Efficient weighted-frequent-regular itemsets mining using interval word segments structure

Recently, weighted-frequent regular itemset mining has been introduced to discover interesting itemsets based on observation of their occurrence behavior and difference of items' importance. The weighted-support and regularity of occurrence are applied to measure interestingness of itemsets. However, with a large amount of items to be considered, the mining process consumes high computational time and memory. Thus, to efficiently mine such itemsets, this paper introduces a single-pass algorithm, called WFRIM-IWS. The interval word segments structure (IWS) is utilized for maintaining occurrence information of each itemset. A new looked up table is designed to quickly calculate frequency (weighted-support) and regularity of an itemset from IWS. Moreover, to early prune search space, the concept of overestimated weighted-frequency and global/local maximum weights are then applied. Experiments on synthetic and real datasets show that the proposed WFRIM-IWS outperforms the previous algorithm in the terms of computational time and memory consumption.