An Efficient Approach for Interactive Mining of Frequent Itemsets

There have been many studies on efficient discovery of frequent itemsets in large databases. However, it is nontrivial to mine frequent itemsets under interactive circumstances where users often change minimum support threshold (minsup) because the change of minsup may invalidate existing frequent itemsets or introduce new frequent itemsets. In this paper, we propose an efficient interactive mining technique based on a novel vertical itemset tree (VI-tree) structure. An important feature of our algorithm is that it does not have to re-examine the existing frequent itemsets when minsup becomes small. Such feature makes it very efficient for interactive mining. The algorithm we proposed has been implemented and its performance is compared with re-running Eclat, a vertical mining algorithm, under different minsup. Experimental results show that our algorithm is over two orders of magnitude faster than the latter in average.