Mining Top-K Closed Frequent Traversal Sequences from Session Streams

Approximate mining top-k closed frequent traversal sequence (Topk_CFTS) has two methods, namely, false-positive oriented and false-negative oriented. Most false-positive approaches require relaxation ratio rho to control memory consumption and mining accuracy. However, it is difficult for users to provide a proper rho value. A higher rho may reduce output precision, and a smaller rho will make memory consumption large. To resolve the conflict, this paper designs a regulatory factor to adjust rho value, and proposes an algorithm, TStream, based on false-negative method to maintain the set of Topk_CFTS in session streams with time-sensitive sliding windows. Experiments show that TStream algorithm performs much better than many established algorithms for mining Topk_CFTS.