Abstract: Sequential pattern mining has been an essential mining task and an active research area in recent years. However, existing sequential pattern mining algorithms are designed for closed itemsets or simple closed sectuential patterns,and can hardly extract composite sequential patterns, an important class of patterns consisting of several short segments separated by gaps. An efficient algorithm for mining frequent closed composite sequences with any number of segments of different lengths, CloCSP, was proposed. It adopts a novel composite strategy called Mixed Composite, which not only can produce all of closed composite sequential patterns, but also can efficiently prune the composite space and simultaneously check the sequential patterns closure, accordingly reduces the cost in both runtime and space usage. Experiments on both synthetic and real data have demonstrated that CloCSP can significantly discover all of closed composite sectuential patterns.

Key words: Frequent sequences,Closed composite sequences,Composite Motif,Data mining