KR102053175B1 - Method and system for detecting frequent pattern in graph streams - Google Patents

Abstract

A frequent pattern detection method and a frequent pattern detection system in a graph stream are disclosed. According to an embodiment of the present invention, a method of detecting a frequent pattern in a graph stream includes preprocessing a graph stream including continuously input graph data as the detection of a frequent pattern is requested, and in the preprocessed graph stream. Counting the number of occurrences for each of the patterns identified in the sliding window unit, detecting a frequent pattern frequently occurring in the graph stream among the patterns while moving the sliding window, based on the occurrence frequency; And outputting a list of patterns to the terminal requesting the detection.

Description

Frequency pattern detection method and graph pattern detection system in a graph stream {METHOD AND SYSTEM FOR DETECTING FREQUENT PATTERN IN GRAPH STREAMS}

The present invention relates to a technique for gradually detecting a frequent pattern from a continuously input graph stream using a sliding window method, and unnecessary calculation when detecting a frequent pattern in a later step by using information about the frequent pattern detected in a previous step. The present invention relates to a frequent pattern detection method and a frequent pattern detection system capable of reducing the total calculation amount.

Graph data is a data structure for expressing complex relationships between objects, and is used in various fields such as transportation networks, bioinformatics, chemistry, and social networks. For example, graph data for a traffic network expresses regions as peaks and roads as trunks, and graph data for social networks express users as vertices, and expresses follow and friend relationships as trunks.

Recently, due to the activation of the social network service, the user's friendship changes every second, and the relationship between the sensors also changes from time to time when constructing the sensor network, and the graph data is continuously generated. Such stream data in the form of a graph may be referred to as a graph stream. .

Graph streams are used in various fields such as anomaly detection, real-time trend analysis, and event detection. For example, smart factories are building IoT-based sensor networks to represent the flow of data between sensors in graph streams, process and supply chain intelligence, and anomaly detection, and in social network services, graphing relationships and information consumption patterns between users. The graph stream is used to analyze human networks and analyze real-time trends.

As the graph stream is applied to various fields, researches on various graph stream analysis techniques are being actively conducted.

In particular, studies are being conducted to detect changes in graphs or analyze patterns as dynamic relationships between objects in graphs change. As an example, techniques for analyzing graph streams may vary, such as graph clustering, graph stream classification, subgraph mining, and frequent pattern detection.

Among these, the frequent pattern detection technique is one of analytical techniques frequently used in graph streams and is a method of detecting subgraphs frequently occurring during a specific period. Frequent pattern detection techniques can be used, for example, in smart factories to detect patterns of precursor phenomena for failure prediction, and in social networks to determine patterns of information exchange between users to analyze intimacy between users.

As the utilization of frequent pattern detection in graph streams is increasing, various techniques are actively researched. However, in the existing method, many pointers are used because a tree (DSTree) is constructed to store data for frequent pattern detection. It can take a long time to build a tree. In addition, the conventionally proposed technique does not consider the connectivity between the patterns, and thus simply detects frequent patterns insignificantly, and there is no processing for overlapping operations when detecting the frequent patterns in a sliding window method, thereby reducing the processing speed.

Accordingly, there is a demand for a technique for reducing the total amount of calculation by using prediction information calculated whether the pattern detected in the previous sliding window is frequent or not in the future, and detecting a significant frequent pattern in consideration of the connectivity between the patterns.

According to an exemplary embodiment of the present invention, when detecting a frequent pattern while gradually moving a sliding window in a graph stream input in real time, the existing sliding window method is performed by performing only necessary operations by using information about the frequent pattern detected in the previous sliding window. The purpose of this paper is to solve the problem of redundancy and reduce the total amount of computation.

In addition, the embodiment of the present invention, when the detection of the frequent pattern for one sliding window is completed, whether the frequent pattern is frequent in the next sliding window and the information predicting the number of sliding windows that occur frequently in succession pattern management table It is an object of the present invention to omit an operation for detecting a frequent pattern with respect to a pattern predicted as a frequent pattern even in a subsequent sliding window.

In addition, the embodiment of the present invention is to delete the graph data corresponding to one sliding window in which the frequent pattern detection is completed in the memory unit, so as to secure a memory space for analyzing the new graph data continuously input For the purpose of

In addition, an embodiment of the present invention is to make it possible to check the frequency of more significant patterns by recognizing the patterns connected to each other in consideration of the connectivity between the graph data as one expansion pattern.

According to an embodiment of the present invention, a method of detecting a frequent pattern in a graph stream includes preprocessing a graph stream including continuously input graph data as the detection of a frequent pattern is requested, and in the preprocessed graph stream. Counting the number of occurrences for each of the patterns identified in the sliding window unit, detecting a frequent pattern frequently occurring in the graph stream among the patterns while moving the sliding window, based on the occurrence frequency; And outputting a list of patterns to the terminal requesting the detection.

In addition, the frequent pattern detection system in the graph stream according to an embodiment of the present invention, the pre-processing unit for pre-processing the graph stream consisting of the continuously input graph data as the detection of the frequent pattern, and the pre-processing A frequency pattern detection unit for counting the number of occurrences of each of the patterns identified in the sliding stream unit in the graph stream, and detecting the frequent pattern occurring in the graph stream among the patterns while moving the sliding window based on the occurrence frequency; And an output unit for outputting a list of the frequent patterns to the terminal requesting the detection.

According to an embodiment of the present invention, by using a frequent pattern management table that calculates whether a pattern detected in a previous sliding window will be frequent or not frequently, determine whether to recalculate the next window slide and perform only necessary calculations. By doing so, it is possible to solve the redundancy calculation which is a problem of the existing sliding window method and reduce the total computation amount.

In addition, according to an embodiment of the present invention, using a two-dimensional array structure called DSMatrix it is possible to efficiently store a large number of graph stream data in a small space to analyze the frequent patterns.

In addition, according to an embodiment of the present invention, a pattern that is connected to each other by a vertex or an edge may be recognized as one pattern to detect a more significant pattern.

In addition, according to an embodiment of the present invention, memory space may be secured by deleting the analyzed graph stream data.

1 is a block diagram illustrating a configuration of a frequent pattern detection system in a graph stream according to an embodiment of the present invention.
2 is a diagram illustrating an example of a graph stream continuously input in the frequent pattern detection system in the graph stream according to an embodiment of the present invention.
3A illustrates an example of a matrix configured according to the presence or absence of edges in a frequent pattern detection system in a graph stream according to an embodiment of the present invention.
3B illustrates an example of an initial frequent pattern management table in a frequent pattern detection system in a graph stream according to an exemplary embodiment of the present invention.
FIG. 4 is a diagram illustrating an algorithm for calculating a number of sliding windows in which a frequent pattern will subsequently occur frequently in a frequent pattern detection system in a graph stream according to an exemplary embodiment of the present invention.
5A to 5C are diagrams for describing a process of detecting a frequent pattern using a frequent pattern management table after moving a sliding window in a frequent pattern detection system in a graph stream according to an embodiment of the present invention.
6 is a flowchart illustrating a method of detecting a frequent pattern in a graph stream according to an embodiment of the present invention.

Hereinafter, a frequent pattern detection system in a graph stream according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited or limited by the embodiments. Like reference numerals in the drawings denote like elements.

1 is a block diagram illustrating a configuration of a frequent pattern detection system in a graph stream according to an embodiment of the present invention.

Referring to FIG. 1, the frequent pattern detection system 100 in a graph stream according to an exemplary embodiment of the present invention may include a preprocessor 110, a frequent pattern detector 120, a table recorder 130, and an output unit 140. ), The memory unit 150, and the frequent pattern management table 160.

As the preprocessing unit 110 is requested to detect the frequent pattern, the preprocessor 110 preprocesses the graph stream including the continuously input graph data.

For example, the preprocessor 110 may perform preprocessing to efficiently store the input graph stream in the memory 150.

When the graph stream is input, the preprocessing unit 110 stores the graph stream in the memory unit 150, and when the graph data corresponding to one sliding window (for example, '9') is stored in the memory unit 150. The stored graph data can form a matrix, for example 'DSMatrix'. For example, the preprocessor 110 may store a large amount of graph data in a small space by using a two-dimensional array such as the matrix 310 illustrated in FIG. 3A.

In addition, the preprocessing unit 110 may generate a two-dimensional array structure, for example, 'DSMatrix' so that frequent patterns may be detected in consideration of not only current graph data but also front and rear graph data among consecutively input graph streams. You can save the entered graph data.

In this case, 'DSMatrix' is a boolean type array, so more data can be stored in less space than the existing DSTree, and the presence or absence of edges in the graph data is expressed as '1' or '0', so the graph data can be added or deleted quickly. It is possible to fit the sliding window method.

In the DSMatrix, one window slide may include a predetermined number of graph data (for example, '9') by the terminal in a batch unit.

In addition, if the pre-processing unit 110 stores the graph data defined by the sliding window (for example, '9') in the memory unit 150 storing the graph stream in the order of input, the pre-processing unit 110 stores the graph stream. In the graph data included in the sliding window, a matrix giving a two-dimensional array value of '1' or '0' is formed according to the presence or absence of edges connecting two vertices, and the two-dimensional array value is given as '1'. The subgraph including the edges and two vertices leading to the edges can be identified with the pattern.

For example, referring to FIG. 2, when the graph stream 200 composed of nine graph data a to i corresponding to one sliding window is continuously input, the preprocessing unit 110 graphs in the order of input. The stream 200 may be stored in the memory unit 150 to form a matrix 310 as shown in FIG. 3A.

Here, one sliding window may consist of three batches, each batch may consist of three graph data, and the sliding window may move in one batch unit in the graph stream 200.

Referring to FIG. 3A, the preprocessing unit 110 grants '1' if there are edges 301 to 305 in each of the graph data a to i in the graph stream 200, and gives '0' if not. The matrix 310 can be configured as shown in 3a.

In this case, the preprocessor 110 may express the names (rows) of the respective edges (301 to 305) using the numbers of the two vertices.

For example, the preprocessor 110 expresses the edge connecting the vertex '1' and the vertex '2' as '1/2' 301, and displays the edge ('1/2') 301. ) Are all in the graph data (a to i), so that all column values (Contents) in one row can be given as '1'.

In addition, the preprocessor 110 does not include the edge ('1/3') 302 connecting the vertex ('1') and the vertex ('3') only in the graph data (d). It can be given as '0' and all others as '1'.

The preprocessor 110 may identify a subgraph including an edge having a column value of '1' in the DS matrix 310 and two vertices leading to the edge as a 'pattern'. Through this, the preprocessor 110 may check the occurrence of the pattern consisting of a single edge in the DS matrix 310.

In addition, the preprocessing unit 110 stores the graph data stored in the memory unit 150 every time the graph data is input since the graph streams are continuously input, and deletes the graph data stored in the memory unit 150 when the detection of the frequent pattern is completed. Thus, the frequent pattern can be detected continuously from the graph data to be input next.

The frequent pattern detection unit 120 counts the number of occurrences for each of the patterns identified in the sliding window unit in the preprocessed graph stream, and based on the occurrence frequency, sliding the frequent patterns in the graph stream among the patterns. Detect while moving window.

That is, the frequent pattern detection unit 120 may count the number of occurrences for each pattern in one sliding window, and detect a pattern having the occurrence number more than a threshold as a frequent pattern.

For example, the frequent pattern detector 120 may first determine whether a pattern consisting of a single trunk line is frequent to detect a frequent pattern in a graph stream.

Specifically, the frequent pattern detection unit 120 may add the two-dimensional array values for each pattern, and count the number of frequency in sliding windows (FiS) in which the pattern is generated from the graph data included in the sliding window. .

In this case, when the sliding window is composed of a plurality of batches, the frequent pattern detection unit 120 sums the two-dimensional array values for each of the patterns in the respective batches to generate the patterns in one batch. The number of FiBs may be counted by counting the number of FiBs (Frequency in Batch), and further adding up the number of FiBs counted in each of the plurality of batches.

For example, referring to FIGS. 3A and 3B, the frequent pattern detection unit 120 includes a matrix 310 of FIG. 3A configured in a preprocessing process, within the sliding window of the trunk lines '1/2' 301. In order to count the number of occurrences (FiS number), the number of occurrences (FiB number) in each of the three batches forming one sliding window may be counted and then summed.

The frequent pattern detection unit 120 includes the number of FiBs ('3') in which the trunk line ('1/2') 301 occurs in batch 1 (graph data a, b, and c), and batch 2 (graph data d, e). , f) the number of FiBs generated in '3' and the number of FiBs generated in batch 3 (graph data g, h, i) ('3') are added together, and the number of FiSs ('9') is shown in FIG. 3B. ') Can be counted. Similarly, the frequent pattern detection unit 120 may count the number of FiSs for the other single edges 302 to 305 as shown in FIG. 3B.

The frequent pattern detection unit 120 may detect a pattern having the number of FiSs equal to or greater than a threshold value as a frequent pattern, and include it in the list.

For example, when the threshold value τ is set to '5', the frequent pattern detection unit 120 has trunk lines '1/2', '1/3', and '3' in which the number of FiSs is greater than or equal to the threshold '5' in FIG. 3B. / 4 ') 301, 302, and 305 can be detected as a frequent pattern. In other words, the frequent pattern detector 120 may detect three frequent patterns of a single trunk line from the graph stream 200 of FIG. 2.

As another example, the frequent pattern detection unit 120 may perform an AND operation on the number of occurrences of two patterns consisting of a single edge to count whether or not an extension pattern consisting of two or more edges is frequent.

According to an embodiment, the frequent pattern detection unit 120 selects a plurality of patterns having the same vertices and neighboring each other when the detection of the frequent patterns is completed among the patterns consisting of a single trunk line in the graph stream, and the plurality of patterns An AND operation is performed on each occurrence frequency to count the occurrence frequency for the expansion pattern in which each of the plurality of patterns occurs at the same time, and detect the expansion pattern in which the occurrence frequency is counted above the threshold as a frequent pattern, Can be included.

That is, the frequent pattern detection unit 120 performs an AND operation on the detected two frequent patterns, counts the number of times the two patterns occur at the same time, and detects the frequent occurrence, and continuously expands the pattern in this manner.

For example, the frequent pattern detection unit 120 may select an edge line among three single edges ('1/2', '1/3', and '3/4') 301, 302, and 305 detected as a frequent pattern. By generating an AND operation on the number of occurrences of the '1/2') 301 and the edges ('1/3') 302, the number of occurrences of the extended pattern simultaneously occurring between the two edges 301 and 302 may be counted. .

Referring to the matrix 310 illustrated in FIG. 3A, the frequent pattern detection unit 120 generates a frequency "1 1 1; 1 1 1; 1 1 1" of the trunk line ('1/2') 301 and the trunk line. The number of occurrences of the ('1/3') 302 is obtained by performing an AND operation on "1 1 1; 0 1 1; 1 1 1". It is possible to count the number of occurrences (FiS number '8') of the extended patterns (1/2/3) 306 that 301 and 302 occur at the same time, and since the threshold is exceeded, the corresponding extended pattern 306 may be frequently patterned. Can be detected.

In this case, the frequent pattern detection unit 120 may detect a significant frequent pattern by using frequent patterns that are connected to each other when determining whether an extended pattern in which several patterns occur simultaneously is frequent in an input graph stream.

In other words, the frequent pattern detection unit 120 may detect a frequent pattern of two trunks using two frequent patterns of one trunk connected to the same vertex, and similarly, two trunks including the same trunk By using the two frequent patterns, the frequent pattern including three trunk lines can be detected.

For example, the frequent pattern detection unit 120 may select the trunk line among the three single trunk lines ('1/2', '1/3', and '3/4') 301, 302, and 305 detected as the frequent patterns. In the case of the '1/2') 301 and the trunk line ('3/4') 305, since the overlapping vertices are not connected, the extension pattern in which the two frequent patterns 301 and 305 occur at the same time is determined from Can be excluded.

The output unit 140 outputs the list of the frequent patterns to the terminal that has requested the detection.

That is, the output unit 140 transmits the detected frequent patterns to the terminal, and simultaneously transmits the frequent patterns and the information about the frequent patterns (eg, the number of occurrences) to the frequent pattern management table 160 through the table recording unit 130. Can be saved for use in the next sliding window.

For example, referring to FIG. 3B, the table recording unit 130 may generate the frequency of occurrence of each of the single edge patterns 301 to 305 and the expansion patterns 306 to 308 detected from the input graph stream. The number of FiBs and the number of FiSs per batch in the management tables 160 and 320 may be recorded, respectively.

The output unit 140 may include the patterns 301, 302, 305 to 308 having the number of FiSs greater than or equal to the threshold '5' in the frequent pattern management tables 160 and 320 in the list of the frequent patterns and transmit them to the terminal. .

When the detection of the frequent pattern is completed at any point of the graph stream as described above, the preprocessor 110 deletes the graph data stored in the memory unit 150 to secure a memory space, and the table recording unit 130 A frequent pattern management table 160 that records information about the detected frequent patterns may be provided to use information on previously detected frequent patterns when detecting frequent patterns in a next sliding window.

In detail, when a frequent pattern is detected at any point of the graph stream, the table recording unit 130 consecutively detects whether the frequent pattern is detected again at the next point of the graph stream to which the sliding window is to be moved, or successively. The frequent pattern prediction information of at least one moving point number of the sliding window to be detected is analyzed, and a frequent pattern management table 160 for recording the frequent pattern prediction information and the occurrence frequency for each pattern is provided.

For example, referring to FIG. 3B and FIG. 4, the table recording unit 130 successively detects frequent patterns 301, 302, 304, 305 to 308 detected through the sliding window using the algorithm shown in FIG. 4. The number of frequent sliding windows 'slideNum' may be calculated and recorded in the 'slideNum' item in the frequent pattern management table 320 of FIG. 3B.

According to the algorithm of FIG. 4, the table recorder 130 generates a frequency pattern 'FiS' greater than the threshold 'th' in the frequent patterns 301, 302, 304, and 305 to 308 within one sliding window. The number of sliding windows in which frequent patterns 301, 302, 304, and 305 to 308 are successively frequent may be calculated by dividing into small cases.

First, when the number of occurrences 'FiS' in the sliding window is greater than the threshold 'th', the table recording unit 130 may include graph data a, which is included in the first batch (batch 1) in the current sliding window. When b and c) are deleted, the number of occurrences of the graph data (d to i) included in the remaining batch 2 and the batch 3 is calculated and entered into the remaining batch count ('BatchCount'), and the newly entered batch is next. Assuming that the number of occurrences ('FiB [slideNum + 1]') in the graph data included in 4 are all '0', the value obtained by subtracting 'FiB [slideNum + 1]' from 'BatchCount' is smaller than the threshold value. You can return the number of iterations until you lose as 'slideNum'.

Next, when the occurrence frequency 'FiS' in the sliding window is smaller than the threshold 'th', the table recording unit 130 generates the occurrence frequency 'FiB [in the graph data included in the newly input batch 4. assuming slideNum + 1] ') is all' 1 ', the number of times iterations until' BatchCount 'is subtracted from' FiB [slideNum + 1] 'is smaller than the threshold can be returned as' slideNum'. have.

The table recorder 130 analyzes the frequent patterns 305, 307, and 308 in which 'slideNum' is returned as '0' as not frequent in the next sliding window, and the value where 'slideNum' is not '0' (eg, For example, the frequent patterns 301, 302, 304, and 306 returned as -1 ',' 1 ', etc. are analyzed as frequent patterns in the next sliding window, and the returned' slideNum 'is used as a frequent pattern management table (320). ) Can be recorded.

In this case, the table recording unit 130 may analyze the absolute value of the returned 'slideNum' as the number of sliding windows in which the frequent pattern is frequent. That is, the frequent pattern having an absolute value of 'slideNum' of 1 may mean that the pattern is frequent in one sliding window.

In other words, 'SlideNum' may represent a value calculated by calculating whether each pattern is frequent during the next sliding window or how many times in the next sliding window.

For example, in the frequent pattern management table 160 illustrated in FIG. 3B, the edges 1/2 and 301 in which 'slideNum' is recorded as '1' may be predicted as frequent patterns up to the next sliding window. .

To verify this, referring to the matrix for the next sliding window (batch 2, 3, 4) in FIG. 5B, even if the number of occurrences of the trunk edge 1/2 has been entered as '0' in batch 4, the Since the number of occurrences in the previous batch 2 and batch 3 has already exceeded the threshold at '6', the trunk line 1/2 may be a frequent pattern that also occurs frequently in the next sliding window. However, if the number of occurrences of the edge (1/2) in the next sliding window (batch 3, 4, 5) is all entered as '0' in batch 5, then the number of occurrences in batch 3 and batch 4 before that '3', so it may no longer be a frequent pattern. Therefore, in this case, 'slideNum' of the trunk line 1/2 may be recorded as '1'.

When the new graph data is input, the frequent pattern detection unit 120 moves the sliding window to a point of the graph stream having the new graph data as a configuration, and among the patterns identified in the moved sliding window, the frequent pattern detection unit 120 includes: The first frequent pattern in which the frequent pattern prediction information is recorded in the management table 160 is detected, and among the patterns identified in the moved sliding window, the occurrence frequency counted for the pattern in which the frequent pattern prediction information is not recorded. The second frequent pattern may be additionally detected based on the second frequent pattern, and the first and second frequent patterns may be included in the list.

For example, referring to FIGS. 5A to 5C, when the graph data j, k and l included in one batch are newly input as shown in FIG. 5A, the preprocessor 110 may generate new graph data. The matrix 520 corresponding to the next sliding window (batch 2, 3, 4) may be configured as (4,510) with (j, k, l) as shown in FIG. 5B.

The frequent pattern detection unit 120 moves to the next sliding window including the batches 2, 3, and 4 including the newly inputted batch 4 instead of the batch 1 inputted first, and when detecting the frequent pattern from the next sliding window, In the frequent pattern management tables 160 and 530 illustrated in FIG. 5C, patterns 501, 502, 504 and 506 in which 'slideNum' is not '0' are detected as the first frequent pattern without any arithmetic processing, The number of occurrences may be counted only for the remaining patterns 503, 505, 507, and 508 in which 'slideNum' is '0', and the patterns 503 and 505 in which the FiS count is greater than or equal to the threshold may be additionally detected as the second frequent pattern.

That is, the frequent pattern detection unit 120 adds the first frequent patterns 501, 502, 504, 506 detected in the previous sliding windows (batch 1, 2, 3) and the current sliding windows (batch 2, 3, 4). The detected second frequent patterns 503 and 505 can be included in the list of frequent patterns detected by the sliding windows (batch 2, 3, 4) this time.

As such, the frequent pattern detection unit 120 may perform only necessary calculation by using the frequent pattern prediction information 'slideNum' related to the frequent patterns detected in the previous sliding window, thereby reducing the total amount of computation.

When the detection of the frequent patterns is completed, the table recording unit 130 corrects the frequent pattern prediction information 'slideNum' recorded with respect to the first frequent pattern in the frequent pattern management table 160 and applies the second frequent pattern to the second frequent pattern. The frequent pattern prediction information and the number of occurrences that have been counted can be additionally recorded.

For example, referring to the frequent pattern management table 530 illustrated in FIG. 5C, the table recording unit 130 may define 'slideNum' of the patterns 501, 502, 504, and 506 detected as the first frequent patterns. From 1 'or' -1 'to' 0 ', the number of occurrences corresponding to batch 4 (' FiB batch 4 ') and the current sliding window for the patterns 503 and 505 detected as the second frequent patterns The number of occurrences of 'FiS' can be recorded.

Accordingly, since the table recording unit 130 records 'slideNum' of all patterns as '0', the frequent pattern detection unit 120 detects the frequent patterns in the subsequent sliding windows (batch 3, 4, and 5). Without using the frequent pattern management table 530, the occurrence frequency may be counted for all patterns as in the initial stage to check whether there are frequent occurrences.

As described above, according to the present invention, in order to solve the performance degradation due to the overlap calculation, which is a problem of the sliding window method used in the existing gradual frequent pattern detection technique, frequent pattern management storing information on the frequently detected frequent patterns The new frequency pattern may be detected using the table 160.

According to the present invention, the frequent pattern management table 160 calculates and stores how many sliding windows the frequent detected patterns are in the future, and performs only the necessary calculation in the next sliding window through this value to calculate the total amount of computation. Can be reduced.

2 is a diagram illustrating an example of a graph stream continuously input in the frequent pattern detection system in the graph stream according to an embodiment of the present invention.

In general, since the graph stream is composed of a plurality of graph data continuously input, the frequent pattern detection system of the present invention can detect a frequent pattern by dividing a wide and continuous graph stream into sliding window units.

In other words, the frequent pattern detection system of the present invention may detect a frequent pattern by detecting a frequent pattern in a graph stream corresponding to one sliding window and then moving to the next sliding window.

2 shows a graph stream 200 corresponding to one sliding window. In the present specification, one sliding window may be configured to include a predetermined number (eg, '9') of graph data by the user terminal.

As illustrated in FIG. 2, the graph stream 200 corresponding to one sliding window is composed of nine graph data a to i and may be divided into three batches. Three graph data may be included in one batch, and three batches may form one sliding window.

In the present specification, the sliding window may move in one arrangement unit. Therefore, the frequent pattern detection system of the present invention can move the sliding window by one batch when three graph data constituting one batch are newly input.

3A illustrates an example of a matrix configured according to the presence or absence of edges in a frequent pattern detection system in a graph stream according to an embodiment of the present invention.

Referring to FIG. 3A, when a graph stream (see 200 of FIG. 2) corresponding to one sliding window is input, the frequent pattern detection system stores the graph stream 200 in a memory unit in the order of input. A matrix 310 representing the presence or absence of edges connecting two vertices in each of the nine graph data a to i included in the graph stream 200 may be configured as shown in FIG. 3A.

In the frequent pattern detection system of the present invention, the edge connecting the vertex '1' and the vertex '2' is represented by '1/2' 301, and the edge ('1 / 2 ') 301 to assign a value of' 1 'to graph data without edges (' 1/2 ') 301 and to assign' 0 'to graph data without edges (' 1/2 ') 301. Can be.

Similarly, the frequent pattern detection system of the present invention gives a two-dimensional array value ('1' or '0') according to the presence or absence of the trunk line to the respective graph data (a to i) for the other trunks (302 to 305), The matrix 310 shown in FIG. 3A can be configured.

The frequent pattern detection system of the present invention can count the edges to which the value '1' is assigned in the matrix 310 shown in FIG. 3A as the occurrence of the pattern.

That is, the frequent pattern detection system of the present invention configures the matrix 310 to identify all occurrences of a pattern composed of one edge in the graph stream 200 corresponding to one sliding window.

3B illustrates an example of an initial frequent pattern management table in a frequent pattern detection system in a graph stream according to an exemplary embodiment of the present invention.

Referring to FIG. 3B, the frequent pattern detection system of the present invention may provide an initial frequent pattern management table 320 that records information on the detected frequent patterns when the frequent pattern is detected in one sliding window.

That is, the frequent pattern detection system of the present invention may record information on at least one of the number of occurrences and frequent occurrences of the pattern identified in the sliding window unit and the number of frequent sliding windows thereafter in the frequent pattern management table 320. have.

To this end, the frequent pattern detection system of the present invention counts the number of occurrences (FiB times) in each of three arrangements for each of the trunk lines 301 to 305 in the matrix 310 shown in FIG. In addition, the sums may be counted and the number of occurrences (FiS) of one sliding window may be counted and recorded in the 'FiS' item.

For example, in the frequent pattern detection system of the present invention, the edge ('1/2') 301 has a number of FiBs ('3') generated in batch 1 (graph data a, b, c), and batch 2 ( The number of FiBs ('3') generated in the graph data d, e and f and the number of FiBs generated in the batch 3 (graph data g, h and i) ('3') are counted and summed up, respectively. As shown in FIG. 3B, the number of FiS ('9') can be counted, and all counted occurrences can be recorded in the frequent pattern management table 320. Similarly, the number of FiBs and the number of FiSs for the other single edges 302 to 305 can be recorded. Can be counted and recorded in the frequent pattern management table 320.

Here, the frequent pattern detection system of the present invention can detect the patterns 301, 302, and 305 recorded above the set threshold '5' of the number of FiS as the frequent patterns. That is, the frequent pattern detection system of the present invention can detect three frequent patterns of a single trunk line from the graph stream 200 of FIG. 2.

In addition, the frequent pattern detection system of the present invention, among the three single edges ('1/2', '1/3', '3/4') (301, 302, 305) detected in the frequent pattern, vertex (' 1 ") an AND operation of the number of occurrences of the trunk ('1/2') 301 and the trunk ('1/3') 302 that overlap each other, so that the two trunks 301 and 302 simultaneously occur. The number of occurrences of the pattern ('1/2/3') 306 (FiS number '8') is counted, and similarly to the edge ('1/3') 302 having overlapping vertices ('3'). The number of occurrences of the extended pattern ('1/3/4') 307 where the two edges 302 and 305 occur simultaneously by ANDing the occurrences of the edges ('3/4') 305 (FiS times) '6') can be counted and recorded in the frequent pattern management table 320.

In addition, in the frequent pattern detection system of the present invention, since the two expansion patterns 306 and 307 are connected to each other by overlapping trunk lines ('1/3') 302, the two expansion patterns 306 and 307 are simultaneously generated. The number of occurrences (FiS number '6') of the extended pattern ('1/2/3/4') 308 can be counted by the AND operation, and recorded in the frequent pattern management table 320.

Here, the frequent pattern detection system of the present invention can detect each expansion pattern 305, 306, 307 whose occurrence frequency exceeds a threshold as a frequent pattern. That is, the frequent pattern detection system of the present invention detects three frequent patterns of a single trunk, two frequent patterns of two trunks, and one frequent pattern of three trunks from the graph stream 200 of FIG. can do.

That is, the frequent pattern detection system of the present invention detects the patterns 301, 302, 305 to 308 in which the number of FiSs is greater than or equal to the threshold '5' in the frequent pattern management table 320 in one sliding window of FIG. 2. Included in the list of patterns can be delivered to the terminal.

In addition, the frequent pattern detection system of the present invention includes the number of sliding windows in which frequent patterns 301, 302, 304, 305 to 308 detected frequently in the graph stream 200 corresponding to the sliding window shown in FIG. 'slideNum') may be calculated using the algorithm shown in FIG. 4 and recorded in the 'slideNum' item in the frequent pattern management table 320 of FIG. 3B.

Here, the absolute value of 'slideNum' may mean the number of sliding windows in which frequent patterns 301, 302, 304, and 305 to 308 detected in the sliding window shown in FIG. 2 frequently occur in subsequent sliding windows. have.

For example, a pattern with 'slideNum' of '0' means a pattern that does not occur frequently in the next sliding window. A pattern with 'slideNum' other than '0' (for example, '-1' or '1') It may mean that the pattern is also frequent in the next sliding window.

Therefore, when the frequent pattern detection system of the present invention detects the frequent pattern in the next sliding window, the frequency of occurrence is only counted for the pattern in which the 'slideNum' is '0' in the frequent pattern management table 320 to check whether there is a frequent occurrence. By reducing the amount of computation and eliminating redundant calculations, the performance for frequent detection can be improved.

FIG. 4 is a diagram illustrating an algorithm for calculating a number of sliding windows in which a frequent pattern will subsequently occur frequently in a frequent pattern detection system in a graph stream according to an embodiment of the present invention.

Referring to FIG. 4, in the frequent pattern detection system of the present invention, for each of the patterns 301, 302, 304, and 305 to 308 detected as the frequent patterns in FIG. 3B, the number of occurrences in one sliding window (' FiS ') is divided into a case where the threshold value' th 'is greater than and smaller than the threshold value' th ', thereby calculating the number of sliding windows (' slideNum ') in which each pattern 301, 302, 304, 305 to 308 frequently occurs in succession. .

First, in the frequent pattern detection system of the present invention, when the occurrence frequency 'FiS' in the sliding window is greater than the threshold 'th', the graph data included in the first batch (batch 1) in the current sliding window ( When a, b, and c) are deleted, the number of occurrences of the graph data (d to i) included in the remaining batch 2 and batch 3 is calculated and entered into the remaining batch count ('BatchCount'), and then newly input. Assuming that the number of occurrences ('FiB [slideNum + 1]') in the graph data included in the batch 4 is '0', the value obtained by subtracting 'FiB [slideNum + 1]' from 'BatchCount' is the threshold value. The number of iterations until it gets smaller can be returned as 'slideNum'.

Next, when the frequency of occurrence 'FiS' in the sliding window is smaller than the threshold 'th', the frequency pattern detection system of the present invention generates the frequency of occurrence (') in the graph data included in the newly input batch 4. Assume that FiB [slideNum + 1] ') is all' 1 ', and returns as' slideNum' the number of iterations until 'BatchCount' is subtracted from 'FiB [slideNum + 1]' until it becomes smaller than the threshold. can do.

The frequent pattern detection system of the present invention analyzes the frequent patterns 305, 307, and 308 in which 'slideNum' is returned as '0' as not frequent in the next sliding window, and has a value in which 'slideNum' is not '0'. The frequent patterns 301, 302, 304, and 306 returned as (eg, -1 ',' 1 ', etc.) are analyzed as frequent patterns in the next sliding window, and the returned' slideNum 'is frequently found in FIG. 3B. It can be recorded in the pattern management table 320.

Here, 'SlideNum' may represent a value calculated by calculating whether each pattern is frequent during the next sliding window or how many times the next sliding window is frequent. For example, in the frequent pattern management table 320 illustrated in FIG. 3B, the trunk lines 1/2 301 in which 'slideNum' is recorded as '1' may be predicted as frequent patterns up to the next sliding window. .

5A to 5C are diagrams for describing a process of detecting a frequent pattern using a frequent pattern management table after moving a sliding window in a frequent pattern detection system in a graph stream according to an embodiment of the present invention.

5A shows an example of a newly input graph stream, FIG. 5B shows a matrix configured after the sliding window moves, and FIG. 5C shows a frequent pattern modified according to the frequent pattern detection after the sliding window moves. The management table is shown.

Referring to FIGS. 5A to 5C, the frequent pattern detection system of the present invention is illustrated in FIG. 5B when three graph data j, k, l forming a batch are newly input as shown in FIG. 5A. New graph data j, k, 1 may be further added as batch 4 510 to the matrix 520 corresponding to the next sliding window (batch 2, 3, 4).

For example, in the frequent pattern detection system of the present invention, since all edges ('1/2', '1/4', '2/4') are all present in the newly input graph data (j, k, l), " 1 " 1 1 ", and since there is no edge ('1/3'), it is assigned as" 0 0 0 ". Because edge ('3/4') is only in the graph data (k), it is assigned as" 0 1 0 ". The two-dimensional array of arrangement 4 510 may be added to the matrix 520.

The frequent pattern detection system of the present invention moves the sliding window to include three newly input graph data (j, k, l) as the batch 4, and detects the frequent pattern from the moved sliding window (next sliding window). In the frequent pattern management table 530 illustrated in FIG. 5C, patterns 501, 502, 504, and 506 in which 'slideNum' is not '0' are detected as a frequent pattern without any arithmetic processing, and 'slideNum' Can be modified from '1' or '-1' to '0'.

That is, if the 'slideNum' is '1' or '-1', since the frequent pattern detection system means a frequent pattern during one sliding window, the frequency of occurrence for checking whether the corresponding pattern 501, 502, 504, 506 is frequent The count or AND operation can be omitted.

In addition, the frequent pattern detection system of the present invention means that if the slideNum is '0', it means that the pattern is not frequent in the subsequent sliding window, and thus the patterns 503, 505, 507, and 508 in which the 'slideNum' is '0' are used. Only the occurrence count can be checked for frequent occurrences.

In the frequent pattern detection system of the present invention, when a new graph is input as shown in FIG. 5A, batch 4 510 is added to the DSMatrix 520 illustrated in FIG. 5B, and FiB and FiS are calculated for basic frequent pattern detection. It may be input to the frequent pattern management table 530 of FIG. 5C.

However, in the frequent pattern detection system of the present invention, if 'slideNum' is not '0', after reducing 'slideNum' to '0', calculation of FiB and FiS may be omitted. That is, in the frequent pattern detection system of the present invention, calculation of FiB and FiS may be omitted for the trunk lines 501, 502, and 504 in the frequent pattern management table 530 of FIG. 5C.

In addition, the frequent pattern detection system of the present invention uses a method similar to the above process when detecting the frequent pattern of the extended pattern composed of several edges, and the edge ('1 /) in the frequent pattern management table 530 of FIG. 5C. Since 'slideNum' of '2/3') 506 is '1', even if FiB and FiS are not calculated, the frequent pattern can be detected, and after changing 'slideNum' to '0', the frequent pattern can be maintained.

On the other hand, in the frequent pattern detection system of the present invention, since 'slideNum' is' 0 'in the case of the trunk line (' 1/3/4 ') 507, which is an extension pattern, the trunk line (' 1) is only applied to the newly inputted batch 4. FiB and FiS may be calculated by performing an AND operation on two single edges ('1/3', '3/4') 502 and 505 constituting (3/4 ') (507).

For example, in the frequent pattern detection system of the present invention, the arrangement 4 portion of the trunk line ('1/3') 502 in FIG. 5B is "0 0 0" and the arrangement of the trunk line ('3/4') 505. Since the 4 part is "0 1 0", count the FiB of batch 4 as' 0 'according to "0 0 0 & 0 1 0 = 0 0 0" and set the number of occurrences of' FiS 'in the next sliding window to' 3 'and' FiS 'is less than the threshold, so the trunk line (' 1/3/4 ') 507 may determine that it is not a frequent pattern. At this time, the frequent pattern detection system of the present invention may detect the edge ('1/3/4') 507 as the frequent pattern and calculate 'slideNum' according to the algorithm shown in FIG. 4 when the 'FiS' is greater than or equal to the threshold. .

Hereinafter, the workflow of the frequent pattern detection system 100 according to the embodiments of the present invention will be described in detail.

6 is a flowchart illustrating a method of detecting a frequent pattern in a graph stream according to an embodiment of the present invention.

The frequent pattern detection method in the graph stream according to the present embodiment may be performed by the frequent pattern detection system 100 described above.

Referring to FIG. 6, in step 610, the frequent pattern detection system 100 checks whether frequent pattern detection for a graph stream is requested from a terminal (for example, a manager terminal).

When the frequent pattern detection is requested by the terminal, in step 620, the frequent pattern detection system 100 preprocesses the graph stream including the continuously input graph data.

The frequent pattern detection system 100 generates, for example, a two-dimensional array structure called 'DSMatrix' so that frequent patterns can be detected in consideration of not only current graph data but also front and rear graph data among consecutively input graph streams. You can save the entered graph data.

In this case, 'DSMatrix' is a boolean type array, so more data can be stored in less space than the existing DSTree, and the presence or absence of edges in the graph data is expressed as '1' or '0', so the graph data can be added or deleted quickly. It is possible to fit the sliding window method.

For example, referring to FIG. 3A, the frequent pattern detection system 100 selects '1' if there are edges 301 to 305 in the graph data a to i included in the graph stream corresponding to one sliding window. If not, the graph 310 may be configured to store '0' and store the graph data.

In step 630, the frequent pattern detection system 100 counts the number of occurrences for each of the patterns identified in the sliding window unit in the preprocessed graph stream, and based on the occurrence frequency, in the graph stream of the patterns Frequent frequent patterns are detected while moving the sliding window.

That is, the frequent pattern detection system 100 counts the number of occurrences ('FiS') for each pattern in the sliding window moving on the graph stream, and detects a pattern having the occurrence frequency more than a threshold as a frequent pattern, and detects the frequent pattern. Calculate the information ('slideNum') in the frequent pattern management table, move the sliding window, and then perform only necessary operations using the information ('slideNum') recorded in the frequent pattern management table to perform frequent patterns. It can be detected quickly.

In addition, the frequent pattern detection system 100 of the present invention may determine whether the expansion pattern composed of two or more trunks is frequent by performing an AND operation on the number of occurrences of two patterns consisting of a single trunk.

For example, the frequent pattern detection system 100 of the present invention includes three single edges ('1/2', '1/3', '3/4') 301, 302, and 305 detected as frequent patterns. Of the edges ('1/2') 301 and the number of occurrences of the edges ('1/3') 302 by ANDing the number of occurrences of the extended pattern at which the two edges 301 and 302 simultaneously occur. You can count.

At this time, the frequent pattern detection system 100 of the present invention can detect a significant frequent pattern from the graph stream by ANDing the number of occurrences by selecting a plurality of patterns having the same vertex and neighboring each other in consideration of the connectivity of the two patterns. Can be.

In step 650, the frequent pattern detection system 100 outputs a list of the frequent patterns to the terminal requesting the detection.

For example, the frequent pattern detection system 100 may output a list regarding the frequent patterns detected for each sliding window to the terminal.

For example, the frequent pattern detection system 100 includes three frequent patterns of single edges detected from the graph stream 200 corresponding to the sliding window shown in FIG. 2, two frequent patterns of two trunks, and three. A list including one frequent pattern of three trunks may be delivered to the terminal.

As described above, according to the present invention, it is possible to reduce the overall calculation amount by using prediction information calculated whether the pattern detected in the previous sliding window is frequent or not frequent, and detect a significant frequent pattern in consideration of the connectivity between the patterns. .

The method according to an embodiment of the present invention can be implemented in the form of program instructions that can be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the media may be those specially designed and constructed for the purposes of the embodiments, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer readable recording media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine code, such as produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described by the limited embodiments and the drawings as described above, various modifications and variations are possible to those skilled in the art from the above description. For example, the described techniques may be performed in a different order than the described method, and / or components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different form than the described method, or other components. Or even if replaced or replaced by equivalents, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are within the scope of the claims that follow.

100: frequent pattern detection system
110: preprocessing unit 120: frequent pattern detection unit
130: table recording unit 140: output unit
150: memory unit 160: frequent pattern management table

Claims (12)

As detection for frequent patterns is requested,
Preprocessing a graph stream comprising continuously input graph data;
Counting the number of occurrences of each pattern identified in units of sliding windows in the preprocessed graph stream;
Detecting, based on the number of occurrences, a frequent pattern occurring in the graph stream among the patterns while moving the sliding window;
If the frequent pattern is detected at any point in the graph stream,
Analyzing the frequent pattern prediction information of at least one of whether the frequent pattern is detected again at a next point of the graph stream to which the sliding window is to be moved, or the number of moving points of the sliding window which subsequently detect the frequent pattern; ;
Providing a frequent pattern management table recording the frequent pattern prediction information and the occurrence frequency for each pattern; And
Outputting a list of the frequent patterns to a terminal that has requested the detection;
Including,
The detecting step,
If new graph data is input, moving the sliding window to a point in the graph stream having the new graph data as a configuration;
Among the patterns identified in the moved sliding window, a first frequent pattern in which the frequent pattern prediction information is recorded in the frequent pattern management table is detected, and among the patterns identified in the moved sliding window, the frequent pattern prediction information is Further detecting a second frequent pattern based on the number of occurrences counted for the unrecorded pattern; And
Including the first and second frequent patterns in the list
Frequency pattern detection method in a graph stream comprising a. The method of claim 1,
When the detection of the frequent pattern is completed among the pattern consisting of a single edge in the graph stream,
The detecting step,
Selecting a plurality of patterns having the same vertex and neighboring each other;
Counting the number of occurrences of the extended pattern in which each of the plurality of patterns occurs simultaneously by performing an AND operation on the number of occurrences of each of the plurality of patterns; And
Detecting an extended pattern whose occurrence number is counted above a threshold value as a frequent pattern and including the same in the list;
Frequent pattern detection method further comprising. delete delete The method of claim 1,
When the number of graph data defined in the sliding window is stored in the memory unit storing the graph stream in the order of input,
The pretreatment step,
Constructing a matrix to which a two-dimensional array value of '1' or '0' is assigned according to the presence or absence of edges connecting two vertices in each of the graph data included in the sliding window; And
Identifying, by the pattern, a subgraph including the edges given by the two-dimensional array value as '1' and the two vertices leading to the edges;
Frequent pattern detection method comprising a. The method of claim 5,
Counting the number of occurrences,
Counting the number of frequency in sliding windows (FIS) in which the pattern is generated in each of graph data included in the sliding window by adding the two-dimensional array values for each of the patterns.
Frequent pattern detection method comprising a. The method of claim 6,
When the sliding window is composed of a plurality of batches,
Counting the number of occurrences,
Counting the number of frequency in batch (FiB) in which the pattern is generated in one batch by adding the two-dimensional array values to each of the patterns in each batch; And
Counting the number of FiSs by further adding up the number of FiBs counted in each of the plurality of batches.
Frequent pattern detection method further comprising. The method of claim 6,
The detecting step,
Detecting as a frequent pattern a pattern having the FiS number greater than or equal to a threshold and including the pattern in the list;
Frequent pattern detection method further comprising. As detection for frequent patterns is requested,
A preprocessing unit for preprocessing a graph stream including continuously input graph data;
Counting the number of occurrences for each of the patterns identified in the sliding window unit in the pre-processed graph stream, and detecting, based on the occurrence frequency, the frequent patterns occurring in the graph stream among the patterns while moving the sliding window. A frequent pattern detection unit;
If the frequent pattern is detected at any point in the graph stream,
Analyzing the frequent pattern prediction information of at least one of whether the frequent pattern is detected again at a next point of the graph stream to which the sliding window is to be moved, or the number of moving points of the sliding window which subsequently detect the frequent pattern; A table recording unit for preparing a frequent pattern management table that records the frequent pattern prediction information and the number of occurrences for each pattern; And
An output unit for outputting a list of the frequent patterns to a terminal requesting the detection
Including,
The frequent pattern detection unit,
When new graph data is input, the sliding window is moved to a point of a graph stream having the new graph data as a configuration,
A first frequent pattern in which the frequent pattern prediction information is recorded in the frequent pattern management table is detected among the patterns identified in the moved sliding window,
Among the patterns identified in the moved sliding window, a second frequent pattern is further detected based on the number of occurrences counted for the pattern for which the frequent pattern prediction information is not recorded.
Including the first and second frequent patterns in the list.
Frequency Pattern Detection System in Graph Stream. The method of claim 9,
When the detection of the frequent pattern is completed among the pattern consisting of a single edge in the graph stream,
The frequent pattern detection unit,
Selecting a plurality of patterns having the same vertices and neighboring each other,
ANDing the number of occurrences for each of the plurality of patterns, counting the number of occurrences for the extended pattern in which each of the plurality of patterns occurs simultaneously,
An extended pattern whose occurrence count is counted above a threshold is detected as a frequent pattern and included in the list.
Frequent pattern detection system. delete delete

Images