Multi-Source Data Stream Online Frequent Episode Mining

Abstract

Online frequent episode mining is more complicated than the traditional static frequent episode mining due to the continuous, unbounded and time-varying data stream. Especially in the multiple data streams, online frequent episode mining is more difficult than the single-source stream, due to the concurrency, global clock loss, and uncertainty of delay caused by the distributed environment. To cope with these problems, we propose a new algorithm. Firstly, the data stream with 'happen-before' relationship among multiple sources is combined on the global data lattice. Next, the traversal on global data lattice generates effective parallel and serial candidate data streams, which guarantee the accuracy of subsequent mining and reduce the number of global sequences during searching process. Then, we use the frequent episode tree to detect the expanding online serial episodes and parallel episodes. Finally, we verify the effectiveness and efficiency of the proposed methods through extensive experiments.