Mining frequent sequences using itemset-based extension

Abstract

In this paper, we systematically explore an itemset-based extension approach for generating candidate sequence which contributes to a better and more straightforward search space traversal performance than traditional item-based extension approach. Based on this candidate generation approach, we present FINDER, a novel algorithm for discovering the set of all frequent sequences. FINDER is composed oftwo separated steps. In the first step, all frequent itemsets are discovered and we can get great benefit from existing efficient itemset mining algorithms. In the second step, all frequent sequcnces with at least two frequent itemsets are detected by combining depth-first search and item set-based extension candidate generation together. A vertical bitmap data representation is adopted for rapidly support counting reason. Several pruning strategies are used to reduce the search space and minimize cost of computation. An extensive set ofexperiments demonstrate the effectiveness and the linear scalability of proposed algorithm.