Constraint-based Sequential Pattern Mining using Search Position Indexes

Abstract

This paper presents a time interval constraint-based sequential pattern mining algorithm using search position indexes and an anti-monotone property of a maximum item. Time interval constraints are useful for many sequential pattern mining applications using sequence data with a timestamp. However, a non constraint based algorithm outputs redundant sequential patterns not satisfying constraints and previous constraint based algorithms tend to be unstable for an item duplication rate per itemset. Although a previous method with indexes and a property of a maximum item tends to be stable for the rate, the method is not applicable to the constraints. In this paper, indexes applicable to the constraints and an algorithm with the indexes are proposed. The indexes are used to skip searching for infrequent patterns. In addition, it is showed that the algorithm is effective for the case that a range of the time intervals is wide.