System and method for parallel mining of association rules in databases

• US 5,842,200 A
• Filed: 07/11/1995
• Issued: 11/24/1998
• Est. Priority Date: 03/31/1995
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Agrawal, Rakesh, Shafer, John Christopher
• Primary Examiner(s): Cosimano, Edward R.
• Application Number:
  US08/500,717
• Time in Patent Office: 1,232 Days
• Field of Search: 364/400, 364/554, 395/201, 395/210, 395/601, 395/603, 395/899, 705/1, 705/10, 707/1, 707/3
• US Class Current: 1/1
• CPC Class Codes: G06F 2216/03 Data mining
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

A multiprocessor including a plurality of processing systems is disclosed for discovering consumer purchasing tendencies. Each processing system of the multiprocessor identifies consumer transaction itemsets that are stored in a database that is distributed among the processing systems and which appear in the database a user-defined minimum number of times, referred to as minimum support. Then, the system discovers association rules in the itemsets by comparing the ratio of the number of times each of the large itemsets appears in the database to the number of times particular subsets of the itemset appear in the database. When the ratio exceeds a predetermined minimum confidence value, the system outputs an association rule which is representative of purchasing tendencies of consumers.

Claims

1. A digital multiprocessor comprising a plurality of processing systems, each processing system including a respective local database having distributed therein data from a transaction database of itemsets purchased in consumer transactions, each processing system including:

means for determining the number of times a candidate itemset appears in the associated local database to establish a local count for the candidate itemset;

means for using the local count to determine whether the number of times a candidate itemset appears in the transaction database exceeds a predefined minimum support value;

means for entering a candidate itemset into a set of large itemsets when the number of times exceeds a predetermined minimum support value, such that the set of large itemsets can be designated as frequently occurring itemsets in transactions;

means for determining, for at least some of the itemsets in the set of large itemsets the number of times selected subsets of the itemsets appear in the transaction database;

means for outputting an association rule when the ratio of the number of times a selected subset appears in the transaction database to the number of times the associated itemset appears in the transaction database exceeds a predetermined minimum confidence value and thereby satisfies a minimum confidence constraint; and

means for exchanging the local count with the other processing systems such that each processing system determines whether the number of times a candidate itemset appears in the transaction database exceeds the predefined minimum support value.

2. The digital multiprocessor of claim 1, wherein each processing system includes means for generating a global set G of large itemsets and partitioning the global set G among the n processing systems such that a local set Gⁱ comprising approximately G/n large itemsets is distributed to each system, and each system establishes the local count using the local set Gⁱ.

3. The digital multiprocessor of claim 1, wherein each processing system further comprises:

means for generating the candidate itemsets by concatenating itemsets in the set of large itemsets in accordance with a predetermined concatenation regime, each of the processing systems being associated with a respective set of 1/n candidate large itemsets;

means for receiving remote portions of the transaction database from the other processing systems; and

means for establishing the local counts by alternatingly determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database and in the remote portions of the transaction database.

4. The digital multiprocessor of claim 3, wherein each processing system further comprises:

means for establishing n-1 receive buffers for receiving the remote portions of the transaction database; and

means for establishing the local counts by determining the number of times one or more of the 1/n candidate large itemsets appear in the remote portions of the transaction database when a receive buffer is full and otherwise determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database.

5. The digital multiprocessor of claim 4, wherein each processing system further comprises:

means for entering a candidate itemset into a local set of large itemsets when the respective local count exceeds a predetermined minimum support value;

means for transmitting the local set of large itemsets to the other processing systems;

means for receiving from the other processing systems respective remote sets of large itemsets; and

means for determining the sum of the local set and the remote sets to establish a global set of large itemsets.

6. A computer-implemented method for identifying frequently recurring itemsets in transactions which are stored in a transaction database, the transaction database being distributed among a first local database associated with a first digital processing system and at least a second local database associated with a second digital processing system, comprising:

(a) determining the number of times a candidate itemset appears in the first local database to establish a local count for the candidate itemset;

(b) exchanging the local counts with the at least second processing system;

(c) using the local count to determine whether the number of times a candidate itemset appears in the transaction database exceeds a predefined minimum support value;

(d) entering a candidate itemset into a set of large itemsets when the number of times exceeds a predetermined minimum support value;

(e) for at least some of the itemsets in the set of large itemsets, determining the number of times selected subsets of the itemsets appear in the transaction database; and

(f) outputting an association rule when the ratio of the number of times a selected subset appears in the transaction database to the number of times the associated itemset appears in the transaction database exceeds a predetermined minimum confidence value and thereby satisfies a minimum confidence constraint.

7. The computer implemented method of claim 6, wherein the transaction database is distributed among n processing systems, further comprising:

generating the candidate itemsets by concatenating itemsets in the set of large itemsets in accordance with a predetermined concatenation regime, each of the processing systems being associated with a respective set of 1/n candidate large itemsets;

establishing n-1 receive buffers;

receiving remote portions of the transaction database from the other processing systems in respective receive buffers; and

establishing the local counts by determining the number of times one or more of the 1/n candidate large itemsets appear in the remote portions of the transaction database when a receive buffer is full, and otherwise determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database.

8. The computer implemented method of claim 6, wherein the transaction database is distributed among n processing systems, and wherein the method includes generating a global set G of large itemsets and partitioning the global set G among the n processing systems such that a local set Gⁱ comprising approximately G/n large itemsets is distributed to each system, and each system establishes the local count using the local set Gⁱ.

9. A system for determining association rules in a transaction database of itemsets distributed among a plurality of local databases, each being associated with a respective processing system, each processing system including:

a large itemset generator for generating candidate large itemsets in accordance with a predetermined concatenation regime;

a counter for counting the number of times selected of the candidate large itemsets appear in the local database to establish a local count;

an exchanger for exchanging data representative of the local count or local database between the processing system and the other processing systems, wherein the system includes;

at least one rule generator for generating association rules based on the local counts.

10. The system of claim 9, wherein each processing system includes means for entering a candidate itemset into a set of large itemsets when the number of times the candidate large itemset appears in the transaction database exceeds a predetermined minimum support value.

11. The system of claim 10, wherein the rule generator determines, for at least some of the itemsets in the set of large itemsets, the number of times selected subsets of the itemsets appear in the transaction database, and wherein the rule generator outputs an association rule when the ratio of the number of times a selected subset appears in the transaction database to the number of times the associated itemset appears in the transaction database exceeds a predetermined minimum confidence value and thereby satisfies a minimum confidence constraint.

12. The system of claim 11, wherein each processing system includes means for exchanging the local counts with the other processing systems such that each processing system determines whether the number of times a candidate itemset appears in the transaction database exceeds the predefined minimum support value.

13. The system of claim 11, wherein the transaction database is distributed among n processing systems and each processing system further comprises:

means for generating the candidate itemsets by concatenating itemsets in the set of large itemsets in accordance with a predetermined concatenation regime each of the processing systems being associated with a respective set of 1/n candidate large itemsets;

means for establishing n-1 receive buffers for receiving respective remote portions of the transaction database from the other processing systems;

means for establishing the local counts by determining the number of times one or more of the 1/n candidate large itemsets appear in the remote portions of the transaction database when a receive buffer is full, and otherwise determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database;

means for entering a candidate itemset into a local set of large itemsets when the respective local count exceeds a predetermined minimum support value;

means for transmitting the local set of large itemsets to the other processing systems;

means for receiving from the other processing systems respective remote sets of large itemsets; and

means for determining the sum of the local set and the remote sets to establish a global set of large itemsets.

14. The system of claim 11, wherein the transaction database is distributed among n processing systems, and each processing system further comprises:

means for generating a global set G of large itemsets and partitioning the global set G among the n processing systems such that a local set Gⁱ comprising approximately G/n large itemsets is distributed to each processing system, and each processing system establishes the local count using the local set Gⁱ.

15. A computer program device comprising:

a computer program storage device readable by a first digital processing system; and

a program means on the program storage device and including instructions executable by the digital processing system for performing method steps for identifying frequently recurring itemsets in transactions which are stored in a transaction database, the transaction database being distributed among a first local database associated with the first digital processing system and at least a second local database associated with a second digital processing system, the method steps comprising;

determining the number of times a candidate itemset appears in the first local database to establish a local count for the candidate itemset;

using the local count to determine whether the number of times a candidate itemset appears in the transaction database exceeds a predefined minimum support value;

entering a candidate itemset into a set of large itemsets when the number of times exceeds a predetermined minimum support value, such that the set of large itemsets can be designated as frequently occurring itemsets in transactions;

for at least some of the itemsets in the set of large itemsets, determining the number of times selected subsets of the itemsets appear in the transaction database;

outputting an association rule when the ratio of the number of times a selected subset appears in the transaction database to the number of times the associated itemset appears in the transaction database exceeds a predetermined minimum confidence value and thereby satisfies a minimum confidence constraint; and

after establishing the local count, exchanging the local counts with the at least second processing system such that each processing system determines whether the number of times a candidate itemset appears in the transaction database exceeds the predefined minimum support value.

16. The program storage device of claim 15, wherein the transaction database is distributed among n processing systems, and wherein the method steps include generating a global set G of large itemsets and partitioning the global set G among the n processing systems such that a local set Gⁱ comprising approximately G/n large itemsets is distributed to each system, and each system establishes the local count using the local set Gⁱ.

17. The program storage device of claim 15, wherein the transaction database is distributed among n processing systems, and the method steps further comprise:

generating the candidate itemsets by concatenating itemsets in the set of large itemsets in accordance with a predetermined concatenation regime, each of the processing systems being associated with a respective set of 1/n candidate large itemsets;

receiving remote portions of the transaction database from the other processing systems; and

establishing the local counts by alternatingly determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database and in the remote portions of the transaction database.

18. The program storage device of claim 17, wherein the method further comprises:

establishing n-1 receive buffers for receiving the remote portions of the transaction database; and

establishing the local counts by determining the number of times one or more of the 1/n candidate large itemsets appear in the remote portions of the transaction database when a receive buffer is full, and otherwise determining the number of times one or more of the 1/n candidate large itemsets appear in the respective local database.

19. The program storage device of claim 18, wherein the method steps further comprise:

entering a candidate itemset into a local set of large itemsets when the respective local count exceeds a predetermined minimum support value;

transmitting the local set of large itemsets to the other processing systems;

receiving from the other processing systems respective remote sets of large itemsets; and

determining the sum of the local set and the remote sets to establish a global set of large itemsets.