A GSP-based efficient algorithm for mining frequent sequences

This paper studies the problem of mining frequent sequences in transactional databases. In [6], Agrawal and Srikant proposed the GSP algorithm for extracting frequently occurring sequences. GSP is an iterative algorithm. It scans the database a number of times depending on the length of the longest frequent sequences in the database. The I/O cost is thus substantial if the database contains very long frequent sequences. In this paper, we extend the candidate generating function used by GSP and propose a new two-stage algorithm ATS. Our algorithm first mines a sample of the database to obtain a rough estimate of the frequent sequences and then refines the solution. Experiment results show that MFS saves I/O cost significantly compared with GSP.