KR101187958B1 - Apparatus and method for extracting association rule - Google Patents

Abstract

PURPOSE: An association rules extracting device and a method thereof are provided to extract association rules without lack of a memory. CONSTITUTION: A partitioning unit(330) partitions a transaction population into N transaction segments. A frequent pattern tree generating unit(340) generates frequent pattern trees according to the N transaction segments. A conditional pattern base generating unit(350) generates conditional pattern bases according to the transaction segments. The conditional pattern base generating unit generates the conditional pattern bases about the transaction population. An association rules extracting unit(360) extracts association rules by using the conditional pattern bases. [Reference numerals] (310) Header table generating unit; (320) Arranging unit; (330) Partitioning unit; (340) Frequent pattern tree generating unit; (350) Conditional pattern base generating unit; (360) Association rules extracting unit; (AA) Association rules extracting device

Description

Association rule extraction device and method {APPARATUS AND METHOD FOR EXTRACTING ASSOCIATION RULE}

Embodiments of the present invention relate to an association rule extraction apparatus and method, and more particularly, to an association rule extraction apparatus and method for extracting association rules without running out of memory in an environment having a large amount of data and limited memory capacity. It is about.

An association rule is a useful pattern expressed as a conditional-conclusion expression of the form "X⇒Y (s, c)" extracted from transaction data such as a customer's purchase list. . Here, X and Y are items present in the transaction data, and "X⇒Y" means that the item Y corresponding to the result occurs together when the item X corresponding to the condition occurs. And, S means "Support" which is the ratio of transaction data in which item X and item Y occur simultaneously in the total transaction data, and C is a transaction including item Y together in the transaction data including item X. "Confidence" is the ratio of data.

Association rule extraction method, popularly known as shopping cart analysis, finds useful patterns among items that exist within a large amount of data, such as merchandise purchase lists of mart customers and movie watching records of TV viewers. Can be used for analysis. For example, an analysis of customers' purchase records at a US mart extracts the rule "Consumers who buy beer buy diapers together", which leads to a close sales position of beer and diapers. There is a famous example that brought.

Representative prior arts of such association rule extraction methods include the Apriori algorithm and the FP-growth algorithm using a frequent pattern tree (FP).

The Apriori algorithm extracts association rules by finding k + 1 frequent item sets using k frequent item sets. That is, if the k set of items is frequent, the Apriori algorithm may also have (k + 1) sets of items that contain all of the items contained in that set of items, but if the set of k items is not frequent, The association rule is extracted using an iterative approach that the set of k + 1 items that contain all of the included items is also not frequent.

Such Apriori algorithm has to generate candidate item set of k + 1 item sets that are likely to occur frequently. For this purpose, it is necessary to access database repeatedly, which requires a lot of storage space and long processing time. There is this.

The FP-growth algorithm generates a frequent pattern tree after sorting the number of occurrences (the frequency of occurrences) of items in a plurality of transaction data including at least one item by item, and generates an association rule using the generated frequent pattern tree. Extract Since the FP-growth algorithm does not need to access the database repeatedly and does not generate candidate item sets for frequent pattern extraction, the association rule extraction speed is high.

1 to 2, the association rule extraction method (that is, the association rule extraction method using the frequent pattern tree) according to the FP-growth algorithm will be described in detail below.

First, for all items included in a plurality of transaction data belonging to a transaction population, a frequency of frequency for items (that is, frequent items) appearing in the transaction data above a desired (preset) minimum support degree is calculated.

The table shown in FIG. 1A is an example of a plurality of transaction data (transaction population) including at least one item. Here, 'TID' means each transaction data, and 'List of Item_IDs' means an item included in each transaction data. In this case, if the minimum support is 20% of the total transaction data, items appearing in two or more of the total nine transactions are extracted as frequent items.

Next, a header table is generated by sorting the frequent items according to the number of occurrences (frequency). The header table refers to a table in which frequent items appearing more than the minimum support are sorted according to the number of appearances. 'Item_ID' means frequent items and 'Support count' means number of occurrences of each item. FIG. 1B shows a header table in the above example.

When the header table is completed, the transaction collation / purification process removes items (that is, non-frequent items) not included in the header table from the plurality of transaction data and sorts the remaining frequent items in the order of the header table by transaction data. Do this. Figure 1 (c) shows the results of the alignment / purification process in the above example.

If only frequent items remain for each transaction data through the above process, items in each transaction data are inserted into nodes in order to generate a frequent pattern tree. The root node of the generated frequent pattern tree is a null node, and each item is inserted into the tree starting from the root node through the following process.

Step 1: Select the root node as the current node

Step 2: It is determined whether there is a child node whose node value is the first item (that is, the item with the most occurrence number) among the sorted items included in the transaction data among the child nodes of the current node.

Step 3-1: If there is no child node as described above, adding the node having the first item as a node value as a child node of the current node, and setting the number of occurrences of the added child node to 1 After that, select the added child node as the current node.

Step 3-2: If the above child node exists, the number of occurrences of the child node is increased by one, and the child node is selected as the current node.

Step 4: Remove the first item from the item list included in the transaction data.

Thereafter, the second to fourth steps are repeatedly performed until all items included in the plurality of transaction data are removed.

FIG. 2A illustrates an example of a frequent pattern tree generated through the above-described process in the above example.

Subsequently, the association rule is extracted using the generated frequent pattern tree.

To do this, first extract a conditional pattern base for an item in the header table from the frequent pattern tree. A conditional pattern base for a given item means a set of items from a parent node of a (condition) item to a child node of the root node in the frequent pattern tree. One or more such sets of items may be extracted from the frequent pattern tree. The conditional pattern base includes a number of times the condition item and the pattern base appear together, and the number of occurrences of the conditional pattern base is set as the frequency of the condition node in the frequent pattern tree.

For example, in the frequent pattern tree shown in (a) of FIG. 2, the conditional pattern base of I5 has two (I2, I1) and (I2, I1, I3), and the frequency of I5 in the frequent pattern tree is respectively Since it is 1, the frequency of the two paths including it is also 1. FIG. 2B shows the conditional pattern base extracted in the above example.

Looking at the conditional pattern base shown in (b) of FIG. 2, it is easy to determine how many times the condition item appears like other items. For example, looking at the conditional pattern base of I5, I5 appears with (I2, I1, I3), 1 times (I2, I1), 2 with (I2), and 2 with (I1), respectively. It can be seen that. Here, the number of occurrences is the same as the number of occurrences (frequency) in the transaction data, so the frequent item set that appears more than 20% (twice) and includes I5 as (I2, I5), (I1, I5), It is determined by (I2, I1, I5), from which an association rule can be generated that includes I5 at the condition or conclusion.

However, the association rule extraction method according to the conventional FP-growth algorithm is large enough to store and access the frequent pattern tree when the amount of transaction data is huge and the size of the frequent pattern tree for expressing it is large. The disadvantage of using memory is that.

In order to solve the problems of the prior art as described above, the present invention proposes an association rule extraction apparatus and method for extracting association rules without running out of memory in an environment having a large amount of data and limited memory capacity.

Other objects of the present invention may be derived by those skilled in the art through the following examples.

According to a preferred embodiment of the present invention to achieve the above object, based on the capacity of the memory used for the extraction of the association rule (Population) consisting of a plurality of transaction data (Transaction Data) of one or more transaction data A divider for dividing into N (the integers of 2 or more) transaction segments each including; A frequent pattern tree generation unit generating a frequent pattern tree for each of the N transaction segments; Create a conditional pattern base for each transaction segment by using the frequent pattern tree generated for each transaction segment, and generate a conditional pattern base for the transaction population by integrating the conditional pattern base generated for each transaction segment. part; And an association rule extracting unit extracting an association rule using a conditional pattern base for the transaction population.

The division unit may determine the number N of the transaction segments based on the number of the plurality of transaction data and the capacity of the memory.

Analyzing the plurality of transaction data each including at least one item, extracting two or more frequent items appearing in the transaction data more than a predetermined minimum support, and extracting the two or more frequent items And a header table generator for generating a header table arranged according to the number of occurrences after matching the occurrence number of items, wherein the conditional pattern base generator further generates the conditional pattern base for each transaction segment by using the header table. can do.

The frequent pattern tree generation unit removes an item (non-frequent item) that is not two or more frequent items from the plurality of transaction data, and frequently occurs for each of the N transaction segments by using a plurality of transaction data from which the non-frequent item is removed. You can create a pattern tree.

In addition, according to another embodiment of the present invention, the transaction population consisting of a plurality of transaction data based on the capacity of the memory used for the extraction of association rules, N (integer of 2 or more) each containing one or more transaction data Dividing into; Generating a frequent pattern tree for each of the N transaction segments; Generating a conditional pattern base for each transaction segment using the frequent pattern tree generated for each transaction segment; Generating a conditional pattern base for the transaction population by integrating the conditional pattern base generated for each transaction segment; And extracting an association rule using a conditional pattern base for the transaction population.

An apparatus and method for extracting association rules according to the present invention has an advantage of extracting association rules without running out of memory in an environment having a large amount of data and a limited memory capacity.

1 and 2 are diagrams for explaining a conventional association rule extraction method according to the FP-growth algorithm.
3 is a block diagram showing a schematic configuration of an association rule extraction apparatus according to an embodiment of the present invention.
4 to 7 are views for explaining a concept of extracting an association rule through the association rule extraction apparatus according to an embodiment of the present invention.
8 is a flowchart illustrating the overall flow of the association rule extraction method according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

3 is a block diagram showing a schematic configuration of an association rule extraction apparatus according to an embodiment of the present invention.

Referring to FIG. 3, the association rule extracting apparatus 300 according to an embodiment of the present invention may include a header table generator 310, an alignment / purifier 320, a divider 330, and a frequent pattern tree generator ( 340, a conditional pattern base generator 350, and an association rule extractor 360. Hereinafter, the function of each component will be described in detail.

The header table generating unit 310 includes at least one item each, and analyzes a plurality of transaction data belonging to a transaction population to analyze two or more items appearing in transaction data more than a predetermined minimum support. The frequent items are extracted, the extracted two or more frequent items are matched with the number of occurrences of the corresponding frequent items, and a header table is generated according to the number of occurrences. The sorting / purging unit 320 removes the non-frequent items not included in the header table from the plurality of transaction data and performs a transaction data sorting / purifying process of sorting the remaining frequent items in the order of the header table for each transaction data. To perform.

The operations of the header table generator 310 and the sorter / purifier 320 are the same as the operation of generating and sorting / purging the header table in the association rule extraction method according to the conventional FP-growth algorithm described above. The description will be omitted.

The divider 330 divides a transaction population consisting of a plurality of transaction data into N transaction segments each including one or more transaction data, based on a capacity of a memory used for extracting an association rule.

That is, the division unit 330 divides the transaction population by dividing only the capacity that can be processed at one time in the memory (that is, grouping a plurality of transaction data) to generate two or more transaction segments. Unlike the transaction population, the divided transaction segment may not include all the items in the header table determined to be frequent. 4 illustrates an example in which the division unit 330 divides a transaction population into two transaction segments.

According to an embodiment of the present invention, the divider 330 may determine the number N of transaction segments to be divided based on the number of transaction data and the memory capacity included in the transaction population.

That is, the divider 330 may dynamically determine the number of transaction segments in consideration of the total number of transaction data and the memory capacity. As an example, the number N of divided transaction segments may be proportional to the number of transaction data included in the transaction population and inversely proportional to the capacity of the memory.

On the other hand, if the current memory capacity is sufficient to process all the transaction data included in the current transaction population, the partition unit 330 may not perform the creation of the transaction segment (that is, division of the transaction population). have.

The frequent pattern tree generation unit 340 generates a frequent pattern tree for each of the N transaction segments divided (created).

As an example, the frequent pattern tree generator 340 may insert items in each transaction data into nodes in order to generate a frequent pattern tree for each transaction segment. In this case, the frequent pattern tree generation unit 340 may generate a frequent pattern tree for each transaction segment by performing a null node as a root node and performing the first to fourth steps described below for each transaction segment.

Step 1: Select the root node as the current node

Step 2: It is determined whether there is a child node whose node value is the first item (that is, the item with the most occurrence number) among the sorted items included in the transaction data among the child nodes of the current node.

Step 3-1: If there is no child node as described above, adding the node having the first item as a node value as a child node of the current node, and setting the number of occurrences of the added child node to 1 After that, select the added child node as the current node.

Step 3-2: If the above child node exists, the number of occurrences of the child node is increased by one, and the child node is selected as the current node.

Step 4: Remove the first item from the item list included in the transaction data.

Thereafter, the second to fourth steps are repeatedly performed until all items included in all the transaction data present in each transaction segment are removed.

As described above, when the frequent pattern tree is generated for each transaction segment, the generation of the frequent pattern tree for a specific transaction segment does not affect other transaction segments or the procedure for generating the frequent pattern tree. The frequent pattern tree does not need to be stored in memory. Accordingly, the association rule extracting apparatus 300 according to an embodiment of the present invention may extract the association rule using all the transaction data included in the transaction population even using a smaller memory.

According to an embodiment of the present invention, the frequent pattern tree generator 340 may sequentially generate the frequent pattern tree with a predetermined time gap for the N transaction segments.

In FIG. 5, the frequent pattern tree generation unit 340 generates an frequent pattern tree for each transaction segment in the example of FIG. 4.

The conditional pattern base extractor 350 generates a conditional pattern base for each transaction segment by using the frequent pattern tree generated for each transaction segment.

More specifically, according to an embodiment of the present invention, the conditional pattern base generator 350 may include a header table (original header table) for a transaction population generated by the frequent pattern tree and header table generator 310 for each transaction segment. Conditional pattern base for each transaction segment can be created by using.

That is, the conditional pattern base generator 350 generates the conditional pattern base for each transaction segment by using a conventional technique of generating the conditional pattern base from the frequent pattern tree. In this case, the conditional pattern base generator 350 may generate a conditional pattern base for each transaction segment by using a header table (original header table) generated by using a transaction population before splitting rather than a header table for the transaction segment. . Therefore, even if a frequent item exists in the header table, an item not included in the divided transaction segment is not included in the conditional pattern base for the transaction segment.

FIG. 6 illustrates an example of generating a conditional pattern base for each transaction segment by using the frequent pattern tree generated in the example of FIG. 5.

In addition, the conditional pattern base generation unit 350 generates a conditional pattern base for the transaction population by integrating the conditional pattern base generated for each transaction segment.

That is, the conditional pattern base generation unit 350 fuses the conditional pattern base for each transaction segment by item, and is identical to the conditional pattern base for the transaction population extracted at one time according to the association rule extraction method using a conventional frequent pattern tree. Create a conditional pattern base. As an example, the conditional pattern base generator 350 may integrate the conditional pattern base for each transaction segment as follows.

(1) Among the condition items of the conditional pattern base of each transaction segment, if there is a condition item included only in the conditional pattern base of one transaction segment, the condition item is included in the final conditional pattern base list.

(2) For the condition items included in the conditional pattern base of two or more transaction segments, all pattern bases subject to the item in the conditional pattern base of each transaction segment are collected and fused as follows.

(2-1) If the collected pattern base list has the same pattern base, add the pattern base to the final conditional pattern base list for the condition item, but the final frequency is the conditional pattern of all transaction segments in which the pattern base appears. Determined by the sum of frequencies at the base.

(2-2) Add all the pattern bases in the collected pattern base list except the pattern base added in (2-1) to the final conditional pattern base list for the condition item.

(3) Repeat (1) to (2) for all items in the header table (original header table) for the transaction population to form the final conditional pattern base list.

FIG. 7 illustrates an example of generating a conditional pattern base for a transaction population by integrating the conditional pattern base for each transaction segment generated in the example of FIG. 6.

Finally, the association rule extractor 360 extracts the association rule using the integrated conditional pattern base (ie, the conditional pattern base for the transaction population).

Extracting the association rule from the conditional pattern base is a well known technique, and thus a detailed description thereof will be omitted.

Meanwhile, each component described with reference to FIG. 3 may mean software or hardware such as a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC). However, the above components are not limited to software or hardware, and may be configured to be recorded in an addressable storage medium and may be configured to execute one or more processors. The functions provided in the above components may be implemented by more detailed components, or may be implemented as one component that performs a specific function by combining a plurality of components.

8 is a flowchart illustrating the overall flow of the association rule extraction method according to an embodiment of the present invention. Hereinafter, a process performed for each step will be described.

First, in step S810, a transaction population composed of a plurality of transaction data each including at least one item is analyzed to extract two or more frequent items appearing in the transaction data with a predetermined minimum support or more.

In operation S820, the extracted two or more frequent items are matched with the number of occurrences of the corresponding frequent items, and then a header table is generated according to the number of occurrences.

Subsequently, in step S830, a transaction population consisting of a plurality of transaction data is divided into N transaction segments each containing one or more transaction data based on the capacity of the memory used for extracting the association rule.

As an example, in step S830, the number (N) of transaction segments to be divided is determined based on the number of transaction data and the memory capacity included in the transaction population, and the transaction population is converted into transaction segments according to the determined number. Can be divided

Next, in step S840, a frequent pattern tree is generated for each N transaction segments.

According to an embodiment of the present invention, in step S840, the two or more non-frequent items (non-frequent items) are removed from a plurality of transaction data, and N is obtained by using a plurality of transaction data from which non-frequent items are removed. A frequent pattern tree can be created for each transaction segment.

Thereafter, in step S850, the conditional pattern base is generated for each transaction segment by using the frequent pattern tree generated for each transaction segment.

According to an embodiment of the present invention, in step S850, the conditional pattern base may be generated for each transaction segment by using the header table for the transaction population before the partition rather than the header table for the transaction segment.

Meanwhile, the generating of the frequent pattern tree for each N transaction segments (S840) and the generating of the conditional pattern base for each N transaction segments (S850) may include generating a conditional pattern base of the frequent pattern tree. May be sequentially performed at, or may be performed simultaneously in N processing apparatuses.

In operation S860, the conditional pattern base generated for each transaction segment is integrated to generate a conditional pattern base for the transaction population. That is, in step S860, the conditional pattern base for each transaction segment is fused for each item, and the same conditional pattern base as the conditional pattern base for the transaction population extracted at one time is extracted according to the association rule extraction method using a conventional frequent pattern tree. Create

Finally, in step S870, the association rule is extracted using the conditional pattern base on the transaction population.

Embodiments of the association rule extraction method according to the present invention have been described so far, and the configuration of the association rule extraction device 300 described above with reference to FIG. 3 is also applicable to the present embodiment. Hereinafter, a detailed description will be omitted.

In addition, embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks, and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Examples of program instructions, such as magneto-optical and ROM, RAM, flash memory and the like, can be executed by a computer using an interpreter or the like, as well as machine code, Includes a high-level language code. The hardware device described above may be configured to operate as one or more software modules to perform the operations of one embodiment of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and limited embodiments and drawings. However, it should be understood that the present invention is not limited to the above- Various modifications and variations may be made thereto by those skilled in the art to which the present invention pertains. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

300: association rule extraction apparatus 310: header table generation unit
320: alignment / purification part 330: division part
340: frequent pattern tree generator 350: conditional pattern base generator
360: association rule extractor

Claims (8)

Partitioning a transaction population consisting of a plurality of transaction data into N (integer or greater) two transaction segments each containing one or more transaction data based on the amount of memory used to extract the association rule. Installment;
A frequent pattern tree generation unit generating a frequent pattern tree for each of the N transaction segments;
Create a conditional pattern base for each transaction segment by using the frequent pattern tree generated for each transaction segment, and generate a conditional pattern base for the transaction population by integrating the conditional pattern base generated for each transaction segment. part; And
Association rule extraction unit for extracting an association rule using a conditional pattern base for the transaction population
Association rule extraction apparatus comprising a. The method of claim 1,
And the division unit determines the number (N) of the transaction segments based on the number of the plurality of transaction data and the capacity of the memory. The method of claim 1,
Analyzing the plurality of transaction data each including at least one item, extracting two or more frequent items appearing in the transaction data more than a predetermined minimum support, and extracting the two or more frequent items A header table generation unit that matches a number of occurrences of an item and generates a header table sorted according to the number of occurrences
Further comprising:
And the conditional pattern base generator further generates a conditional pattern base for each transaction segment using the header table. The method of claim 3,
The frequent pattern tree generation unit removes an item (non-frequent item) that is not two or more frequent items from the plurality of transaction data, and frequently occurs for each of the N transaction segments by using a plurality of transaction data from which the non-frequent item is removed. Association rule extraction apparatus, characterized in that for generating a pattern tree. Dividing a transaction population consisting of a plurality of transaction data into N transaction segments each comprising one or more transaction data based on a capacity of a memory used for extracting an association rule;
Generating a frequent pattern tree for each of the N transaction segments;
Generating a conditional pattern base for each transaction segment using the frequent pattern tree generated for each transaction segment;
Generating a conditional pattern base for the transaction population by integrating the conditional pattern base generated for each transaction segment; And
Extracting an association rule using a conditional pattern base on the transaction population
Association rule extraction method comprising a. The method of claim 5,
Analyzing the plurality of transaction data each including at least one item, extracting two or more frequent items appearing in the transaction data by a predetermined minimum support degree, and matching the two or more frequent items with the number of occurrences of the frequent items. Generating a header table sorted according to the number of occurrences
Further comprising:
The generating of the conditional pattern base may further include generating a conditional pattern base for each transaction segment using the header table. The method according to claim 6,
The generating of the frequent pattern tree may include removing the two or more non-frequent items (non-frequent items) from the plurality of transaction data, and using the plurality of transaction data from which the non-frequent items are removed, the N transactions. An association rule extraction method comprising generating a frequent pattern tree for each segment. A computer-readable recording medium having recorded thereon a program for performing the method of any one of claims 5 to 7.

Images