Mining Temporal Sequential Patterns Based on Multi-Granularities

Sequential pattern mining is an important data mining problem that can extract frequent subsequences from sequences. However, the times between successive items in a sequence is typically used as user-specified constraints to pre-process the input data or to prune the pattern search space. In either cases, the times cannot be used to identify item intervals of sequential patterns. In this paper, we introduce a form of multi-granularity sequence patterns, which is a sequential pattern where each transition time is annotated with multi-granularity boundary interval and average time derived from the source data rather than the user-predetermined time interval or only a typical time. Then we present a novel algorithm, MG-PrefixSpan, of multiple granularity sequential patterns based on PrefixSpan[, which discovers all such patterns. Empirical evaluation shows that MG-PrefixSpan scales up linearly as the size of database, and has a good scalability with respect to the length of sequence and the size of transaction.