A new framework of mining association rules with time-windows on real-time transaction database

This paper dedicates efforts to discover the part-time association rules in real-time transactional database by extending the traditional minsup-minconf based frame-work to a new one - the minsup, minconf and minwin based framework. We propose a more general form for association rule, i.e., the Association Rule with Time-Windows (ARTW), to properly integrate the temporal association rules together with the normal ones. New notions like Frequent Itemset with Time-Windows (FITW) are also defined, and an Apriori-like algorithm, named TW-Apriori, is developed to fast generate the FITWs. Computational experiments are conducted on two datasets - a synthetic dataset and a real database. Both experiments show that large number of ARTWs ignored previously can be discovered under the new framework; many of them are even very strong rules and valuable for market decisions. The efficiency of the proposed TW-Apriori algorithm is also proven feasible since it can finish the calculation within one minute and the length of the calculation time is nearly proportional to the number of ARTWs found.