KR20140120711A - Method and apparatus for mining closed frequent pattern using pararell processing - Google Patents

Abstract

The present invention relates to a method and an apparatus to mine a closed frequent pattern using parallel processing. According to an aspect of the present invention, a method to mine a closed frequent pattern comprises the steps of: generating and storing projection data using a frequent item as a prefix pattern in a database; allocating the stored projection data to a plurality of processing nodes in accordance to the priority of the stored projection data; generating a frequent pattern using the allocated projection data; comparing the frequent pattern with a closed frequent pattern list to update the closed frequent pattern list; and generating and storing projection data using the frequent pattern as a prefix pattern when the frequent pattern is added to the closed frequent pattern list.

Description

[0001] METHOD AND APPARATUS FOR MINING CLOSED FREQUENTIAL PATTERN USING PARALLEL PROCESSING [0002]

To a closed frequent pattern mining method and apparatus using parallel processing.

The amount of data that accumulates with the development of IT technology, including the Internet, is increasing exponentially. It is becoming increasingly difficult to analyze this huge data and extract useful information. As an effort to solve this problem, data mining techniques have been researched and developed.

Among the data mining techniques, association rules mining is a technique to find out how the items in the data are related. To do this, we first find a frequent pattern and find the association rules from it. However, as the amount of data increases, the number of frequent patterns to be extracted increases, and in recent years, the number of frequent patterns is too large to analyze. The concept of expressing a large number of frequent patterns called closed frequent patterns as a pattern has been proposed.

In order to extract a closed frequent pattern, it was found that all the frequent patterns were found and the closed frequent patterns were found in the frequent patterns found. However, it took too long to extract the frequent patterns and many frequent patterns eventually resulted in one closed frequent patterns Which is inefficient. Thus, techniques have been developed to extract frequent closed patterns directly from the data.

Basically, algorithms that find closed frequent patterns have exponential time complexity. All possible patterns are searched by a depth-first search method or a width-first search method to check whether a desired closed closed pattern is desired. In this case, the patterns to be inspected can be structured in the form of a tree, and the closed patterns may not be a closed pattern in which the patterns belonging to a subtree rooted at a specific pattern are closed. The performance of algorithms depends heavily on how quickly it can be found without a direct search.

It is an object of the present invention to provide a closed frequent pattern mining method and apparatus using parallel processing.

A closed frequent pattern mining method using parallel processing according to an aspect includes generating and storing projection data that prefixes a frequent item in a database and storing the projection data in a plurality of processing nodes according to a priority among the stored projection data, Generating a frequent pattern using the allocated projection data, updating the closed frequent pattern list by comparing the frequent pattern with a closed frequent pattern list, and updating the closed frequent pattern list using the closed frequent pattern list A step of generating and storing projection data having the frequent pattern as a prefix pattern may be included.

According to an aspect, the allocating step may allocate the stored projection data according to a depth-first search order based on a prefix pattern of the stored projection data.

According to an aspect of the present invention, the step of generating and storing projection data having a frequent item as a prefix pattern in the database includes a step of searching a frequent item in the database, a step of generating projection data in which the frequent item is a prefix pattern, And storing the projected projection data.

According to an aspect of the present invention, the step of generating the frequent pattern may include a step of finding a frequent item in the allocated projection data, and a step of generating a frequent pattern by combining the preliminary pattern of the allocated projected data and a frequent item of the allocated projected data Step < / RTI >

According to an aspect, the updating of the closed frequent pattern list may update the closed frequent pattern list according to whether a closed frequent pattern candidate having the same degree of support as the frequent pattern exists in the closed frequent pattern list.

According to an aspect, the updating of the closed frequent pattern list may include adding the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate having the same degree of support as the frequent pattern in the closed frequent pattern list can do.

According to an aspect of the present invention, the step of updating the closed frequent pattern list may further include the step of, when there is no closed frequent pattern candidate that is a subset of the frequent patterns among the closed pattern candidates having the same degree of support as the frequent pattern, You can add it to a closed frequent pattern list.

According to an aspect of the present invention, in the step of updating the closed frequent pattern list, when there is no closed frequent pattern candidate having the same support record as the support pattern of the frequent pattern among the closed pattern candidates having the same support degree as the frequent pattern, A frequent pattern can be added to the closed frequent pattern list.

According to an aspect of the present invention, the step of updating the closed frequent pattern list may further include the step of, when the support degree of the frequent pattern is less than a predetermined value, storing the same support record as the support record of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern A closed frequent pattern candidate having the frequent pattern candidate as a subset of the closed pattern candidates having the same degree of support as the frequent pattern is determined if the closed frequent pattern candidate having the same degree of support as the frequent pattern is present, It can be determined whether or not it exists.

According to an aspect of the present invention, generating and storing projection data having the frequent pattern as a prefix pattern may generate only one projection data from the allocated projection data when the number of stored projection data is equal to or larger than a preset value, The received projection data may be stored with the next item to be projected.

Closed frequent pattern mining apparatus using parallel processing according to one aspect includes projection data providing unit for storing projection data generated at a plurality of processing nodes and allocating stored projection data to a plurality of processing nodes in accordance with priority, Generating a frequent pattern using the closed frequent pattern management unit storing the closed frequent pattern list including the candidates and the projection data allocated from the projection data providing unit to update the closed frequent pattern list, When added to the pattern list, it may include a plurality of processing nodes that generate projection data that makes the frequent pattern a pre-pattern.

According to an aspect, the projection data providing unit may allocate the stored projection data according to a depth-first search order based on a prefix pattern of the stored projection data.

According to an aspect, any one of the plurality of processing nodes may generate frequent items in the database, and then generate projection data with the frequent items as a prefix pattern.

According to an aspect, the plurality of processing nodes may search a frequent item in the allocated projection data, and then combine the found frequent items with the prefix pattern of the allocated projection data to generate a frequent pattern.

According to an aspect, the plurality of processing nodes can update the closed frequent pattern list according to whether a closed frequent pattern candidate having the same degree of support as the frequent pattern exists in the closed frequent pattern list.

According to an aspect, the plurality of processing nodes may add the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate having the same degree of support as the frequent pattern in the closed frequent pattern list.

According to one aspect, when there are no closed frequent pattern candidates that include the frequent patterns as a subset among the closed frequent patterns having the same degree of support as the frequent patterns, the plurality of processing nodes may convert the frequent patterns into the closed frequent patterns You can add it to the list.

According to an aspect of the present invention, when there are no closed frequent pattern candidates having the same support record as the support pattern of the frequent pattern among the closed frequent pattern candidates having the same degree of support as the frequent pattern, Can be added to the closed frequent pattern list.

According to an aspect of the present invention, when the degree of support of the frequent pattern is less than a predetermined value, the plurality of processing nodes determine whether a closed frequent pattern having the same support record as the support record of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern Determining whether or not a closed frequent pattern candidate having a subset of the frequent patterns among the closed pattern candidates having the same degree of support as the frequent pattern exists if the degree of support of the frequent pattern is equal to or greater than a predetermined value, It can be judged.

According to an aspect of the present invention, the plurality of processing nodes generate only one projection data from the allocated projection data when the number of projection data stored in the projection data providing unit is equal to or greater than a preset value, It can be saved together with the frequent items to be projected.

Parallelization can reduce the time taken for frequent pattern mining that is closed.

Furthermore, it is possible to reduce the search space for closed frequent pattern mining by preferentially generating frequent patterns that are likely to be closed frequent patterns.

1 is an illustration for explaining projection data,
FIG. 2 is a block diagram of a closed frequent pattern mining apparatus using parallel processing according to an embodiment. FIG.
FIGS. 3A to 3F are diagrams for explaining a parallel processing process according to a depth-first search method;
Figure 4 is an illustration of a closed frequent pattern list,
5 is a flowchart showing a process of generating projection data in a database,
6 is a flowchart illustrating a closed frequent pattern mining process by a plurality of processing nodes.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the like elements throughout.

1 is an exemplary diagram for explaining projection data.

Referring to FIG. 1, in the illustrated example, projection data 120, which preforms an exemplary database 110 and frequent item a, preforms projection data 130, which prefixes frequent item b, and frequent item c, And projection data 140 in the form of a pattern are shown.

A frequent item means an item that satisfies the minimum support defined by the user. In addition, a pattern satisfying the minimum support is called a frequent pattern, and a pattern means a set of items.

On the other hand, support means the number of records including a specific pattern, and minimum support means a minimum value of support required for a specific pattern to become a frequent pattern.

In the illustrated example, if the minimum support is 2, the frequent items in database 110 are a, b, and c.

Projection data, on the other hand, is a collection of records that have patterns in a database that are prefixed with a specific frequent item or a specific frequent pattern.

For example, the projection data 120 with the prefix pattern a in FIG. 1 includes patterns with a pattern as a prefix in the database and a record with each pattern. At this time, the prefix pattern a is displayed for all the patterns included in the projection data, so that the prefix pattern a may not be displayed separately.

Meanwhile, the database 110 may be, for example, a set of purchase items per customer. At this time, the record may be identification information of the customer, and each item may mean a product purchased by the customer.

In another example, the database 110 may represent the patient's health screening data. At this time, the record is identification information of the patient, and each item may be indicative of the health examination data. However, the database 110 is not limited to the illustrated ones.

FIG. 2 is a block diagram of a closed frequent pattern mining apparatus using parallel processing according to an embodiment.

Referring to FIG. 2, a frequent pattern mining apparatus 200 according to an embodiment may include a plurality of processing nodes 210.

According to one embodiment, each processing node 210 may be a respective core included in a multicore processor.

According to another embodiment, each processing node 210 may be a single core processor.

According to another embodiment, the processing node 210 may be a respective computing device that constitutes a distributed computing environment.

The plurality of processing nodes 210 may be assigned a job for mining a closed frequent pattern in the projection data providing unit 230 and may process the same in parallel.

More specifically, the plurality of processing nodes 210 may generate a frequent pattern by allocating the projection data from the projection data providing unit 230, and determine whether the generated frequent pattern is a closed frequent pattern .

In this case, a frequent pattern that is not included in another frequent pattern having the same degree of support among frequent patterns can be defined as a closed frequent pattern.

According to one embodiment, any one of the plurality of processing nodes 210 may generate all of the frequent items in the database and then projection data that pre-patterns each frequent item.

For example, given a database 110 such as that shown in FIG. 1, and with a minimum support of 2, the frequent items are a, b, and c and the projection data 120 To 140 may be generated.

On the other hand, generation of projection data that prefixes frequent items and frequent items in the database can be performed at any one of the processing nodes 210 of the plurality of processing nodes 210.

For example, one processing node can be selected arbitrarily considering the current operating state of each processing node or the performance of each processing node.

The projection data providing unit 230 stores the projection data generated at the processing node 210 and outputs the high-priority projection data according to the pseudo depth-first search method among the stored projection data to a plurality of processing units Node 210 in a similar manner.

Meanwhile, according to one embodiment, the projection data providing unit 230 may be implemented with a priority queue. At this time, the priority queue can be implemented based on a data structure such as a heap with a prefix pattern of projection data as a key.

Meanwhile, each of the plurality of processing nodes 210 may receive the projection data from the projection data providing unit 230 to generate a frequent pattern.

According to one embodiment, each of the plurality of processing nodes 210 finds a frequent item from the projection data provided by the projection data providing unit 230, combines the frequent item with the pre-pattern of the projection data to generate a frequent pattern .

For example, referring to FIG. 1, the frequent items in the projection data 120 for the prefix pattern a are b and c. Thus, the processing node that has been provided with the projection data 120 for the prefix pattern a may combine the prefix pattern a and the frequent item b or c to generate frequent patterns {a, b} and {a, c}.

Also, the frequent item in the projection data 130 for the prefix pattern b is c. Thus, the processing node that has received the projection data for the prefix pattern b can combine the prefix pattern b and the frequent item c to generate frequent patterns {b, c}.

Meanwhile, each of the plurality of processing nodes 210 may generate projection data with the generated frequent pattern as a prefix pattern, and store the generated projection data in the projection data providing unit 230.

Meanwhile, the processing node that has completed the generation of the frequent pattern and the projection data may receive the projection data having the highest priority among the projection data stored in the projection data providing unit 230, and repeat the same process.

The closed frequent pattern management unit 250 may store the closed frequent pattern list.

According to one embodiment, each of the plurality of processing nodes 210 may update the closed frequent pattern list by comparing the generated frequent pattern with the closed frequent pattern candidates included in the closed frequent pattern list.

Specifically, each of the plurality of processing nodes 210 determines whether there is a frequent pattern candidate having the same degree of support as the frequent pattern generated at each processing node among the closed frequent pattern candidates included on the closed frequent pattern list, The frequent pattern list can be updated.

At this time, if there is no frequent pattern candidate having the same degree of support as the frequent pattern generated on the closed frequent pattern list, the generated frequent pattern is likely to be a closed frequent pattern, so that it can be added to the closed frequent pattern list.

According to an embodiment, when there is no closed frequent pattern candidate that is a subset of the frequently generated frequent pattern candidates having the same degree of support as the generated frequent pattern, the generated frequent pattern is classified into a closed frequent pattern list . ≪ / RTI >

That is, if there is no closed frequent pattern candidate having the same degree of support as the frequent pattern generated on the closed frequent pattern list, or if there is no closed frequent pattern candidate including the generated frequent pattern as a subset even if it exists, Frequent patterns are likely to be closed frequent patterns, so they can be added to a closed frequent pattern list.

On the other hand, a closed frequent pattern candidate included in a subset of the frequent patterns generated with the same support as the frequent patterns generated from the closed frequent pattern candidates is determined as not a closed frequent pattern, and thus can be deleted from the closed frequent pattern list.

According to another embodiment of the present invention, if there is no closed frequent pattern candidate having the same support record as the frequent pattern generated from the closed frequent pattern candidates having the same degree of support as the generated frequent pattern, You can add it to the pattern list.

On the other hand, if there is a closed frequent pattern candidate that includes a frequent pattern generated from a closed frequent pattern candidate having the same support as the generated frequent pattern and a frequent pattern generated with the same supporting record, It can not be a closed frequent pattern, so it is not added to the closed frequent pattern list.

According to one embodiment, when the support degree of the created frequent pattern is less than a preset value, the plurality of processing nodes 210 may generate a support record of the frequent pattern generated among the closed pattern candidates having the same degree of support as the created frequent pattern, It can be determined whether there is a closed frequent pattern candidate with the same support record.

At this time, when the support degree of the generated frequent pattern is equal to or greater than a predetermined value, it can be determined whether there is a closed frequent pattern candidate having a subset of the generated frequent patterns among the closed pattern candidates having the same degree of support as the generated frequent patterns .

In other words, as the length of the frequent pattern is generally shorter, the support degree becomes lower. When the frequency of the frequent pattern is shorter, the support degree becomes larger. Therefore, when the frequency of the frequent pattern is less than a predetermined value, So that the amount of computation can be reduced.

According to an exemplary embodiment, when the generated frequent pattern is not added to the closed frequent pattern list, the plurality of processing nodes 210 do not generate projection data having the frequent pattern as a prefix pattern, Can be reduced.

FIGS. 3A to 3F are diagrams for explaining a parallel processing process according to a depth-first search method.

It is assumed that the same database as the database 110 shown in FIG. 1 is given in FIGS. 3A to 3F, and the minimum support is assumed to be two. It is also assumed that there are two processing nodes.

3A shows a search space using frequent patterns that can be generated in the database 110. FIG. In the illustrated example, the nodes on the tree represent frequent patterns that can be generated in the database 110, which may be the prefix pattern of the projection data.

On the other hand, according to the depth-first search method, {a, -} {a, b} -> {a, b, c} -> {a, c} -> {b } -> {b, c} -> {c}.

At this time, according to one embodiment, the priority among the frequent items can be set so that the frequent items having the lowest support have the highest priority. However, the present invention is not limited thereto, and it can be determined by various methods such as alphabetical order.

Meanwhile, in the example shown in FIG. 3B, it is assumed that the priority order of the frequent items is determined in alphabetical order.

Referring to FIG. 3B, processing node 1 may find frequent items in database 110. In the illustrated example, the frequent items are a, b, and c.

Thereafter, projection data Da, Db, and Dc can be generated with each frequent item as a prefix pattern. At this time, the generated projection data is the same as the projection data 120 to 140 shown in Fig. Meanwhile, the generated projection data may be stored in the projection data providing unit 230.

According to one embodiment, the projection data providing unit 230 may provide the projection data stored using the priority queue 300 to the processing nodes 1 and 2 according to the priority according to the depth-first search method.

Referring to FIG. 3C, the projection data Da and Db having the high priority {a} and {b} as the prefixes according to the depth-first search method among the projection data stored in the priority queue 300 are stored in the processing nodes 1 and 2 .

Referring to FIGS. 3D and 3E, the processing nodes 1 and 2 respectively assigned the projection data Da and Db can generate frequent patterns using the allocated projection data.

Specifically, referring to FIG. 1, b and c are frequent items since the support degrees of b and c in the projection data Da (120) are more than minimum support. Therefore, the patterns {a, b} and {a, c} combining the prefix pattern a and the frequent item b or c in the projection data become frequent patterns.

Then, the processing node 1 may generate projection data Dab and Dac, which are prefix patterns of the generated frequent patterns, and store the generated projection data Dab and Dac in the priority queue 300.

Similarly, the processing node 2 may generate a frequent pattern after finding the frequent item in the projection data Db. Referring to FIG. 1, in the projection data Db 130, c is a frequent item. Therefore, {b, c} is a frequent pattern.

On the other hand, the processing node 2 may generate projection data Dbc having the generated frequent pattern as a prefix pattern, and store the generated projection data Dbc in the priority queue 300.

Then, among the projection data stored in the priority queue 300, high priority Dab, Dac on the depth-first search method can be assigned to the processing nodes 1 and 2 as in the example shown in FIG. 3F.

Meanwhile, according to one embodiment, the plurality of processing nodes 210 can generate only one projection data when the number of projection data stored in the priority queue is equal to or larger than a predetermined value.

Specifically, in the example shown in FIG. 3A, when the number of preset projection data is 2, the number of projection data stored in the priority queue 300 exceeds a predetermined value. Therefore, unlike the example shown in FIGS. 3C and 3D, the processing node 1 and the processing node 2, which respectively receive Da and Da in the priority queue 300, can generate only one projection data, respectively.

3C and 3D, the processing node 1 may generate only the projection data Dab and store it in the priority queue 300. FIG. At this time, the processing node 1 may store Da and frequent item c together in the priority queue 300 instead of generating Dac, so that the processing node 1 can be provided to the processing node 1 or 2 so that the Dac can be generated.

Meanwhile, the processing nodes 1 and 2 can update the closed frequent pattern by comparing the generated frequent pattern with the closed frequent pattern candidates of the closed frequent pattern list. At this time, if the generated frequent pattern is not added to the closed frequent pattern list, projection data having the frequent pattern as a prefix pattern may not be generated.

That is, if the generated frequent pattern is not a closed frequent pattern, a frequent pattern existing in a subtree rooted by the frequent pattern can not also be a closed frequent pattern. Therefore, by not generating the projection data having the frequent pattern as the prefix pattern, it is possible to reduce the search space by pruning the frequent pattern having the frequent pattern as the root.

4 is an exemplary view of a closed frequent pattern list.

Referring to FIG. 4, the closed frequent pattern list 400 may store closed frequent pattern candidates found so far. At this time, the closed frequent pattern list 400 may include the degree of support of the closed frequent pattern candidate.

Also, according to one embodiment, the closed frequent pattern list 400 may further include a support record of closed frequent pattern candidates.

According to one embodiment, the closed frequent pattern management unit 250 may continuously update the closed frequent pattern list 400 by comparing the generated frequent patterns with the closed frequent pattern candidates.

For example, if a frequent pattern {b, c} with a support score of 2 is generated, a closed frequent pattern candidate {a, b, c} having the same degree of support as the frequent pattern generated in the closed frequent pattern list 400 is found . At this time, {b, c} is a subset of {a, b, c} and can not be added to the closed frequent pattern list.

On the other hand, if the generated frequent pattern {b, c} has a degree of support of 3, the frequent pattern {b, c} can be added to the closed frequent pattern list since there is no closed frequent pattern candidate having the same degree of support.

On the other hand, if the generated frequent pattern is a pattern {a, b, d} with a support degree of 4, the closed frequent pattern candidate {a, b} in the closed frequent pattern list is equal to {a, b, d} is a subset of a, b, d}, {a, b} can be removed from the closed frequent pattern list and {a, b, d} can be added to the closed frequent pattern list.

As another example, if a frequent pattern {b} with a score of 4 is generated, a closed frequent pattern candidate {a, b} with a score of 4 on the closed frequent pattern list can be found.

In this case, if the record supporting the frequent pattern {b} is {R1, R2, R3, R4}, the frequent pattern {b} is a frequent pattern list 400).

On the other hand, if a frequent pattern {d} with a support score of 2 is generated and the supporting record is {R5, R6}, then the frequent pattern {d} Lt; / RTI >

On the other hand, if the frequent pattern {a, b, c, d} with support 2 is generated and the supporting record is {R1, R2}, there is a closed frequent pattern candidate {a, b, c} However, since the closed frequent pattern candidates {a, b, c} are deleted and the frequent patterns {a, b, c, d} are {a, b, c} Can be added to the closed frequent pattern list.

5 is a flowchart showing a process of generating projection data in a database.

Referring to FIG. 5, any one of the processing nodes 210 finds a frequent item with a degree of support in the database higher than the minimum support (510), and generates projection data having a prefix pattern of each frequent item (530).

Thereafter, the generated projection data may be stored in the projection data providing unit 230 (550).

6 is a flowchart illustrating a closed frequent pattern mining process by a plurality of processing nodes.

Referring to FIG. 6, each of the plurality of processing nodes 210 may receive the projection data having the highest priority among the projection data stored in the projection data providing unit 230 in parallel (610).

At this time, the priority of the projection data can be determined according to the priority according to the depth-first search method based on the prefix pattern.

Meanwhile, each of the plurality of processing nodes 210 may search for a frequent item in the allocated projection data (620) and generate a frequent pattern (630). At this time, the frequent pattern can be generated by combining the detected frequent item with the prefix pattern of the projection data.

If a frequent item is generated, it can be determined whether a closed frequent pattern candidate having the same support as the frequent pattern generated in the closed frequent pattern list and including a frequent pattern as a subset exists (640).

At this time, if there is no closed frequent pattern candidate having the same support as the frequent pattern generated in the closed frequent pattern list and including a frequent pattern as a subset, the generated frequent pattern is added to the closed frequent pattern list (650) The projection data having the generated frequent pattern as a prefix pattern may be generated and stored in the projection data providing unit 230 (660).

Meanwhile, when frequent pattern generation and projection data generation for all frequent items of the assigned projection data are completed (670), the projection data having the highest priority according to the depth-first search method among the projection data stored in the projection data providing unit 230 Data may be extracted (680, 610) and the process of 620 to 680 may be repeated.

On the other hand, when projection data stored in the projection data providing unit 230 does not exist, all the procedures are terminated and the closed frequent pattern candidates included in the closed frequent pattern list are determined as closed frequent patterns.

Meanwhile, the embodiments of the present invention can be embodied as computer readable codes on a computer readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer apparatus is stored. Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like.

Also, functional programs, codes, and code segments for implementing the present invention can be easily deduced by programmers of the art to which the present invention belongs.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

210: processing node 230: projection data provider
250: closed frequent pattern management unit

Claims (20)

Generating and storing projection data in which a frequent item is a prefix pattern in a database;
Assigning the stored projection data to a plurality of processing nodes according to a priority among the stored projection data;
Generating a frequent pattern using the allocated projection data;
Updating the closed frequent pattern list by comparing the frequent pattern with a closed frequent pattern list; And
And generating and storing projection data having the frequent pattern as a prefix pattern when the frequent pattern is added to the closed frequent pattern list. The method according to claim 1,
Wherein the assigning comprises:
Wherein the stored projection data is allocated according to a priority according to a depth-first search method based on a prefix pattern of the stored projection data. The method according to claim 1,
The step of generating and storing projection data that prefixes a frequent item in the database includes:
Searching a frequent item in the database;
Generating projection data with the frequent item as a prefix pattern; And
And storing the generated projection data. ≪ Desc / Clms Page number 19 > The method according to claim 1,
The generating of the frequent pattern may include:
Finding a frequent item from the allocated projection data; And
And generating a frequent pattern by combining the prefix pattern of the allocated projection data and the frequent item of the allocated projection data. The method according to claim 1,
Wherein updating the closed frequent pattern list comprises:
And updating the closed frequent pattern list according to whether a closed frequent pattern candidate having the same degree of support as the frequent pattern exists in the closed frequent pattern list. 6. The method of claim 5,
Wherein updating the closed frequent pattern list comprises:
And adding the frequent pattern to the closed frequent pattern list if there is no closed frequent pattern candidate having the same degree of support as the frequent pattern in the closed frequent pattern list. 6. The method of claim 5,
Wherein updating the closed frequent pattern list comprises:
And a closed frequent pattern mining using parallel processing for adding the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate having a subset of the frequent patterns among the closed pattern candidates having the same degree of support as the frequent pattern Way. 6. The method of claim 5,
Wherein updating the closed frequent pattern list comprises:
Wherein when the closed frequent pattern candidate having the same support record as the support pattern of the frequent pattern does not exist among the closed pattern candidates having the same degree of support as the frequent pattern, Closed frequent pattern mining method. 6. The method of claim 5,
Wherein updating the closed frequent pattern list comprises:
Determining whether a closed frequent pattern candidate having the same support record as the support record of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern exists if the support degree of the frequent pattern is less than a preset value,
A closed frequent pattern using parallel processing for determining whether there is a closed frequent pattern candidate that is a subset of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern when the support degree of the frequent pattern is equal to or greater than a predetermined value, Mining method. The method according to claim 1,
Generating and storing projection data having the frequent pattern as a prefix pattern,
A frequent pattern mining method using parallel processing for generating only one projection data from the allocated projection data and storing the allocated projection data together with a frequent item to be projected next when the number of stored projection data is equal to or larger than a preset value . A projection data provider for storing projection data generated at a plurality of processing nodes and assigning the stored projection data to a plurality of processing nodes according to a priority;
A closed frequent pattern manager that stores a closed frequent pattern list containing closed frequent pattern candidates; And
Generating a frequent pattern using the projection data allocated from the projection data providing unit to update the closed frequent pattern list, and when the frequent pattern is added to the closed frequent pattern list, CLAIMS What is claimed is: 1. A closed frequent pattern mining device using parallel processing comprising a plurality of processing nodes for generating data. 12. The method of claim 11,
Wherein the projection data providing unit comprises:
Wherein the stored projection data is allocated according to a priority according to a depth-first search method based on a prefix pattern of the stored projection data. 12. The method of claim 11,
Wherein any one of the plurality of processing nodes comprises:
A closed frequent pattern mining device using parallel processing to generate projection data with a frequent item as a prefix pattern after a frequent item is found in a database. 12. The method of claim 11,
The plurality of processing nodes comprising:
Wherein the frequent item is found in the allocated projection data and then the frequent item is combined with the prefix pattern of the allocated projection data to generate a frequent pattern. 12. The method of claim 11,
The plurality of processing nodes comprising:
Wherein the closed frequent pattern list is updated based on whether a closed frequent pattern candidate having the same degree of support as the frequent pattern exists in the closed frequent pattern list. 16. The method of claim 15,
The plurality of processing nodes comprising:
And adding the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate having the same degree of support as the frequent pattern in the closed frequent pattern list. 16. The method of claim 15,
The plurality of processing nodes comprising:
A closed frequent pattern using parallel processing for adding the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate that includes the frequent pattern as a subset among the closed frequent patterns having the same degree of support as the frequent pattern; Mining device. 16. The method of claim 15,
The plurality of processing nodes comprising:
A parallel processing for adding the frequent pattern to the closed frequent pattern list when there is no closed frequent pattern candidate having the same support record as the support record of the frequent pattern among the closed frequent pattern candidates having the same degree of support as the frequent pattern; Closed frequent pattern mining device. 16. The method of claim 15,
The plurality of processing nodes comprising:
Determining whether a closed frequent pattern candidate having the same support record as the support record of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern exists if the support degree of the frequent pattern is less than a preset value,
A closed frequent pattern using parallel processing for determining whether there is a closed frequent pattern candidate that is a subset of the frequent pattern among the closed pattern candidates having the same degree of support as the frequent pattern when the support degree of the frequent pattern is equal to or greater than a predetermined value, Mining device. 12. The method of claim 11,
The plurality of processing nodes comprising:
A parallel processing for generating only one projection data from the allocated projection data and storing the allocated projection data together with a frequent item to be projected next when the number of projection data stored in the projection data providing unit is equal to or larger than a preset value Frequently used pattern mining device.

Images