of attribute patterns including all attributes except the lowest priority attribute among the attributes included in the attribute pattern identified in the process of S44 ( S45 ).

[0112] Specifically, if the attribute pattern identified in the process of S44 is "over 30," the attribute pattern containing all attributes included in the attribute pattern identified in the process of S44, except for the attribute with the lowest priority, is "none." Therefore, in this case, the pattern determination unit 114 determines whether the frequency of occurrence of negative examples corresponding to "over 30" is lower than the frequency of occurrence of negative examples corresponding to "none."

[0113] As a result, if Figure 8 As shown, when it is determined that the frequency of occurrence of negative examples of the attribute pattern determined in the process of S44 is not lower than the frequency of occurrence of the attribute pattern having all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern determined in the process of S44 ("No" in S51), the pattern determination unit 114 determines each of the other attribute patterns formed by adding other attributes to the attribute pattern determined in the process of S44 as an attribute pattern that is not subjected to the process of S53 (S52).

[0114] That is, if it is determined that the frequency of occurrence of negative examples of the attribute pattern determined in the process of S44 is not lower than the frequency of occurrence of attribute patterns having all attributes except the lowest-priority attribute among the attributes included in the attribute pattern determined in the process of S44, the information processing device 1 can determine that the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S44 does not include a minimally exposed pattern. Therefore, for example, if it is detected that the frequency of occurrence has not decreased as described above, the information processing device 1 decides to omit the execution of the determination regarding the attribute pattern determined in the process of S44 and the execution of the determination regarding the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S44.

[0115] Thus, the information processing device 1 can reduce the computational processing cost required to extract extremely small exposure patterns.

[0116] On the other hand, when it is determined that the frequency of occurrence of the attribute pattern determined in the processing of S44 is lower than the frequency of occurrence of the attribute pattern having all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern determined in the processing of S44 ("Yes" in S51), the pattern determination unit 114 determines whether the attribute pattern determined in the processing of S44 is a revealing pattern (S53).

[0117] If the attribute mode determined in the process of S44 is the exposure mode ("YES" in S54), the information management unit 111 stores the mode information 133 indicating the attribute mode determined in the process of S44 in the information storage area 130 (S55). Thereafter, the mode determination unit 114 performs the processes from S52 onwards.

[0118] Specifically, if the attribute pattern determined in S44 is a revealing pattern, it can be determined that the additional attribute pattern formed by adding other attributes to the attribute pattern determined in S44 is not a minimal revealing pattern. Therefore, if the attribute pattern determined in S44 is determined to be a revealing pattern, the pattern determination unit 114 omits the determination of the additional attribute pattern formed by adding other attributes to the attribute pattern determined in S44.

[0119] Then, after the process of S52 or when it is determined that the attribute mode specified in the process of S44 is not the exposure mode (No in S54 ), the mode determination unit 114 performs the processes after S42 .

[0120] Afterwards, when it is determined in the processing of S42 that all attribute patterns have been determined in the processing of S33, the processing of S44 or the processing of S52 ("Yes" in S42), the information sending unit 115 of the information processing device 1 sends the information indicating the extremely small exposure mode in the pattern information 133 stored in the information storage area 130 to the operation terminal 2 (S43).

[0121] Specifically, in this case, the mode determination unit 114 determines the minimal exposure mode based on the attribute mode indicated by the mode information 133 stored in the information storage area 130. Furthermore, the information transmission unit 115 transmits information indicating the minimal exposure mode determined by the mode determination unit 114. Figures 6 to 8 A specific example of the mode determination processing described in is described.

[0122] [Specific example of mode determination processing (1)]

[0123] Figure 16 and Figure 17 Yes Figures 6 to 8 A diagram illustrating a specific example of the mode determination processing described in .

[0124] Figure 16 and Figure 17 The example shown is a tree structure that establishes a corresponding relationship between each attribute pattern and each node. Figure 16 and Figure 17 In the example shown, the numbers on the left that correspond to each node represent Figure 14 The frequency of occurrence of the positive example of the attribute pattern corresponding to each node in the data described in the figure, and the number on the right side of the corresponding relationship between each node indicates the frequency of occurrence of the positive example of the attribute pattern corresponding to each node in the data described in the figure. Figure 14 The frequency of occurrence of negative examples of the attribute pattern corresponding to each node in the data described in . In addition, the following describes the case where the frequency of occurrence of positive examples of each attribute pattern is greater than 1 and the frequency of occurrence of negative examples of each attribute pattern is 0, and each attribute pattern is determined to be a revealing pattern.

[0125] (Search for attribute patterns that do not contain attributes)

[0126] First, in the process of S32, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the first search order among the attribute patterns included in the table described in is "None". Figure 16 As shown, the mode determination unit 114 establishes a correspondence between the determined “none” and the root node, for example.

[0127] Here, in Figure 16In the example shown, the appearance frequencies of positive examples and negative examples corresponding to the attribute pattern “none” are both “3.” Therefore, in the process of S33 , the pattern determination unit 114 determines that “none” is not a revealing pattern.

[0128] (Search for attribute patterns containing "over 30 years old")

[0129] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the second search order in the attribute pattern included in the table described in is "over 30 years old". Figure 16 As shown, the mode determination unit 114 associates the identified “over 30 years old” with the child node of the root node, for example.

[0130] Here, in Figure 16 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "over 30" as the attribute pattern, i.e., "1", is lower than the frequency of occurrence of negative examples of the node corresponding to "none", i.e., "3". Therefore, in the process of S53, the pattern determination unit 114 determines whether "over 30" is a revealing pattern. Specifically, Figure 16 In the example shown, the appearance frequencies of the positive examples and the negative examples corresponding to “over 30” are both “1.” Therefore, in the process of S53 , the pattern determination unit 114 determines that “over 30” is not a revealing pattern.

[0131] (Search for attribute patterns containing "30s" and "married")

[0132] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the third search order in the attribute pattern included in the table described in is "more than 30 years old, married". Figure 16 As shown, the pattern determination unit 114 establishes a correspondence between the identified “in their 30s, married” and the child node corresponding to “in their 30s”, for example.

[0133] Here, in Figure 16 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "over 30 years old, married" is lower than the frequency of occurrence of negative examples of the node corresponding to "over 30 years old ... Figure 16 In the example shown, the appearance frequencies of positive examples and negative examples corresponding to "over 30 years old, married" are "1" and "0," respectively. Therefore, in the process of S53, the pattern determination unit 114 determines that "over 30 years old, married" is a prominent pattern.

[0134] Therefore, in the process of S52, pattern determination unit 114 decides to omit the determination of the attribute pattern formed by adding the new attribute to "30s, married." Specifically, in this case, pattern determination unit 114 omits the determination of "30s, married, male," the attribute pattern corresponding to the fourth search order.

[0135] (Search for attribute patterns containing "30s" and "male")

[0136] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the fifth search order in the attribute pattern included in the table described in is "over 30 years old, male". Figure 16 As shown, the pattern determination unit 114 establishes a correspondence between the determined “over 30s, male” and the second child node corresponding to “over 30s”, for example.

[0137] Here, in Figure 16 In the example shown, the frequency of occurrence of negative examples for the node corresponding to "30s, male," i.e., "1," is not lower than the frequency of occurrence of negative examples for the node corresponding to "over 30," i.e., "1." Therefore, the pattern determination unit 114 does not determine whether "30s, male" is a revealing pattern in the process of S53.

[0138] (Search for attribute patterns containing "married")

[0139] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the sixth search order in the attribute pattern included in the table described in is "married". Figure 16 As shown, the mode determination unit 114 establishes a correspondence between the determined "married" and the second child node of the root node, for example.

[0140] Here, in Figure 16 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "married", i.e., "1", is lower than the frequency of occurrence of negative examples of the node corresponding to "none", i.e., "3". Therefore, in the process of S53, the pattern determination unit 114 determines whether "married" is a revealing pattern. Specifically, Figure 16 In the example shown, the appearance frequencies of positive examples and negative examples corresponding to "married" are "3" and "1," respectively. Therefore, in the process of S53, the pattern determination unit 114 determines that "married" is not a revealing pattern.

[0141] (Search for attribute patterns containing "married" and "male")

[0142] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the seventh search order in the attribute pattern included in the table described in is "married, male". Figure 16 As shown, the pattern determination unit 114 establishes a correspondence between the determined “married, male” and the child node corresponding to “married”, for example.

[0143] Here, in Figure 16 In the example shown, the frequency of negative examples for the node corresponding to "married, male," i.e., "1," is not lower than the frequency of negative examples for the node corresponding to "married," i.e., "1." Therefore, in the process of S53, pattern determination unit 114 does not determine whether "married, male" is a revealing pattern.

[0144] (Search for attribute patterns containing "male")

[0145] Next, in the process of S44, the mode determination unit 114 determines whether Figure 14 The attribute pattern corresponding to the eighth search order in the attribute pattern included in the table described in is "male". Figure 16 As shown, the mode determination unit 114 establishes a correspondence relationship between the determined “male” and the third child node of the root node, for example.

[0146] Here, in Figure 16 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "male", i.e., "2", is lower than the frequency of occurrence of negative examples of the node corresponding to "none", i.e., "3". Therefore, in the process of S53, the mode determination unit 114 determines whether "male" is the revealing mode. Specifically, Figure 16 In the example shown, the appearance frequencies of positive examples and negative examples corresponding to “male” are “2” and “2”, respectively. Therefore, in the process of S53 , the pattern determination unit 114 determines that “male” is not a revealing pattern.

[0147] That is, Figure 17 As shown, the mode determination unit 114 omits the execution of determination on other attribute patterns formed by adding other attributes to the attribute pattern (attribute pattern corresponding to "○") as the exposed pattern. Figure 17 As shown, the pattern determination unit 114 omits execution of determination regarding attribute patterns that can be determined as unlikely to be extremely small exposure patterns (attribute patterns corresponding to “×”).

[0148] This allows the pattern determination unit 114 to reduce the number of times it determines whether each attribute pattern is a revealing pattern and extract a revealing pattern that has a very low probability of being a revealing pattern.

[0149] [Mode determination processing (2)]

[0150] Next, Figures 6 to 8 Mode determination processing having contents different from the mode determination processing described in (hereinafter also referred to as other mode determination processing) will be described. Figures 9 to 12 This is a flowchart illustrating other mode determination processing.

[0151] like Figure 9 As shown, the mode determination unit 114 waits, for example, until the mode determination timing arrives (No in S61 ).

[0152] When the pattern determination timing arrives (Yes in S61 ), the pattern determination unit 114 determines the attribute pattern corresponding to the top search order among the plurality of attribute patterns based on the priority of each attribute determined in the process of S14 or S22 ( S62 ).

[0153] Specifically, for example, when the search order of multiple attribute patterns is determined in such a way that the attributes contained in each attribute pattern are in a lexicographic order according to the priority determined in processing such as S14, the pattern determination unit 114 determines the attribute pattern that establishes a correspondence with the search order of the front end.

[0154] Next, the mode determination unit 114 determines whether the attribute mode specified in the process of S62 is the exposure mode ( S63 ).

[0155] As a result, when it is determined that the attribute mode determined in the process of S62 is the exposure mode ("Yes" in S64), as shown in FIG. Figure 10 As shown, the information management unit 111 stores the pattern information 133 indicating the attribute pattern identified in the process of S62 in the information storage area 130 ( S71 ).

[0156] Then, the mode determination unit 114 determines whether all attribute modes have been determined in the process of S62, the process of S75, or the process of S82 (S72). If the mode determination unit 114 determines in the process of S64 that the attribute mode determined in the process of S62 is not a revealing mode ("No" in S64), the process of S72 is also performed.

[0157] As a result, when it is determined that all attribute patterns have not been determined in the processing of S62, the processing of S75 or the processing of S82 ("No" in S72), the pattern determination unit 114 performs a process of redetermining the priority of each attribute included in multiple attribute patterns (hereinafter also referred to as redetermination processing) (S74).

[0158] That is, in Figures 9 to 12 In the other mode determination process shown, each time a determination is made as to whether or not each attribute mode is a reveal mode, the priority of each attribute is re-determined.

[0159] As a result, the information processing device 1 can create more situations in which it is possible to omit the execution of the determination of whether each attribute pattern is a revealing pattern.

[0160] [Redetermination Processing]

[0161] Figure 12 This is a flowchart illustrating the re-determination process.

[0162] The pattern determination unit 114 determines the attribute pattern corresponding to the next search order among the attribute patterns that have not been determined in the processing of S62, the processing of S75, or the processing of S82 (S101) based on the priority of each attribute that has been determined (redetermined) in the processing of S14, the processing of S22, or the processing of S74.

[0163] Then, the pattern determination unit 114 specifies an attribute pattern that includes all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern specified in the process of S101 ( S102 ).

[0164] Next, the pattern determination unit 114 specifies one or more samples corresponding to the appearance set of the attribute pattern specified in the process of S102 ( S103 ).

[0165] Specifically, the pattern determination unit 114 specifies, for example, one or more samples including the appearance set of the attribute pattern specified in the process of S102 , among the plurality of samples included in the data set DN.

[0166] Furthermore, the pattern determination unit 114 determines, for each of the one or more samples determined in the process of S103, one or more attributes whose priority has been determined in the process of S14, the process of S22, or the process of S74 and is lower than the attributes included in the attribute pattern determined in the process of S102 (S104).

[0167] Thereafter, the pattern determination unit 114 re-determines the priority of one or more attributes determined in the process of S104 based on the inclusion relationship of the occurrence set of negative examples in one or more samples determined in the process of S103 for one or more attributes determined in the process of S104 (S105).

[0168] Furthermore, when the processing of S103 and subsequent steps has already been performed on the attribute pattern identified in the processing of S102 , the pattern determination unit 114 may omit the execution of the processing of S103 and subsequent steps.

[0169] Return to Figure 10 The pattern determination unit 114 determines the attribute pattern corresponding to the next search order among the attribute patterns not determined in the processing of S62, S75 or S82 based on the priority of each attribute determined in the processing of S74 (S75).

[0170] Then, the pattern determination unit 114 determines whether the appearance frequency of negative examples of the attribute pattern identified in the process of S75 is lower than the appearance frequency of attribute patterns including all attributes except the lowest priority attribute among the attributes included in the attribute pattern identified in the process of S75 ( S76 ).

[0171] As a result, if Figure 11 As shown, when it is determined that the frequency of occurrence of negative examples of the attribute pattern determined in the process of S75 is not lower than the frequency of occurrence of the attribute pattern having all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern determined in the process of S75 ("No" in S81), the pattern determination unit 114 determines each of the other attribute patterns formed by adding other attributes to the attribute pattern determined in the process of S74 as an attribute pattern that is not subjected to the process of S83 (S82) based on the attribute patterns that have not been determined in the process of S62, the process of S75 or the process of S82.

[0172] That is, when it is determined that the frequency of occurrence of the attribute pattern determined in the process of S75 is not lower than the frequency of occurrence of attribute patterns having all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern determined in the process of S75, the information processing device 1 determines that the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S75 does not include a minimal exposure pattern. Therefore, for example, when it is detected that the frequency of occurrence has not decreased as described above, the information processing device 1 decides to omit the execution of the determination regarding the attribute pattern determined in the process of S75 and the execution of the determination regarding the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S75.

[0173] Furthermore, the information processing device 1 is used to Figure 4 and Figure 5 In the ranking determination process described in , as described above, many situations where execution of the determination can be omitted are created, and the priority of each attribute included in a plurality of attribute patterns is determined in advance.

[0174] Thus, the information processing device 1 can reduce the computational processing cost required to extract extremely small exposure patterns.

[0175] On the other hand, when it is determined that the frequency of occurrence of the attribute pattern determined in the processing of S75 is lower than the frequency of occurrence of the attribute pattern having all attributes except the attribute with the lowest priority among the attributes included in the attribute pattern determined in the processing of S75 ("Yes" in S81), the pattern determination unit 114 determines whether the attribute pattern determined in the processing of S75 is a revealing pattern (S83).

[0176] If the attribute mode determined in the process of S75 is the exposure mode ("YES" in S84), the information management unit 111 stores the mode information 133 indicating the attribute mode determined in the process of S75 in the information storage area 130 (S85). Thereafter, the mode determination unit 114 performs the processes from S82 onwards.

[0177] That is, if the attribute pattern determined in the process of S75 is a revealing pattern, it can be determined that the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S75 is not a minimal revealing pattern. Therefore, if the attribute pattern determined in the process of S75 is determined to be a revealing pattern, the pattern determination unit 114 omits the execution of the determination regarding the additional attribute pattern formed by adding other attributes to the attribute pattern determined in the process of S75.

[0178] Then, after the process of S82 or when it is determined that the attribute mode specified in the process of S75 is not the exposure mode (No in S84 ), the mode determination unit 114 performs the processes after S72 .

[0179] As a result, when it is determined in the processing of S72 that all attribute modes have been determined in the processing of S63, the processing of S75 or the processing of S82 ("Yes" in S72), the information sending unit 115 sends the information representing the extremely small exposure mode in the mode information 133 stored in the information storage area 130 to the operation terminal 2 (S73).

[0180] Specifically, in this case, the mode determination unit 114 determines the minimal exposure mode based on the attribute mode indicated by the mode information 133 stored in the information storage area 130. Furthermore, the information transmission unit 115 transmits information indicating the minimal exposure mode determined by the mode determination unit 114. Figures 9 to 12 A specific example of other mode determination processing described in is described.

[0181] [Specific example of mode determination processing (2)]

[0182] Figures 18 to 24 Yes Figures 9 to 12A diagram illustrating a specific example of other mode determination processing described in .

[0183] Figure 19 、 Figure 21 、 Figure 22 as well as Figure 24 The example shown is a tree structure that establishes a corresponding relationship between each attribute pattern and each node. Figure 19 In the examples shown in , the numbers on the left side of each node correspond to the frequency of occurrence of positive examples of the attribute pattern corresponding to each node, and the numbers on the right side of each node correspond to the frequency of occurrence of negative examples of the attribute pattern corresponding to each node. Furthermore, the following description assumes that each attribute pattern is determined to be a revealing pattern when the frequency of occurrence of positive examples of each attribute pattern is greater than 1 and the frequency of occurrence of negative examples of each attribute pattern is 0.

[0184] [Specific example of data set]

[0185] First of all, Figures 9 to 12 A specific example of the data set D used in the description of the other mode determination processing described in will be described. Figure 18 This is a diagram illustrating a specific example of the data set D. Specifically, Figure 18 (A) is a diagram illustrating a specific example of a data set DP containing positive examples. Figure 18 (B) is a diagram illustrating a specific example of a data set DP containing negative samples.

[0186] Figure 18 The dataset D shown has the same Figure 13 The same project as described in dataset D.

[0187] Specifically, in Figure 18 In the dataset DP shown in (A), "c" and "e" are set in the "attribute" of the sample with "ID" "1", and "b", "c", and "e" are set in the "attribute" of the sample with "ID" "2". In addition, "a", "b", and "d" are set in the "attribute" of the sample with "ID" "3", and "a", "b", and "d" are set in the "attribute" of the sample with "ID" "4".

[0188] On the other hand, Figure 18 In the dataset DN shown in (B), the "attribute" of the sample with "ID" "5" is set to "d", and the "attribute" of the sample with "ID" "6" is set to "c" and "d". Figure 18In the data set DN shown in (B), "a", "b" and "c" are set in the "attributes" of the sample with "ID" "7", and "a", "b", "c" and "d" are set in the "attributes" of the sample with "ID" "8".

[0189] [Specific Examples of Other Mode Determination Processes]

[0190] Next, Figures 9 to 12 A specific example of other mode determination processing described in is described.

[0191] (Search for attribute patterns that do not contain attributes)

[0192] First, in the process of S62, the mode determination unit 114 determines the attribute mode corresponding to the search order of the front end, that is, "none". Figure 19 As shown, a corresponding relationship is established between the determined "none" and the root node.

[0193] Here, in Figure 19 In the example shown, the appearance frequencies of positive examples and negative examples corresponding to “none” are both “4.” Therefore, in the process of S63 , the pattern determination unit 114 determines that “none” is not a revealing pattern.

[0194] (Search for attribute patterns containing "a")

[0195] Next, in the process of S74 (redetermination process), the pattern determination unit 114 re-determines the priority of each attribute included in the plurality of attribute patterns to the order of "a", "b", "c", and "d". Then, in the process of S75, the pattern determination unit 114 determines the attribute pattern corresponding to the next search order, that is, "a", based on the priority of each attribute. In addition, for example, Figure 19 As shown, the mode determination unit 114 establishes a correspondence between the determined "a" and the child node corresponding to the root node.

[0196] Here, in Figure 19 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "a", i.e., "2", is lower than the frequency of occurrence of negative examples of the node corresponding to "none", i.e., "4". Therefore, in the process of S81, the pattern determination unit 114 determines whether "a" is a revealing pattern. Specifically, Figure 19 In the example shown, the appearance frequencies of the positive examples and the appearance frequencies of the negative examples corresponding to “a” are both “2.” Therefore, in the process of S83 , the pattern determination unit 114 determines that “a” is not a revealing pattern.

[0197] (Search for attribute patterns containing "a" and "d")

[0198] Next, in the process of S74, pattern determination unit 114 determines "a, b" as the attribute pattern corresponding to the next search order based on the previously determined (redetermined) priorities of each attribute, namely, "a," "b," "c," and "d." Furthermore, pattern determination unit 114 determines "a" as the attribute pattern that includes all attributes included in the determined attribute pattern, excluding the attribute with the lowest priority.

[0199] Next, in the process of S74, the mode determination unit 114 will Figure 18 The samples corresponding to the third and fourth rows of the data set DN described in (B) are determined to be samples containing the determined attribute pattern, that is, "a". Figure 20 As shown in (A), the mode determination unit 114 determines that the attributes "b" and "c" have a lower priority than "a" among the attributes included in the sample corresponding to the third row. Figure 20 As shown in (A) of FIG, mode determination unit 114 identifies attributes "b," "c," and "d" as having lower priorities than "a" among the attributes included in the sample corresponding to the fourth row. Furthermore, mode determination unit 114 re-determines the priorities of the identified attributes to "d," "b," and "c" in that order.

[0200] Then, in the process of S75, the pattern determination unit 114 determines "a, d" as the attribute pattern corresponding to the next search order based on the priority of each attribute including "b", "c", and "d" ("b", "c", and "d"). Figure 19 As shown, the pattern determination unit 114 establishes a correspondence between the determined "a, d" and the child node corresponding to "a".

[0201] Here, in Figure 19 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "a, d", i.e., "1", is lower than the frequency of occurrence of negative examples of the node corresponding to "a", i.e., "2". Therefore, in the process of S81, the pattern determination unit 114 determines whether "a, d" is a revealing pattern. Specifically, Figure 19 In the example shown, the frequencies of appearance of positive examples and negative examples corresponding to "a, d" are "2" and "1," respectively. Therefore, in the process of S83, the pattern determination unit 114 determines that "a, d" is not a revealing pattern.

[0202] (Search for attribute patterns containing "a", "b", and "d")

[0203] Next, in the process of S74, pattern determination unit 114 determines "a, b, d" as the attribute pattern corresponding to the next search order based on the previously determined (redetermined) priorities of each attribute, namely, "d," "b," and "c." Furthermore, pattern determination unit 114 determines "a, d" as the attribute pattern that includes the attributes included in the determined attribute pattern, excluding the attribute with the lowest priority.

[0204] Next, in the process of S74, the mode determination unit 114 will Figure 18 The samples corresponding to the fourth row among the samples included in the dataset DN described in (B) are determined to be samples including "a, d" as the attribute pattern. Figure 20 As shown in (B), the mode determination unit 114 identifies "b" and "c," which have lower priorities than "d," among the attributes included in the sample corresponding to the fourth row. Furthermore, the mode determination unit 114 re-determines the priorities of the identified attributes to "b" and "c," respectively.

[0205] Then, in the process of S75, the pattern determination unit 114 determines "a, b, d" as the attribute pattern corresponding to the next search order based on the priorities of the attributes that have been re-determined. Figure 19 As shown, the pattern determination unit 114 establishes a correspondence relationship between the determined "a, b, d" and the child nodes corresponding to "a, d".

[0206] Here, in Figure 19 In the example shown, the frequency of occurrence of negative examples for the node corresponding to "a, b, d," i.e., "1," is not lower than the frequency of occurrence of negative examples for the node corresponding to "a, d," i.e., "1." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "a, b, d" is a revealing pattern.

[0207] (Search for attribute patterns containing "a", "c", and "d")

[0208] Next, in the process of S75, the pattern determination unit 114 determines "a, c, d" as the attribute pattern corresponding to the next search order based on the priorities ("b" and "c") of the attributes including the already determined "b" and "c". Figure 19 As shown, the pattern determination unit 114 establishes a correspondence between the determined “a, c, d” and the second child node corresponding to “a, d”.

[0209] Here, in Figure 19In the example shown, the frequency of occurrence of negative examples for the node corresponding to "a, c, d," i.e., "1," is not lower than the frequency of occurrence of negative examples for the node corresponding to "a, d," i.e., "1." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "a, c, d" is a revealing pattern.

[0210] (Search for attribute patterns containing "a" and "b")

[0211] Next, in the process of S75, the pattern determination unit 114 determines "a, b" as the attribute pattern corresponding to the next search order based on the priority of each attribute including "b", "c", and "d" that has been determined ("d", "b", and "c"). Figure 21 As shown, the mode determination unit 114 establishes a correspondence between the determined "a, b" and the second child node corresponding to "a".

[0212] Here, in Figure 21 In the example shown, the frequency of occurrence of negative examples for the node corresponding to "a, b," namely, "2," is not lower than the frequency of occurrence of negative examples for the node corresponding to "a," namely, "2." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "a, b" is a revealing pattern.

[0213] (Search for attribute patterns containing "a" and "c")

[0214] Next, in the process of S75, the pattern determination unit 114 determines "a, c" as the attribute pattern corresponding to the next search order based on the priority of each attribute including "b", "c", and "d" that has been determined ("d", "b", and "c"). In addition, for example, Figure 21 As shown, the mode determination unit 114 establishes a correspondence between the determined "a, c" and the third child node corresponding to "a".

[0215] Here, in Figure 21 In the example shown, the frequency of occurrence of negative examples for the node corresponding to "a, c," namely, "2," is not lower than the frequency of occurrence of negative examples for the node corresponding to "a," namely, "2." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "a, c" is a revealing pattern.

[0216] (Search for attribute patterns containing "b")

[0217] Next, in the process of S75, the pattern determination unit 114 determines "b" as the attribute pattern corresponding to the next search order based on the priority of each attribute including "a", "b", "c" and "d" that has been determined ("a", "b", "c" and "d"). Figure 22 As shown, the mode determination unit 114 establishes a correspondence between the determined “b” and the second child node corresponding to “none”.

[0218] Here, in Figure 22 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "b", i.e., "2", is lower than the frequency of occurrence of negative examples of the node corresponding to "none", i.e., "4". Therefore, in the process of S81, the pattern determination unit 114 determines whether "b" is a revealing pattern. Specifically, Figure 22 In the example shown, the frequencies of appearance of positive examples and negative examples corresponding to “b” are “3” and “2,” respectively. Therefore, in the process of S83 , the pattern determination unit 114 determines that “b” is not a revealing pattern.

[0219] (Search for attribute patterns containing "b" and "d")

[0220] Next, in the process of S74, pattern determination unit 114 determines "b, c" as the attribute pattern corresponding to the next search order based on the priorities ("a," "b," "c," and "d") of the attributes including the determined "a," "b," "c," and "d." Furthermore, pattern determination unit 114 determines "b" as the attribute pattern including the attributes included in the determined attribute pattern, excluding the attribute with the lowest priority.

[0221] Next, in the process of S74, the mode determination unit 114 will Figure 18 The samples corresponding to the third and fourth rows, respectively, among the samples included in the data set DN described in (B) are determined to be samples including "b" as the attribute pattern. Figure 23 As shown in (A), the mode determination unit 114 determines that the attribute "c" has a lower priority than "b" among the attributes included in the sample corresponding to the third row. Figure 23 As shown in (A), the mode determination unit 114 identifies attributes "c" and "d" as having lower priorities than "b" among the attributes included in the sample corresponding to the fourth row. Furthermore, the mode determination unit 114 determines the priorities of the identified attributes in the order of "d" and "c."

[0222] Then, in the process of S75, the pattern determination unit 114 determines "b, d" as the attribute pattern corresponding to the next search order based on the priorities of the attributes that have been re-determined. Figure 22 As shown, the pattern determination unit 114 establishes a correspondence between the determined "b, d" and the child node corresponding to "b".

[0223] Here, in Figure 22 In the example shown, the frequency of occurrence of negative examples of the node corresponding to "b, d", i.e., "1", is lower than the frequency of occurrence of negative examples of the node corresponding to "b", i.e., "2". Therefore, in the process of S81, the pattern determination unit 114 determines whether "b, d" is a revealing pattern. Specifically, Figure 22 In the example shown, the frequencies of appearance of positive examples and negative examples corresponding to "b, d" are "2" and "1," respectively. Therefore, in the process of S83, the pattern determination unit 114 determines that "b, d" is not a revealing pattern.

[0224] (Search for attribute patterns containing "b", "c", and "d")

[0225] Next, in the process of S74, pattern determination unit 114 determines "b, c, d" as the attribute pattern corresponding to the next search order based on the previously determined (redetermined) priorities of each attribute, namely "d" and "c." Furthermore, pattern determination unit 114 determines "b, d" as the attribute pattern that includes the attributes included in the determined attribute pattern, excluding the attribute with the lowest priority.

[0226] Next, in the process of S74, the mode determination unit 114 will Figure 18 The samples corresponding to the fourth row among the samples included in the dataset DN described in (B) are determined to be samples including "b, d" as the attribute pattern. Figure 23 As shown in (B), the mode determination unit 114 determines that the attribute "c" has a lower priority than "d" among the attributes included in the sample corresponding to the fourth row.

[0227] Then, in the process of S75, the pattern determination unit 114 determines "b, c, d" as the attribute pattern corresponding to the next search order. Figure 22 As shown, the pattern determination unit 114 establishes a correspondence relationship between the determined "b, c, d" and the child nodes corresponding to "b, d", for example.

[0228] Here, in Figure 22 In the example shown, the frequency of occurrence of negative examples for the nodes corresponding to "b, c, d," i.e., "1," is not lower than the frequency of occurrence of negative examples for the nodes corresponding to "b, d," i.e., "1." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "b, c, d" is a revealing pattern.

[0229] (Search for attribute patterns containing "b" and "c")

[0230] Next, in the process of S75, the pattern determination unit 114 determines "b, c" as the attribute pattern corresponding to the next search order based on the priority of each attribute including the already determined "a", "b", "c" and "d" ("a", "b", "c" and "d"). Figure 24 As shown, the mode determination unit 114 establishes a correspondence relationship between the determined "b, c" and the second child node corresponding to "d".

[0231] Here, in Figure 24 In the example shown, the frequency of occurrence of negative examples for the node corresponding to "b, c," namely, "2," is not lower than the frequency of occurrence of negative examples for the node corresponding to "b," namely, "2." Therefore, in the process of S82, the pattern determination unit 114 decides not to determine whether "b, c" is a revealing pattern.

[0232] As described above, information processing device 1 in this embodiment determines the priority of each attribute included in multiple attribute patterns based on the inclusion relationship between the occurrence sets of each attribute included in multiple samples. Furthermore, information processing device 1 determines whether each attribute pattern in the multiple attribute patterns is a revealing pattern based on the search order determined based on the determined priority.

[0233] Here, whenever determining whether each attribute pattern is a revealing pattern, information processing device 1 determines whether the frequency of occurrence of the first attribute pattern (the attribute pattern to be determined next) in multiple samples is lower than the frequency of occurrence of the second attribute pattern (an attribute pattern containing all attributes included in the first attribute pattern except the attribute with the lowest priority) in multiple samples. If it is determined that the frequency of occurrence of the first attribute pattern is not lower than the frequency of occurrence of the second attribute pattern, information processing device 1 does not make a determination regarding the first attribute pattern.

[0234] Specifically, in this embodiment, information processing device 1 determines that the first attribute pattern is not a minimally revealed pattern if the frequency of occurrence of the first attribute pattern is not lower than the frequency of occurrence of the second attribute pattern. Therefore, if the processing objective is to extract minimally revealed patterns, information processing device 1 omits the determination of the first attribute pattern based on the fact that the frequency of occurrence of the first attribute pattern is not lower than the frequency of occurrence of the second attribute pattern.

[0235] Furthermore, in order to create a plurality of first attribute patterns that can omit execution of determination as described above, the information processing device 1 in this embodiment preliminarily sorts the attributes included in the plurality of attribute patterns.

[0236] Thus, the information processing device 1 can reduce the computational processing cost required to extract the exposure pattern. Specifically, the information processing device 1 can reduce the computational processing cost required to extract an extremely small exposure pattern.

[0237] Description of Reference Numerals

[0238] 1: information processing device; 2: operation terminal; 10: information processing system; NW: network.

Claims

1. A non-transitory computer-readable storage medium storing a pattern search program, characterized in that: At least one computer is caused to perform the following processing: Continuously monitoring a communication device of the at least one computer to detect whether the communication device receives a command signal from an operation terminal via a network; In response to a received instruction signal, dynamically determining a priority of each attribute included in a plurality of attribute patterns stored in a memory, each of the plurality of attribute patterns including a combination of one or more attributes, the dynamic determination of the priority being performed for each attribute based on an inclusion relationship between negative samples from the plurality of samples relative to an occurrence set of each attribute included in the plurality of attribute patterns; In response to the priority determined for each attribute, for a first attribute pattern that is each of the plurality of attribute patterns, dynamically determining whether an occurrence frequency of the first attribute pattern in the plurality of samples is less than an occurrence frequency of a second attribute pattern in the plurality of samples, the second attribute pattern having all attributes except an attribute with the lowest priority among the attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is lower than the appearance frequency of the second attribute pattern in the plurality of samples, determining whether the first attribute pattern is a revealing pattern is performed on the first attribute pattern by determining whether the first attribute pattern is a revealing pattern according to a search order based on a priority obtained by dynamically determining attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is not lower than the appearance frequency of the second attribute pattern in the plurality of samples, excluding the first attribute pattern from the processing target of determining the exposure pattern; as well as When the first attribute pattern is determined as the exposure pattern by performing the exposure pattern determination, the first attribute pattern that is the exposure pattern among the plurality of attribute patterns is automatically stored in the memory.

2. The non-transitory computer-readable storage medium according to claim 1, wherein The determination of the above priorities includes: For each two attributes with the above inclusion relationship, generate a directed edge from the attribute with the smaller occurrence set to the attribute with the larger occurrence set. Generate a directed acyclic graph based on each of the directed edges generated above, Based on the generated directed acyclic graph, the priority of each attribute included in the plurality of attribute patterns is determined.

3. The non-transitory computer-readable storage medium according to claim 2, wherein: The determination of the priority includes: referring to the directed acyclic graph, determining the priority of each attribute included in the plurality of attribute patterns so that the attribute of the output source in each directed edge is ranked higher than the attribute of the output destination.

4. The non-transitory computer-readable storage medium according to claim 3, wherein: When determining the exposure pattern, the search order is determined based on the dynamically determined priority so that the attributes included in the first attribute pattern are in lexicographic order according to the priority.

5. The non-transitory computer-readable storage medium according to claim 4, wherein The plurality of attribute patterns include an attribute pattern that does not include any of the attributes. When determining the exposure pattern, the attribute pattern that does not include any of the attributes is selected as the first attribute pattern, and the first attribute pattern is earlier than the remaining attribute patterns of the plurality of attribute patterns.

6. The non-transitory computer-readable storage medium according to claim 1, wherein The determination of the priorities includes determining the priorities of the attributes included in the plurality of attribute patterns based on the occurrence frequencies of the negative samples in the plurality of samples relative to the attributes included in the plurality of attribute patterns.

7. The non-transitory computer-readable storage medium according to claim 6, wherein: The determination of the priorities includes determining the priorities of the attributes included in the plurality of attribute patterns such that an attribute with a greater frequency of appearance in the plurality of samples has a higher priority.

8. The non-transitory computer-readable storage medium according to claim 1, wherein Executing the determination of the above-mentioned exposure pattern includes: when the appearance frequency of the above-mentioned first attribute pattern in the above-mentioned multiple samples is lower than the appearance frequency of the above-mentioned second attribute pattern in the above-mentioned multiple samples, performing the above-mentioned determination on the other attribute pattern formed by adding other attributes to the above-mentioned first attribute pattern.

9. The non-transitory computer-readable storage medium according to claim 1, wherein: The determining of the exposure mode includes re-determining the first attribute mode each time the determination is made.

10. The non-transitory computer-readable storage medium according to claim 9, wherein Performing the above-mentioned determination of the exposure mode includes: Among the plurality of samples, one or more samples corresponding to the occurrence set of the second attribute pattern are determined. For each of the one or more determined samples, one or more attributes included in the plurality of attribute patterns whose priority is lower than that of the attributes included in the second attribute pattern are determined; Based on the inclusion relationship between the one or more samples and the occurrence set of the one or more attributes, the priority of the one or more attributes determined above is re-determined. The first attribute pattern is re-determined according to the search order based on the re-determined priority.

11. A pattern search device, characterized in that: have: communications equipment; one or more memories; and One or more processors, coupled to the one or more memories and the communication device, configured to perform the following processing: Continuously monitoring the communication device to detect whether the communication device receives a command signal from the operation terminal via the network; In response to a received instruction signal, dynamically determining a priority of each attribute included in a plurality of attribute patterns stored in a memory, each of the plurality of attribute patterns including a combination of one or more attributes, the dynamic determination of the priority being performed for each attribute based on an inclusion relationship between negative samples from the plurality of samples relative to an occurrence set of each attribute included in the plurality of attribute patterns; In response to the priority determined for each attribute, for a first attribute pattern that is each of the plurality of attribute patterns, dynamically determining whether an occurrence frequency of the first attribute pattern in the plurality of samples is less than an occurrence frequency of a second attribute pattern in the plurality of samples, the second attribute pattern having all attributes except an attribute with the lowest priority among the attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is lower than the appearance frequency of the second attribute pattern in the plurality of samples, determining whether the first attribute pattern is a revealing pattern is performed on the first attribute pattern by determining whether the first attribute pattern is a revealing pattern according to a search order based on a priority obtained by dynamically determining attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is not lower than the appearance frequency of the second attribute pattern in the plurality of samples, excluding the first attribute pattern from the processing target of determining the exposure pattern; as well as When the first attribute pattern is determined as the exposure pattern by performing the exposure pattern determination, the first attribute pattern that is the exposure pattern among the plurality of attribute patterns is automatically stored in the memory.

12. A pattern search method, characterized in that: Make the computer execute the following processing: Continuously monitoring the communication device of the computer to detect whether the communication device receives a command signal from the operation terminal via the network; In response to a received instruction signal, dynamically determining a priority of each attribute included in a plurality of attribute patterns stored in a memory, each of the plurality of attribute patterns including a combination of one or more attributes, the dynamic determination of the priority being performed for each attribute based on an inclusion relationship between negative samples from the plurality of samples relative to an occurrence set of each attribute included in the plurality of attribute patterns; In response to the priority determined for each attribute, for a first attribute pattern that is each of the plurality of attribute patterns, dynamically determining whether an occurrence frequency of the first attribute pattern in the plurality of samples is less than an occurrence frequency of a second attribute pattern in the plurality of samples, the second attribute pattern having all attributes except an attribute with the lowest priority among the attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is lower than the appearance frequency of the second attribute pattern in the plurality of samples, determining whether the first attribute pattern is a revealing pattern is performed on the first attribute pattern by determining whether the first attribute pattern is a revealing pattern according to a search order based on a priority obtained by dynamically determining attributes included in the first attribute pattern; When the appearance frequency of the first attribute pattern in the plurality of samples is not lower than the appearance frequency of the second attribute pattern in the plurality of samples, excluding the first attribute pattern from the processing target of determining the exposure pattern; as well as When the first attribute pattern is determined as the exposure pattern by performing the exposure pattern determination, the first attribute pattern that is the exposure pattern among the plurality of attribute patterns is automatically stored in the memory.

Images