Maintaining and Mining Sequential Patterns in Incremental Sequence database

Abstract

Real world databases are dynamic, new data are stored into database over time. The most naïve solution of mining sequential patterns over an incremental database is to rerun the database from scratch, which didn’t take the advantage of previous work. The other way is that merge the new sequential patterns set with previous discovered sequential patterns. However, Algorithms that pruned off infrequent sequences are essentially not suitable for merging due to the information loss. In this paper, we proposed a novel algorithm IMSP that transform original sequence database into a frequency data model. In the model constructing process, no candidates were generated and with only one database scan. The advantages of IMSP are proven by example.