Abstract

Sequential pattern mining is an important data mining method for determining time-related behavior in sequence databases. Recent years have seen a growth of discovering sequential patterns with time constraints such as minimum gap, maximum gap and duration. In this paper, we study various time constraints from both sequence level and event level in one general framework. On the sequence level, duration and timeliness constraints are used to balance important sequential patterns in both the long run and in recent periods. While on the event level, we incorporate minimum gap and maximum gap constraints to confine time period between adjacent events. Based on the pattern-growth strategy, an algorithm, called SE-Growth (Sequence Event Growth), is proposed to discover time-constrained sequential patterns based on suffix projection. Extensive experimental results show that the SE-Growth algorithm is efficient, effective and scalable.

Paper Details