Mining general temporal association rules for items with different exhibition periods

In this paper we explore a new model of mining general temporal association rules from large databases where the exhibition periods of the items are allowed to be different from one to another. Note that in this new model, the downward closure property which all prior Apriori-based algorithms relied upon to attain good efficiency is no longer valid. As a result, how to efficiently generate candidate itemsets form large databases has become the major challenge. To address this issue, we develop an efficient algorithm, referred to as algorithm SPF (standing for Segmented Progressive Filter) in this paper The basic idea behind SPF is to first segment the database into sub-databases in such a way that items in each sub-database will have either the common starting time or the common ending time. Then, for each sub-database, SPF progressively filters candidate 2-itemsets with cumulative filtering thresholds either forward or backward in time. This feature allows SPF of adopting the scan reduction technique by generating all candidate k-itemsets (k>2) from candidate 2-itemsets directly. The experimental results show that algorithm SPF significantly outperforms other schemes which are extended from prior methods in terms of the execution time and scalability.