Efficient Mining of High Utility Sequential Patterns Over Data Streams

High utility sequential pattern mining has emerged as an important topic in data mining. Although several preliminary works have been conducted on this topic, the existing studies mainly focus on mining high utility sequential patterns (HUSPs) in static databases and do not consider the streaming data. Mining HUSPs over data streams is very desirable for many applications. However, addressing this topic is not an easy task. First, streaming data come continuously in high speed and the mining result should be instantly available when users request it. Second, we need to overcome the problem of combinatorial explosion of a large search space. Third, pruning search space for HUSP mining is difficult because the downward closure property does not hold for the utility of sequences. In this paper, we propose a new framework for mining high utility sequential patterns over data streams, which has not been explored previously. A novel data structure named HUSP-Tree is proposed to maintain the essential information for mining HUSPs. HUSP-Tree can be easily updated when new data arrive and old data expire in a data stream. An efficient and single-pass algorithm named HUSPStream is proposed to generate HUSPs from HUSP-Tree. When data arrive at or leave from a sliding window, HUSP-Stream incrementally updates HUSP-Tree online to find HUSPs based on previous mining results. HUSP-Stream uses a new utility estimation model to more effectively prune the search space. Experimental results on real and synthetic datasets show that our algorithm outperforms the state-of-the-art algorithms and serves as an efficient solution to the new problem of mining high utility sequential patterns over data streams.