JP5478229B2 - Data analysis system and method - Google Patents

Description

  The present invention relates to a data analysis system and method, and more particularly to a data mining technique for clarifying regularity of appearance order of data included in a database.

  With the widespread use of data collection environments such as mobile terminals, IC cards, and IC tags, it has become possible to acquire and store a large amount of human behavior and the state of things from time to time. There is a need to analyze this large amount of accumulated human behavior and object state data, extract characteristic or typical behavior patterns and state patterns, and apply them to marketing and healthcare. As a solution, data mining is known in which a large amount of data is analyzed and useful regularities and patterns buried in the data are extracted. In particular, there is time-series pattern mining as a technique for analyzing a pattern of appearance order on the time axis of data.

  For example, consider time-series pattern mining of credit card usage data. The history of the use of the credit card at the store by the customer is recorded as the use data of the credit card, the use date, the use store, and the use amount. As a pattern that appears in common with a plurality of customers, a time-series pattern that is an order-related pattern can be extracted from a large amount of credit card usage data. If a time-series pattern such as “Customers who purchased at store A often make expensive purchases at store B” is extracted, it can be seen that store A and store B have a relationship with concurrent purchase behavior. , Can be used for store location and sales strategy. In addition, it is possible to extract a typical access pattern of the user from the access log of the website, that there is an access that is not included in the extracted access pattern, and an access pattern that has not been extracted from past data. It can be used to detect abnormal access by determining whether it has been extracted with new data. In addition, it is possible to extract operation status patterns and normal operation patterns that are likely to cause failures from the operation history of construction machinery and the history of failures and maintenance. It can be useful for detection.

Research on extracting time-series patterns from a large amount of data has been conducted in the field of data mining. For example, there are methods described in Patent Document 1 and Non-Patent Document 1. In the methods of Patent Literature 1 and Non-Patent Literature 1, the support level (appearance frequency of the appearance frequency) set in advance by the user from a database including a combination of items (data items, events) and an identifier indicating a time stamp (time) or appearance order. A time series pattern that is equal to or greater than the minimum value (representing the ratio to all data) is extracted. A time-series pattern is a pattern including the order of appearance of a combination of items (item set), and a time-series pattern composed of one or more n item sets is represented by <(IS1)... (ISn)>. The Here, (IS1),..., (ISn) are item sets, and the item set includes one or more items. The support degree of a certain time series pattern is the ratio of the time series data including the time series pattern in all the time series data. A time series pattern that is equal to or greater than the minimum support is called a frequent time series pattern. Frequent time-series pattern extraction is performed by creating a candidate time-series pattern (candidate time-series pattern), counting the frequency at which the candidate time-series pattern appears in the time-series data by reading data from the database, and the minimum support level. This is performed by selecting a time-series pattern having the above frequency.

  As another problem of time series pattern mining, there is also a problem of extracting a time series pattern (repeated time series pattern) repeated in one time series data. For example, consider repeated time series pattern extraction of credit card usage data. A time-series pattern that is a usage pattern that is repeated a plurality of times can be extracted from usage data of one user that has been used for a long period of time. If a time-series pattern such as “Purchase at store C, then purchase at store D, then often at store E” is extracted, store C, store D, and store E are determined. It can be seen that it is regularly used in order, which can be useful for advertising strategy and sales strategy, and for segmentation that classifies user types based on the tendency of regular usage patterns.

  Research to extract repeated time series patterns has been conducted in the fields of data mining and bioinformatics. For example, there are methods described in Patent Documents 2 and 3. According to the method described in Patent Document 2, it is possible to extract a regularly performed pattern by extracting a time series pattern repeated more than a predetermined frequency in one person's time series data. In the method described in Patent Document 3, a pattern that is determined to be statistically repeated is extracted without counting the exact number of times that are actually repeated. Unlike Patent Document 2, repeated patterns are considered to be meaningless parts, and the removal of them is the purpose of repeated pattern extraction in the field of biotechnology like Patent Document 3. In Patent Document 2, repeated patterns are considered to be meaningful parts, and the purpose is to find them together with the exact number of appearances.

JP-A-8-263346 JP 2001-229202 US Patent Application Publication 2003/0068617

R. Agrawal, R. Srikant, "Mining Sequential Patterns: Generalizations and Performance Improvements", in proceedings of International Conference on Extending Database Technology, 1996

  In actual analysis or analysis, not all data is always analyzed. For example, in human behavior analysis, the behavior that happens to be performed may be removed as noise, and only behavior that has been repeated more than a certain number of times may be the analysis target. There is also a need to extract the actions that many users perform regularly as significant action patterns. In purchasing analysis, knowing the purchase pattern of customers who make regular and repeated purchases and promoting purchases in that pattern can help to create customers who make regular and repeated purchases.

  As described above, Patent Document 1 and Non-Patent Document 1, and Patent Documents 2 and 3 respectively extract time series patterns that appear in common to a plurality of customers in time series data of a large number of persons, and time series data of one person. It is possible to extract a time series pattern that is repeated a plurality of times. However, it has not been considered to extract a time series pattern considering both of them.

  Simply, a pattern satisfying both conditions can be extracted by first extracting repeated patterns from the time-series data of individual customers and extracting patterns extracted by a predetermined number or more of them from them. However, since the purchase pattern to be repeated for each customer is different, the simple combination method performs an unnecessary pattern search process on a huge amount of time-series data, and a huge amount of unnecessary processing occurs. It is difficult.

  An object of the present invention is a time series pattern that is repeated a predetermined number of times or more in each time series data, and a frequently repeated time series pattern that is a time series pattern that satisfies the condition of the number of repetitions in a time series data of a predetermined number or more. To provide a data analysis system and method for extraction.

  Another object of the present invention is to provide a data analysis system and method that reduce the amount of search processing by using checkpoints and calculating the upper limit of the number of repetitions and the number of appearances.

  In order to achieve the above object, in the present invention, a computer including a processing unit and a storage unit is used, and a plurality of sets of information indicating an event, an ID to which the event belongs, and an order relationship between events are stored. The data is time-series data in which events having the same ID are arranged according to the order relationship, and a duplicate permutation in which one or more events are arranged in the forward direction is a time-series pattern. In order to extract a frequent repeating time series pattern that is a time series pattern that is repeated a predetermined number of times in series data, a step of counting the number of repetitions in each time series data for a time series pattern with an unknown number of repetitions, A step of counting the number of time-series data that is equal to or greater than a predetermined number of repetitions, and the number of time-series data counted is a predetermined number or more. Configuring the data analysis system and method for performing the steps in the processing unit for extracting time series pattern as a.

In order to achieve the above object, in the present invention, a computer including a processing unit and a storage unit is used to store a plurality of sets of events, IDs to which events belong, and information indicating the order relationship between events. The time-series data in which the events having the same ID are arranged according to the order relation, and the overlapping permutation in which one or more events are arranged in the forward direction are set as a time-series pattern. In order to extract a frequent repeated time series pattern that is a time series pattern repeated for a predetermined number of times in each time series data, a first step for setting checkpoints at predetermined intervals in each time series data, and each time series For a time-series pattern whose number of repetitions in the data is unknown, the time-series pattern of each time-series data is in the range from the check point to the next check point. A second step of counting the number of times the process is repeated;
The upper limit value of the number of repetitions of the time series pattern in the time series data is calculated from the sum of the number of repetitions up to the already counted check point and the number of repetitions of each event appearing after the check point. A fourth step of counting the number of time series data for which the calculated upper limit value is equal to or greater than a predetermined number of repetitions, and a time series pattern for extracting the time series pattern for which the number of time series data is equal to or greater than a predetermined number A data analysis system and method for executing the fifth step and the sixth step of repeating the second to fifth steps up to the last check point for the extracted time series pattern in the processing unit are configured.

  According to the present invention, it is possible to extract a time series pattern that is repeated a predetermined number of times or more in individual time series data and that satisfies the repetition condition in a predetermined number or more of time series data. Become.

  Further, the upper limit value of the number of repetitions is calculated in the process of counting the number of repetitions of each time series data for each data processing unit of the present invention, and when the number of repetitions is less than the predetermined number of repetitions, the subsequent number of repetitions is avoided. The amount of analysis processing can be reduced by counting the appearance frequency for each data processing unit.

It is a figure which shows the system configuration example of a 1st Example. It is a figure which shows an example of the user interface based on a 1st Example. It is the flowchart which showed the relationship between user operation and system operation | movement based on a 1st Example. It is a flowchart figure which shows the outline | summary of the frequent repetition time series pattern extraction process based on 1st Example. It is a flowchart figure which shows the frequent repeated item extraction process based on 1st Example. It is a flowchart figure which shows the time series data read-out process based on a 1st Example. It is a flowchart figure which shows the repeating pattern count process based on 1st Example. It is a flowchart figure which shows the counting process of the repetition frequency of the specific time series pattern with respect to specific time series data based on 1st Example. It is a flowchart figure which shows the appearance frequency counting process in this invention based on a 1st Example. It is a figure which shows the system configuration example of a 3rd Example. It is a figure which shows the example of a user interface based on a 3rd Example. It is the figure which showed the relationship between user operation and system operation | movement based on a 3rd Example. It is a flowchart figure which shows the modification of the counting process of the repetition frequency of the specific time series pattern with respect to specific time series data based on 3rd Example. It is an image figure which shows the relationship between the time series data reading process based on 1st Example, and the time series data read from a memory | storage device.

  Embodiments of the present invention will be described below with reference to the drawings.

  First, the configuration of data used in various embodiments will be described. The database is composed of a set of records, and the record is composed of a combination of events (items) (an item set), an identifier to which the combination of the events belongs (time series data ID), and an identifier indicating a time stamp or an order relationship. Data expressing one or more records having the same time-series data ID as a set of item sets arranged in the order of time stamps or identifiers indicating order relations is called time-series data. Here, the item is a discrete value. When the item is a continuous value, it is possible to associate the continuous value with the discrete value by dividing the item by range division or the like and assigning a specific discrete value to each division. Further, by classifying discrete values into groups and associating each group with a specific discrete value, it is also possible to associate with discrete values not included in the item.

  For example, Tables 1 and 2 show examples of credit card usage data. Table 1 is a tabular representation. In the case of credit card usage data, one record means one use of a customer, the time series data ID is “card ID”, and the time stamp is “use date”. The event becomes “usage content”. Table 2 is a representation of time-series data format. One time-series data means a long-term usage history (usage history) of a customer, and a list of events is arranged in the order used. It becomes contents.

  The time series pattern is an overlapping permutation of item sets, and the time series pattern composed of one or more n item sets (IS1),..., (ISn) is <(IS1) ... (ISn)>. It is expressed as This is called the number of repetitions, which is the number of times a certain time series pattern appears in one piece of time series data. A certain time series pattern has an appearance frequency, which is the number of time series data in which the time series pattern appears more than a predetermined number of repetitions, and a statistical value of the number of repetitions in the time series data as evaluation values. Here, the number of time-series data is equal to the number of types of different time-series data IDs.

  In the example of data shown in Tables 1 and 2, the number of time-series data is 3, for example, the number of repetitions in the data with the card ID card01 of the time-series pattern <(store A) (store B)> is 2. . Also, for example, the time series pattern <(store A) (store B)> has a card ID of card01, the number of repetitions is 2, card02, 1 and card03 is 3, so the predetermined number of repetitions is set to 2 or more. In this case, the appearance frequency is 2, which corresponds to card01 and card03. The statistical value of the number of repetitions is 2.5 on average, 3 on the maximum, and 2 on the minimum.

  FIG. 1 is a diagram illustrating a configuration example of the data analysis system according to the first embodiment. This system includes a processor 101 that is a processing unit, a memory 102 that constitutes a storage unit, and a storage device 103. The processor 101 and the memory 102 constitute a computer 100, and data to be analyzed is stored in the storage device 103. The time-series pattern extraction program of the present embodiment is stored in the memory 102, and the processing shown in FIG.

  As shown in FIG. 1, the memory 102 stores setting value information 107 to checkpoint information 111 in addition to the execution program 106. The setting value information 107 holds analysis target data, a minimum value of the number of repetitions, a minimum value of appearance frequency, and a setting value of a processing unit of the analysis target data in the form of data or a file. The time-series data information 108 includes time-series data ID and time-series data for the time-series data read from the storage device 103, for example, card01, <(store A) (store C) (store A) (store B, high-value settlement). ) (Store C)> and hold in a table format or a list format.

  For each item appearing in the data to be analyzed, the item information 109 is an item, a time-series data ID, the number of repetitions in the time-series data, and the time-series pattern to be searched, such as (Store A, card 01, 3, 0). A combination of the number of repetitions (called a count value) to be used for counting is held in a table format or a list format. The search time-series pattern information 110 includes the search time-series pattern and time-series data for the time-series pattern to be searched, such as (<(store A) (store B, high-price payment)>, card01, 1, 0). A combination of the ID, the counted number of repetitions and the pattern position of the counted time series pattern is held in a table format or a list format.

  Further, the memory 102 holds the start position of the time-series data for counting up the number of repetitions as checkpoint information 111. This check point will be described later. Further, the computer 100 is connected to an input device 104 including a keyboard and a mouse, and an output device 105 including a display and a printer.

  FIG. 2 shows an example of the user interface of this embodiment. The user interface 200 includes an analysis data specifying unit 201 that specifies data to be analyzed, a checkpoint specifying unit 202 that specifies a processing unit of analysis target data, and a minimum number of repetitions of a time series pattern to be extracted (minimum number of repetitions). Minimum repeat count input unit 203 for designating, minimum appearance frequency input unit 204 for designating the minimum value of appearance frequency (minimum appearance frequency), execution button 205 for instructing execution of processing, extracted time series pattern and its evaluation value Is displayed on the result display unit 206.

  The user designates the analysis target data with the analysis data designation unit 201, extracts the processing unit into the checkpoint designation unit 202, and extracts the minimum number of repetitions of the time series pattern to be extracted into the minimum repetition number input unit 203. Are input to the minimum appearance frequency input unit 204, respectively. Then, a time series pattern extraction process is started by the execution button 205.

  In the extracted time series pattern, a list of item sets constituting the time series pattern, a statistical value of the number of repetitions as an evaluation value of the time series pattern, and an appearance frequency are displayed on the result display unit 206. In the result display unit 206, the time series pattern is displayed using a table format, but it may be displayed by a transition diagram having the item set constituting the time series pattern as a node.

  The analysis data specifying unit 201, the checkpoint specifying unit 202, the minimum repetition count input unit 203, and the minimum appearance frequency input unit 204 correspond to the input device 104, and the result display unit 206 corresponds to the output device 105. It goes without saying that the input device 104 and the output device 105 can be formed as an integrated configuration by using a display or the like that functions as a touch panel.

  FIG. 3 is a diagram illustrating an example of a flow of an operation by the user and an operation by the system in the time-series pattern extraction process of the present embodiment.

  First, the user designates the data to be analyzed, inputs the processing unit of the analysis target data, and inputs the minimum repetition frequency and minimum appearance frequency of the time series pattern to be extracted (301). Next, the execution of time series pattern extraction processing is started by instructing execution (302).

  The data analysis system acquires the analysis data, the data processing unit, the minimum number of repetitions, and the minimum appearance frequency simultaneously with the execution instruction, stores the acquired data in the memory 102, and executes the execution program stored in the memory 102 by the processor 101 (303). . The execution program extracts frequent repeated time series patterns by reading time series data from the storage device 103, counting the number of repetitions, and counting the appearance frequency (304). The details of the frequent repeat time series pattern extraction processing procedure will be described later. Finally, the extracted time series pattern is output to the output device 106 (305). The user checks the time series pattern output to the output device 106 (306), thereby ending the time series pattern extraction process.

  FIG. 4 is a flowchart for explaining the overall processing procedure of the time-series pattern extraction process in this embodiment.

  First, in response to the input process (301), the user inputs the analysis data to be analyzed, the processing unit of the analysis data, the minimum number of repetitions, and the minimum appearance frequency to the input device 104 (401). The processor 101 holds the input analysis target data, the analysis processing unit, the minimum number of repetitions, and the minimum appearance frequency in the memory 102. In the memory 102, the analysis target data is stored as the database name or file name, the data processing unit, the minimum number of repetitions, and the minimum appearance frequency are stored as numerical values in the setting value information 107, and the head of the data position is stored in the checkpoint information 111. 0 is set (401).

  Next, in response to the execution program processing (304), the processor 101 executes the execution program 106 stored in the memory 102, the frequent repeated item extraction processing 402, and the setting of a time-series pattern to be searched. Frequently repeated time-series patterns are extracted by 403, time-series data reading processing 404 from the storage device 103, repeated pattern counting processing 405, and appearance frequency counting processing 406. The extracted time series pattern is output to the output device 105 (409).

  The frequent repeated item extraction process 402 in FIG. 4 reads time-series data from the storage device 103, counts the number of repetitions of each item for each time-series data, and counts the appearance frequency, thereby minimizing the number of repetitions and the minimum appearance frequency. Extract all items that satisfy both conditions.

  FIG. 5 is a flowchart for explaining in detail the procedure of the frequent repeated item extraction process 402 in FIG. First, the item information 109 is initialized and emptied (501). The item information 109 holds items, time series data IDs, repetition times in the time series data of the time series data IDs, and count values for the item to be searched in a table format or a list. Next, one time-series data is read from the storage device 103, and the time-series data ID and time-series data are held in the time-series data information 109 (502). Next, the number of repetitions of each item appearing in the time series data is counted (503), and for an item that is greater than or equal to the minimum number of repetitions, the item, the time series data ID of the time series data, the time series data ID of the item A set of 0 that is the initial value of the number of repetitions and count value in the time series data is registered in the item information (504). The same processing is repeated for all time series data (505).

  When the processing for all the time series data is completed, the number (appearance frequency) of the time series data ID registered in the item information 109 is counted for each item registered in the item information 109 (506). A set of an item whose frequency is less than the minimum appearance frequency, a time series ID, the number of repetitions, and a count value is deleted from the item information (507).

  When the frequent repeated item extraction process 402 described with reference to FIG. 5 is completed, the items, the time series data ID, the number of repetitions in the time series data, and the count value of all items that satisfy both the minimum number of repetitions and the minimum appearance frequency. A set of initial values is stored in the item information 109 of the memory 102. Here, it is not necessary to store the time series data ID in the time series data information 108 for the time series data ID for which the number of repetitions has been completed, only the time series data ID is stored and the time series data is deleted. It doesn't matter.

  Returning to FIG. 4, the next candidate search time series pattern is set (403). The candidate search time-series pattern is composed of two or more items, and is a time-series pattern whose number of repetitions and appearance frequency are unknown in each time-series data. The search time series pattern information 110 holds a set of a time series pattern, a time series data ID, an initial value of the counted number of repetitions, and an initial value of the position of the counted time series pattern. Here, the search time series pattern information 110 is used to count the number of repetitions in each time series data of the time series pattern to be searched. Note that 0 is set for both the initial value of the counted number of repetitions and the initial value of the counted time series pattern position.

  4 reads time-series data from the storage device 103, and stores the read time-series data in the time-series data information 108 of the memory 102.

  FIG. 6 is a flowchart for explaining in detail the procedure of the time-series data reading process 404 in FIG. First, when the time series data of each time series data ID of the time series data information 108 in the memory 102 is held, only the time series data ID is held and the time series data is deleted (601). Next, the data position of the current time series data is read from the check point information 111 (603), and for each time series data ID stored in the time series data information 108, the analysis processing unit of the set value information 106 from the current data position. Is read from the storage device 103 (604), the item registered in the item information 109 is selected, and the time series data information 108 is used as the time series data of the time series data ID. (605), and for each item appearing in the time series data, the number of times the item appears is added to the count value of the item information 109 (606). The same processing is repeated for all time series data (607). When the processing for all the time series data is completed, the time series data information 108 holds the time series data of each time series data ID for the processing unit of the data of the setting information 106.

  Returning to FIG. 4 again, the repeated pattern counting process 405 is performed next. The repetitive pattern counting process 405 uses the time series data information 108, the item information 109, and the search time series pattern information 110 in the memory 102, and repeats the time series pattern held in the search time series pattern information 110 in each time series data. Count up the number of times.

FIG. 7 is a flowchart for explaining in detail the procedure of the repetitive pattern counting process 405 for the one time series data of FIG. For each time series pattern held in the search time series pattern information 110 (701), the number of repetitions is counted (702), and the number of repetitions of the search time series pattern in the time series data is counted. Next, an upper limit value of the number of repetitions in the time series data of the search time series pattern is calculated (703). The upper limit value of the number of repetitions is based on the number of repetitions counted in the time series data, the counted position in the time series pattern, the number of repetitions in the time series data and the count value of each item constituting the search time series pattern Is calculated by the following equation.

When the counted pattern position is the initial value of the time series pattern:
(Upper limit number of repetitions) = (Number of repeated repetitions) + min {Number of repetitions of item a-Count value of item a}
Here, a is an item constituting a search time series pattern.
When the counted pattern position is not the initial value of the time series pattern:
(Maximum number of repetitions) = (Number of repeated repetitions) + min {Number of repetitions of item a-Count value of item a} + 1
Here, a is an item constituting a search time series pattern.

  The exact value of the number of repetitions of the time series pattern is the sum of the number of repetitions in the time series data for which the process for counting the number of repetitions has been completed up to now and the number of repetitions in the time series data for which the process for counting the number of repetitions has not been processed. . When the search time series pattern spans time-series data that has been counted and unprocessed time-series data, it is necessary to add 1 to the above-mentioned sum. The exact value of the number of iterations of the search time series pattern in the time series data for which the number of iterations has not been processed is unknown until the counting process is completed.

  However, the number of repetitions of a time series pattern has a property that the number of repetitions of a certain time series pattern does not exceed the number of repetitions of each item constituting the time series pattern. Therefore, the number of repetitions of the search time series pattern in the unprocessed time series data cannot be higher than the minimum value of the number of repetitions of items constituting the search time series pattern in the unprocessed time series data. Formula 1 uses the above-mentioned property of the number of repetitions of the time-series pattern, and configures the upper limit value of the number of repetitions in the time-series data, the number of repetitions counted in the time-series data, the counted position, and the search time-series pattern It is a mathematical formula calculated from the number of repetitions and the count value in the time-series data of each item.

  When the value calculated by Equation 1 does not satisfy the minimum number of repetitions, even if the search time series pattern is counted up to the end including the unprocessed time series data of the time series data ID, the search time series Since it can be seen that the pattern does not satisfy the minimum number of repetitions in the time series data, information corresponding to the time series data of the search time series pattern is deleted from the search time series pattern information (705). The counting process of the search time series pattern in the time series data is omitted.

  FIG. 8 is a flowchart for explaining in detail the procedure of the repeat count counting process 702 in one time series data for one search time series pattern of FIG.

  First, the pattern position that has been counted in the time-series data ID to be processed for the time-series pattern to be searched is acquired, and if the acquired pattern position is not the initial value, the next pattern position is set as the current pattern position, and the initial value In the case of (1), the head pattern position is set to the current pattern position (801). The data position of the time-series data at which the repetition count starts is set at the head of the time-series data (802). Thereafter, in order from the item set at the current data position (803), the data position including the item set at the current pattern position of the search time series pattern is searched (804).

  If the item set at the current data position of the time series data does not include the item set at the current pattern position of the search time series pattern, it is checked whether the current data position is the end of the time series data (809). The current data position is shifted backward by one, and the process (804) and subsequent steps for checking whether the item set at the current data position of the time series data includes the item set at the current pattern position of the time series pattern are repeated. In the case of the end, the process is terminated.

  If the item set at the current data position of the time series data includes the item set at the current pattern position of the search time series pattern, it is checked whether the current pattern position is the end of the search time series pattern (805). In the case of the end, the number of repeated iterations counted in the search time series pattern information 110 is increased by 1 (806), and the top pattern position is set to the pattern position of the counted time series pattern (807). If it is not the end, the current pattern position is set to the pattern position of the time series pattern already counted in the search time series pattern information 110 (808). It is checked whether the current data position is the end of the time series data (809). If not, the current data position is shifted backward by one (810), and the item set of the current data position of the time series data is The process returns to the process (804) for checking whether the item set at the current pattern position of the time series pattern is included. When the processing is completed to the end, the current pattern position is set to the time-series pattern position counted in the time-series data ID of the search time-series pattern information 110 (811), and the process ends.

  Returning to FIG. 4 again, the appearance frequency counting process 406 is performed next. The appearance frequency counting process 404 counts the number of appearances of the search time series pattern.

  FIG. 9 is a flowchart for explaining the procedure of the appearance frequency counting process 404 in detail. For each search time series pattern registered in the search time series pattern information 110 of the memory 102 (901), the number of types of time series data ID of the search time series pattern is counted (902), and the minimum appearance of the set value information 107 Search time series patterns that are less than the number of times are deleted from the search time series pattern information 110 (904). A search time series that may satisfy both the minimum repetition frequency and minimum appearance frequency conditions in the search time series pattern information 110 at the time when the same processing for all search time series patterns registered in the search time series pattern information is completed. Only patterns are registered.

  Returning to FIG. 4 again, next, the time series data ID not included in the search time series pattern information 110 is deleted from the time series data information (407).

When the above processing is completed up to the end of all time series data, the statistical value of the number of repetitions and the statistical value of the appearance frequency are extracted as the time series pattern extracted from the search time series pattern registered in the search time series pattern information 110. (409). The search time-series pattern information 110 stores the exact value of the number of repetitions in each time-series data of each time-series pattern, so that it is possible to calculate a statistical value related to the number of repetitions, and the appearance of each time-series pattern. Since the frequency and the total number of time series data are known, it is possible to calculate the statistical value of the appearance frequency.

  In the result display unit 206 of FIG. 2 of the output device 105 of FIG. 1, the average value, maximum value, and minimum value are used as the statistical values of the number of repetitions, and the frequency and the ratio to the total number of time series data as the statistical values of the appearance frequency Shown as an example.

  FIG. 14 is an image diagram of processing units by checkpoints of time series data read from the storage device 103 for the time series data reading process 404, the repetition pattern counting process 405, and the appearance frequency counting process 406 in the flowchart of FIG. 4 of the present embodiment. It is.

  One time-series data is indicated by a single straight line. In the first time-series data reading process, first, time-series data from the beginning of data 1 to the first check point is read. A repeated pattern counting process is performed on the read time-series data. When the repeated pattern counting process for data 1 is completed, the time series data from the beginning of data 2 which is the next time series data to the first check point is read, and the repeated pattern counting process is performed. Appearance frequency counting processing is performed when the time-series data reading processing from the top to the first check point and the repeated pattern counting processing are completed for all time-series data.

  In the first time-series data reading process, time-series data from the beginning to the first check point is read for each time-series data. If the result of the appearance frequency counting process is greater than or equal to the minimum appearance frequency, 2 times from the first checkpoint for time series data where the upper limit of the number of repetitions in the repeated pattern counting process up to the first checkpoint is greater than or equal to the minimum number of repetitions. Time series data reading processing and repetitive pattern counting processing are performed up to the first check point, and appearance frequency counting processing is performed when all time series data is completed.

  If the result of the appearance frequency counting process is greater than or equal to the minimum appearance frequency, the same process is repeated. It is possible to obtain the number of repetitions and the appearance frequency in the time-series data that is equal to or greater than the minimum number of repetitions when the last checkpoint is completed. If the result of the appearance frequency counting process up to a checkpoint in the middle is less than the minimum appearance frequency, it can be seen that the search time series pattern does not satisfy the minimum appearance frequency. The process is omitted. Further, for each time series data, when the upper limit of the number of repetitions as a result of the repetition pattern counting process up to a checkpoint in the middle is less than the minimum number of repetitions, the search time series pattern is the minimum number of repetitions in the time series data. Therefore, the processing is omitted for the time series data after the checkpoint in the middle. As a result, it is possible to avoid processing for a search time series pattern that is not a frequently repeated time series pattern, and to reduce the search processing load.

  As described above, according to this embodiment, it is possible to obtain a time series pattern that satisfies both the minimum value of the number of repetitions in each time series data and the minimum value of the appearance frequency in all time series data. Further, in this embodiment, a search that cannot satisfy the minimum number of iterations by calculating the upper limit value of the number of iterations in the course of the analysis process and counting the appearance frequency using the calculated upper limit value of the number of iterations. By reducing the number of iterations of the search time series pattern that cannot satisfy the minimum number of occurrences of the time series pattern, the range of time series data subject to the number of iterations is reduced, reducing the analysis processing load. It becomes possible to do.

  In the first embodiment described above, the number of repetitions and the appearance frequency are used, but the repetition rate that is a quotient of each time series data length or period of the number of repetitions and the number of all time series data of the appearance frequency. The analysis process can be similarly performed by using a certain appearance rate (support level).

  In this embodiment, the processing unit of the time series data is input to the checkpoint specifying unit 202. However, by setting a predetermined value in the checkpoint information of the set value information 107 of the memory 102 in advance, the user can It is also possible to omit the input of the processing unit.

  Next, as a second embodiment, taking credit card usage data as an example, the processing of the execution program 106 of the data analysis system in FIG. 1 and various information stored in the memory 102 will be described. Here, Table 3 shows an example of usage data of the credit card to be analyzed in the present embodiment. As is apparent from Table 3, it is assumed that the data of the time series data number 20 is stored in the storage device 103. In the input device 104, it is assumed that the data processing unit is 5, the minimum number of repetitions is 3, and the minimum appearance frequency is 5 and is stored in the setting value information 107 of the memory 102.

  The processor 101 executes the execution program 106 stored in the memory 102 and first performs a frequent repeated item extraction process 402. For example, when this item extraction process 402 reads card01 time-series data from the storage device 103 and counts the number of repetitions of each item, the item information 109 in the memory 102 stores (store A, card01, 3, 0). ), (Store B, card 01, 4, 0), (Store C, card 01, 3, 0), (Store D, card 01, 1, 0), (Store E, card 01, 1, 0), (high value payment, card01, 3, 0) are stored. Since the minimum number of appearances is 3, the items “Store D” and “Store E” are deleted from the item information, and the item information 109 in the memory 102 includes (Store A, card 01, 3, 0), (Store B, card01, 4, 0), (store C, card01, 3, 0), (high value settlement, card01, 3, 0).

  Here, in the search time series pattern setting process 403, <(store A) (store B, high-priced payment)> and <(store C) (store A)> are used as search time series patterns whose number of candidate repetitions is unknown. And <(Store C) (Store B)> are set.

  Next, in the time-series data reading process 404 of the execution program 106, the processor 101 reads the time-series data stored in the storage device 103 and stores it in the time-series data information 108 of the memory 102 for each processing unit.

  Other than the frequently repeated items are deleted from the read time-series data and stored in the time-series data information 108, the number of repetitions of each item is counted, and the number of repetitions information of the item information 109 is updated. For example, in the time-series data of card01, as the time-series data in the first time-series data reading process, <(store A) (store C) (store A) (store B, high-value settlement) (store E) (store C)> Is read, items other than the frequently repeated items are deleted, and card01, <(store A) (store C) (store A) (store B, high-value settlement) (store C)> is held in the time-series data information 108. . In addition, the count value of the item information 109 is (store A, card 01, 3, 2), (store B, card 01, 4, 1), (store C, card 01, 3, 2) (high payment, card 01, 3, 1 ) And update.

  Next, in the repeated pattern counting process 405, the number of repetitions in each time series data of the search time series pattern is counted. For example, time series data of card01 will be described. First, when the search time-series pattern <(store A) (store B, high-price payment)> is counted, the number of repetitions that have been counted from the search time-series pattern information 110 of the memory 102 as card01 information of the search pattern. 0, 0 is read as the pattern position of the counted time series pattern.

  Thereafter, for the time series data of card01, the data position where the item set (store A) of the first pattern position of the search pattern appears in order from the item set of the top data position of the time series data information 108 of the memory 102 is searched. The data position is detected first. Since the item set at the first data position of the search pattern is not the end, the position of the time-series pattern in which the search pattern information is counted is updated to 1.

  Next, the data position where the item set (store B, high-priced payment) at the second pattern position of the search pattern appears is searched for at the second and subsequent data positions of the time series data, and detected at the fourth data position. Since the item set at the second pattern position of the search pattern is the end, the number of times the search pattern information has been counted is incremented by 1, and the pattern position of the time series pattern is updated to 0. Again, the data position where the first item set (store A) of the search pattern appears is searched from the fifth data position of the time series data, but even if the data position is searched up to the fifth data position which is the end of the time series pattern. Since it is not detected, the pattern position of the counted time series pattern whose time series data ID of the time series pattern <(store A) (store B, expensive payment)> of the search time series pattern information 110 is card01 is updated to 1. The number counting process ends.

  Further, when the search pattern <(store C) (store A)> is counted, the number of repeated iterations counted is 0 as the information of card01 of the search pattern from the search time series pattern information 110 of the memory 102. Pattern position 0 is read out. Thereafter, for the time series data of card01, the data position where the item set (store C) of the first pattern position of the search pattern appears in order from the item set of the first data position of the time series data information 108 of the memory 102 is searched. The second data position is detected. Since the item set (store C) at the first pattern position of the search pattern is not the end, the current pattern position is updated to 2 (808 in FIG. 8).

  Next, the data position where the item set (store A) at the second pattern position of the search pattern appears is searched after the third data position of the time series data and detected at the third data position. Since the item set (store A) at the second pattern position of the search pattern is the end, the number of times the search pattern information has been counted is incremented by 1 (same 806), and the current pattern position is updated to the top (same 807). .

  Again, the data position where the item set (store C) at the first pattern position of the search pattern appears is searched after the third data position of the time-series data and detected at the fifth data position. Since the item set at the first pattern position of the search pattern is not the end, the current pattern position is updated to 2. Again, an attempt is made to find the data position at which the item set (store A) at the second pattern position of the search pattern appears at the fifth and subsequent data positions of the time series data. Is the end (same as 809), the pattern position of the time-sequential pattern counted in the search pattern information 110 is set to 2 (same as 811), and the repeat count counting process 702 is terminated.

  Next, calculation 703 of the upper limit value of the number of repetitions is performed in the flow of FIG. For example, for the search pattern <(store A) (store B, high-price payment)> in card01, the search time-series pattern information 110 includes (<(store A) (store B, high-price payment)>, card01, 1, 0). The item information 109 holds (Store A, card 01, 3, 2), (Store B, card 01, 4, 1), and (high value settlement, card 01, 3, 1). The value is calculated as 2 (= 1 + 1).

  Since the calculated upper limit value is less than the minimum number of repetitions (3 in this embodiment), it can be seen that the search pattern cannot exceed the minimum number of repetitions. By deleting the search pattern information from the search pattern information 110, the processor 010 cancels the count counting process for the card01 search pattern, and omits the repeat counting process after the second and subsequent time-series data reading processes.

  For the search pattern <(store C) (store A)>, the search pattern information is (<(store C) (store A)>, card01, 1, 1), and the item information is (store A, card01, 3, 2) and (Store C, cards 01, 3, 2) are held, and therefore the upper limit value is calculated as 3 (= 1 + 1 + 1) according to Equation 1. Since the calculated upper limit value is equal to or greater than the minimum number of repetitions, the search pattern may be equal to or greater than the minimum number of repetitions. Therefore, it is possible to omit the repeat counting process after the second and subsequent time series data read processing. Absent.

  Next, the appearance frequency counting process 406 of the repeated time series pattern extraction process of FIG. 4 is performed. The processor 101 counts the number of types of the time series data ID of each time series pattern from the time series pattern and time series data ID stored in the search time series pattern information 110 of the memory 102, and the time series pattern becomes less than the minimum appearance frequency. Is deleted from the search pattern information. For example, for the search pattern <(store C) (store B)>, the search time series pattern information 110 includes (<(store C) (store B)>, cards 01, 1, 1), (<(store C) ( Store B)>, card 05, 2, 1), (<(store C) (store B)>, card 08, 2, 0) are stored. It can be seen that the appearance frequency of the search pattern is 3, which does not satisfy the minimum appearance frequency (5 in this embodiment). In this case, the processor 101 deletes information including the search pattern from the search pattern information 110. The time series pattern deleted from the search pattern information 110 can omit the repeat counting process after the second time series data reading process.

  Following the appearance frequency counting process 406, the time-series data information 108 is updated 407. The time series data ID that needs to be subjected to the repeat count counting process is held in the search time series pattern information 110. Since the time series data ID that is not held in the search time series pattern information 110 does not need to perform the process of counting the number of repetitions, the time series data reading process for the second and subsequent times is unnecessary. If there is a time series data ID that is not held in the search time series pattern information 110, the time series data ID is deleted from the time series data information 108. For example, when card02, card04, and card07 are not held as time-series data IDs, they are deleted from the time-series data information 108.

  The above processing is repeated until the end of each time series data. For example, in the second time-series data reading process of the present embodiment, the 6th to 10th ranking positions of the time-series data are read, and for card01, <(store B) (store B, expensive payment) (store A) (Store C) (Store B, high-value payment)> is stored in the memory 102 as time-series data information 108, and when the search pattern <(Store C) (Store A)> is counted, the search time-series pattern in the memory 102 From the information 110, as the information of card01 of the search pattern, the counted number of repetitions is 1, and the position of the counted time series pattern is 1. Thereafter, for the time series data of card01, a process of searching for a rank position where the second item set (store A) of the search pattern appears in order from the top item set of the time series data information 108 of the memory 102 is started.

  As described above in detail, according to the processing of the present embodiment, it is possible to extract a time series pattern that satisfies both the minimum number of repetitions and the minimum appearance frequency conditions while avoiding unnecessary repetition number counting processing.

  In the case of the present embodiment, for example, for the search pattern <(store A) (store B, high-price payment)> in card01, the repeat count counting process after the second time-series data reading process can be omitted. Further, for example, the time-series data of card02, card04, and card07 can omit the second and subsequent time-series data read processing.

  Next, as a third embodiment, a case will be described in which the data analysis system sets data to be analyzed in which each time-series data is separated. There may be a case where it is desired that a time series pattern extending over a predetermined phrase break in one time series data is not included in the time series data. For example, when an action pattern in a unit of one day is considered in the analysis of a person's action pattern, the action pattern across the date must not be counted.

  FIG. 10 is a diagram illustrating an example of a system configuration of the third embodiment. In this system, the memory 102 having the system configuration shown in FIG. 1 stores the time-series data delimiter conditions in the analysis target data as delimiter conditions 1001 in the form of conditional expressions.

  FIG. 11 shows an example of a user interface of this embodiment. The user interface 1100 has a configuration in which a delimiter condition setting unit 1101 for setting delimiter conditions for time-series data in analysis target data is added to the user interface 200 of FIG. The user designates the analysis target data with the analysis data designation unit 201, extracts the processing unit into the checkpoint designation unit 202, and extracts the minimum number of repetitions of the time series pattern to be extracted into the minimum repetition number input unit 203. The minimum appearance frequency is input to the minimum appearance frequency input unit 204 and the time series data delimiter condition is input to the delimiter condition setting unit 1101. A time series pattern extraction process is started by the execution button 205. In the extracted time series pattern, a list of item sets constituting the time series pattern, a statistical value of the number of repetitions as an evaluation value of the time series pattern, and an appearance frequency are displayed on the result display unit 206.

  FIG. 12 is a diagram illustrating a flow of an operation by the user and an operation by the system in the time-series pattern extraction process of the present embodiment. First, the user designates data to be analyzed, inputs the processing unit of the analysis target data, sets the minimum number of repetitions and the minimum frequency of appearance of the time series pattern to be extracted, and the time series data delimitation conditions using the input device 104 Input from the unit 1101 (1201). The subsequent processing is the same as in FIG.

  Overall processing sequence of time series pattern extraction when a condition is set for the time series data to be analyzed, frequent repeated item extraction processing, candidate search time series pattern setting processing, time series data reading processing, appearance frequency counting The processing is the same as the processing procedure described above. In the time-series pattern extraction process when the time-series data is divided and the condition is set, the repetition count counting process in one time-series data with respect to one search time-series pattern in the repetition pattern counting process is different from the above-described processing procedure. .

  FIG. 13 is a flowchart for explaining in detail the procedure in the case where a delimiter condition is set in the time series data to be analyzed in the repetition count counting process 702 in one time series data for one search time series pattern. The content of the processing is the same from the processing 801 for setting the pattern position for counting up the number of repetitions in FIG. 8 to the processing 809 for checking whether processing has been performed up to the end of the time series data. It is checked whether or not the current data position is the end of the time series data (809). If it is not the end, it is checked whether or not the current data position satisfies the delimitation condition 1001 of the setting value information 107 in the memory 102 (1301). ).

  If the division condition is satisfied, the current pattern position is returned to the first pattern position (1302), the current data position is shifted backward by one (810), and the item set at the current data position of the time series data is the time series pattern. The process of checking whether the item set at the current pattern position is included (804) and subsequent steps are repeated. If the delimiter condition is not satisfied, the current data position is shifted backward by one (810), and it is checked whether the item set at the current data position of the time series data includes the item set at the current pattern position of the time series pattern ( Step 804) and subsequent steps are repeated. The case of the end is the same as described above.

  As described above, according to the present embodiment, the time-series pattern segmentation condition is set, and when the search time-series pattern repetition count is counted, the search pattern By returning the pattern position for counting up the number of repetitions to the top, it is possible to avoid counting up the number of repetitions when straddling the punctuation of the time series pattern. This makes it possible to extract a time series pattern even when a break is set in the time series data.

  As a fourth embodiment, a data analysis system for analyzing access log data of a website will be described. That is, taking the access log data of the Web site as an example, the repetition count counting process 702 for one time-series data for one search time-series pattern performed in the execution program 106 in the data analysis system described above will be described. In the case of Web access log data, one record means one access of a user, the time series data ID is the user ID, the time stamp is the access date and time, and the event is the URL of the accessed page. The Web site access log has a session number representing a series of access units, and records having the same session number are accesses in the same session.

  For example, user01 of the data in Table 4 has three sessions with session numbers 100, 101, and 102. Session number 100 means that page A is accessed first, then page B is accessed, and page C is accessed last. Here, it is assumed that the access log data to be analyzed is the data shown in Table 4 and stored in the storage device 103. In the input device 104, it is assumed that “session number of i-th record ≠ session number of (i + 1) -th record” is set as a delimiter condition, and is stored in the setting information of the memory 102.

  It is assumed that the search pattern <(page A) (page B)> in the time series data of user01 is counted as the repetition count counting process 702 in one time series data for one search time series pattern. (100, Page A) (100, Page B) (100, Page A) (101, Page A) (101, Page B) (101, Page D) (101, page E) (101, page D) (102, page D) are held, and the number of repeated repetitions counted from the search pattern information 110 of the memory 102 as information of user01 of the search pattern. 0, the pattern position of the counted time series pattern is read as 0 It was to be. In this embodiment, the search is performed in order from the data position including the item set (page A) at the head pattern position of the search pattern and the head data position of the time-series data information 108 in the memory 102 (corresponding to 803 in FIG. 13). .

  First, it is detected that the item set (page A) at the head pattern position of the search pattern appears in the item set at the head data position of the time-series data. Since the current pattern position is not the end of the search pattern (No in 805), the current pattern position is set to 2 which is the next pattern position (808).

  Next, it is checked whether the current data position is at the end (step 809). Since the current data position is not at the end, it is checked whether the current data position is immediately after (step 1301). As described above, since the session number is set to be different as the phrase break condition in the delimiter condition 1001 of the memory 102, the session number at the current data position is compared with the session number at the next data position, and both are equal to 100. Since it is the session number, the current data position is set to 2 (810), and it is checked again whether the item set at the current pattern position is included in the current data position (804). In this embodiment, when the current pattern position is 2, it is detected that the current data position is 2 and the current pattern position item set (page B) appears at the current data position, and the search pattern information 110 in the memory 102 is counted. The number of repeated repetitions is updated to 1 (same 806), the current pattern position is set to the top (same 807), and the current data position is set to 3.

  Next, since the item set (page A) at the current pattern position is not included in the item set (page C) at the current data position, the current data position is shifted to the next data position. Here, since the session number of the current data position is 100 and the session number of the next data position is 101, it is detected that there is a partition (1301), and the current data position is shifted to the next data position. Before, the head pattern position is set to the current pattern position (1302). As a result, it is possible to avoid counting the number of repetitions of the time-series pattern across the session numbers 100 and 101. Thereafter, the repeat count counting process is continued until the end of the time series data.

  In this embodiment, a list of session number and item set pairs is stored in the time series data information, and a conditional expression is set as a delimiter condition. However, the phrase break condition is a predetermined symbol, and the time series data The analysis process can be similarly performed by using data in which a predetermined symbol is added to a phrase break. For example, in the data of Table 4, if the session number is different and set as a break, and the symbol representing the phrase break is set to “.”, The analysis processing is performed in the same manner as the present embodiment using the time series data of Table 5 It becomes possible to do.

  The present invention described above in detail relates to a data analysis system and method for databases and data warehouses, and is particularly useful as a data mining technique for clarifying regularity of the appearance order of data by analyzing records in the database. is there.

DESCRIPTION OF SYMBOLS 100 ... Computer 101 ... Processor 102 ... Memory 103 ... Storage device 104 ... Input device 105 ... Output device 106 ... Execution program 107 ... Setting value information 108 ... Time series data information 109 ... Item information 110 ... Search pattern information 111 ... Checkpoint information 1001 ... Separation condition 1101 ... Separation condition setting unit 200 ... User interface 201 ... Analysis data designation unit 202 ... Check point designation unit 203 ... Minimum repetition count input unit 204 ... Minimum appearance frequency input unit 205 ... Execution button 206 ... Result display Department.

Claims (15)

A data analysis system for analyzing data consisting of a plurality of sets of an event, an ID to which the event belongs, and information indicating an order relation between events,
The computer includes the data to be analyzed, a storage unit that stores an execution program that performs analysis processing, and a processing unit that executes the execution program,
The processor is
A first step of storing the data in which events having the same ID are arranged according to the order relation in the storage unit as time-series data;
A second step of counting the number of repetitions in each of the time series data for a time series pattern comprising overlapping permutations of the events;
A third step of counting the number of the time series data in which the number of repetitions is equal to or greater than a predetermined number;
A fourth step of extracting a time series pattern in which the counted number of time series data is equal to or greater than a predetermined number;
A data analysis system characterized by executing A data analysis system according to claim 1, wherein
The processor is
Providing check points at predetermined intervals in each of the time series data, and storing the time series data in the storage unit in a range from a check point to a next check point in the first step;
Counting the number of repetitions from the time-series data in the range stored in the storage unit in the second step for the number of repetitions in each of the time-series data of the time-series pattern with an unknown number of repetitions;
The number of repetitions of the time-series pattern in the time-series data is added by adding the number of times of counting in the range in which the counting is performed and the number of repetitions of the events constituting the time-series pattern in the time-series data that has not been counted. Calculating an upper limit value of
In the third step, counting up the number of time-series data of a time-series pattern in which the upper limit of the number of repetitions is a predetermined number or more;
Repeating the above steps for a time series pattern in which the counted number of time series data is greater than or equal to a predetermined number;
A data analysis system characterized by executing The data analysis system according to claim 1,
The time series data is data having a delimiter,
The processor is
Executing the second step, the third step, and the fourth step for each of the divisions;
A data analysis system characterized by this. The data analysis system according to claim 1,
The time series data is access log data including a session number indicating an access unit to a website.
The processor is
Executing the second step, the third step, and the fourth step for each of the access log data having the same session number;
A data analysis system characterized by this. The data analysis system according to claim 1,
The computer further includes an output unit,
The processing unit outputs the extracted time series pattern together with the number counted in the second step and the number counted in the third step to the output unit.
A data analysis system characterized by this. The data analysis system according to claim 1,
The computer further includes an input unit capable of inputting the predetermined number of times and the predetermined number.
A data analysis system characterized by this. A data analysis system according to claim 3, wherein
The computer further includes an input unit capable of inputting a condition for setting the break in the time series data.
A data analysis system characterized by this. A data analysis method for analyzing data in which a plurality of sets of information indicating an event, an ID to which the event belongs, and an order relationship between the events is stored by a computer including a processing unit and a storage unit,
Data in which the events having the same ID are arranged according to the order relationship is time series data, and a duplicate permutation in which one or more of the events are arranged in a forward direction is a time series pattern,
The processor is
In order to extract a frequent repeated time series pattern that is the time series pattern repeated a predetermined number of times or more in each time series data in a predetermined number or more of the time series data,
For the time-series pattern whose number of repetitions is unknown, counting the number of repetitions in each of the time-series data;
Counting the number of the time-series data in which the number of repetitions is a predetermined number or more as an appearance frequency;
Extracting the time series pattern in which the appearance frequency is a predetermined number or more;
The data analysis method characterized by performing. A data analysis method according to claim 8, comprising:
The time series data is data in which a break exists,
The processing unit executes each of the steps for the time-series data for each of the divisions,
A data analysis method characterized by the above. A data analysis method according to claim 8, comprising:
The time series data is access log data having a session number indicating an access unit to a website.
The processing unit executes each of the steps for each access log data having the same session number.
A data analysis method characterized by the above. The data analysis method according to claim 8, comprising:
The calculator further includes a display unit,
The processing unit displays the number of repetitions and the appearance frequency corresponding to the extracted time-series pattern on the display unit.
A data analysis method characterized by the above. A data analysis method for performing analysis processing of data in which a plurality of sets of information indicating an event, an ID to which the event belongs, and an order relationship between the events is stored by a computer including a processing unit and a storage unit ,
The processor is
Data in which events having the same ID are arranged in accordance with the order relationship is time series data, and a duplicate permutation in which one or more events are arranged in the forward direction is a time series pattern. In order to extract a frequently repeated time series pattern that is the time series pattern repeated a predetermined number of times or more in each series data,
A first step of setting checkpoints at predetermined intervals in each of the time series data;
A second step of counting the number of times the time series pattern is repeated in a range from a check point to the next check point for each time series data for the time series pattern in which the number of repetitions in each time series data is unknown;
The upper limit value of the number of repetitions of the time series pattern in the time series data is calculated from the sum of the number of repetitions up to the already counted check point and the number of repetitions of each event appearing after the check point. Steps,
A fourth step of counting the number of time-series data for which the calculated upper limit value is equal to or greater than a predetermined number of repetitions as the appearance frequency;
A fifth step of extracting a time-series pattern in which the counted appearance frequency is a predetermined number or more;
A sixth step of repeating the second to fifth steps until the last checkpoint for the extracted time-series pattern;
The data analysis method characterized by performing. A data analysis method according to claim 12, comprising:
The calculator further includes an output unit,
The processor is
At the time of processing up to the last checkpoint, the number of repetitions corresponding to the extracted time-series pattern and the appearance frequency are output to the output unit.
A data analysis method characterized by the above. A data analysis method according to claim 12, comprising:
The time series data is data in which a break exists,
The processing unit executes each of the steps for the time-series data for each of the divisions,
A data analysis method characterized by the above. A data analysis method according to claim 12, comprising:
The time series data is access log data including a session number indicating an access unit to a website.
The processing unit executes each of the steps for each access log data having the same session number.
A data analysis method characterized by the above.

Images