A Mediator Exploiting Approach for Mining Indirect Associations from Web Data Streams

Recently, the concept of indirect associations, a new type of infrequent patterns that indirectly connect two rarely co-occurred items via a frequent item set called "mediator", has been shown its power in capturing interesting information over web usage data. Most contemporary indirect association mining algorithms are developed for static dataset. Our previous work has proposed an algorithm, MIA-LM, tailored to streaming data. In this paper, we propose a new efficient algorithm, namely EMIA-LM, for mining indirect associations over web data streams. EMIA-LM employs a mediator-exploiting search strategy, which reduce the search space as well as computation cost for generating indirect associations. Besides, EMIA-LM adopts a compact data structure, alleviating unnecessary data transforming processes and consuming far less memory storage. Preliminary experiments conducted on real Web streaming datasets show that EMIA-LM is superior to the leading HI-mine* algorithm for static data and MIA-LM both in computation speed and memory consumption.