Efficiently enumerating frequent itemsets using combinatorial number system. Issue 2 (2nd April 2016)

• Record Type: Journal Article
• Authors: Tomovic, S.; Stanisic, P.
• Is Part Of: International journal of computer mathematics. Volume 1:Issue 2(2016)
• Page Start: 43
• Page End: 56
• Publication Date: 2016-04-02
• Subjects: Frequent itemset mining -- Apirori algorithm -- combinatorial number system; 68P20 -- 68P15 -- 68P10; Computer systems -- Periodicals; Computer systems; Periodicals; 004
• DOI: 10.1080/23799927.2016.1218934
• Languages: English
• ISSNs: 2379-9927

Abstract

ABSTRACT: In this paper we establish novel theoretical framework for frequent itemset mining problem, as well as new and efficient join-based algorithm. The algorithm maintains one-dimensional array rank, starting from k =2nd iteration. The algorithm uses combinatorial number schema to map k -candidates to indexes of array. At the index r of the array, the algorithm stores RANKC of r th candidate in the lexicographic order, so, where is the r th candidate in lexicographic order in iteration k . Having in hand rank array, the algorithm will join two candidates iff their ranks are equal, making join an operation. Also, we believe that candidate ranking by combinatorial number system can be effectively integrated into pattern-growth algorithms, that are state of the art in frequent itemset mining, and additionally improve their performances.