Multi-Level Linear Location Tree for Efficient Sequential Pattern Mining

Discovering sequential patterns from a large database of sequences is an important problem in the field of knowledge discovery and data mining. Most of the previously used sequential pattern mining data structure follows a complex hash-tree structure which has poor locality. Since the hash-tree uses the repetition of a simple search instead of a consecutive search, the hash-based sequential pattern mining, though it reduces search time, may exhibit low performance. In this paper, we propose a novel multi-level linear location tree (or MLLT in short) with an excellent locality structure and develop an efficient MLLT-tree based sequential pattern mining algorithm for mining the complete set of frequent sequence by linear copy of nodes. Our performance study shows that MLLT runs considerably faster than the hash-tree based GSP algorithm.