MFTPM: Maximum Frequent Traversal Pattern Mining with Bidirectional Constraints

Abstract

An important application of sequential mining technique is maximal frequent traversal pattern mining, since users' traversal pattern and motivation are latent in session sequence at some time segment. In this paper, a Frequent Traversal Pattern Tree structure with dwell time (FTP-Tree) is designed to store, compress the session database, and simplify the configuration of dwell time thresholds during mining. A novel algorithm based on bidirectional constraints, called Maximal Frequent Traversal Pattern Mining (MFTPM) is presented, which traverses quickly FTP-Tree and discovers maximal frequent traversal patterns from the session sequences. Experimental results show that MFTPM can significantly reduce the average execution time and the storage space for mining maximal frequent traversal patterns. Our performance study shows that MFTPM performs muth better than previous approaches in the time constraint environment.