Mining top-rank-k frequent weighted itemsets using WN-list structures and an early pruning strategy

Abstract

Frequent weighted itemsets (FWIs) are a variation of frequent itemsets (FIs) that take into account the different importance or weights for each item. Many algorithms have been introduced for mining FWIs recently. However, the traditional algorithms for mining FWIs produce a large number of FWIs which causes difficulties when applied with intelligent systems. Therefore, this study first introduces the problem of mining top-rank-k FWIs from weighted databases that combines the mining and ranking phases into one without finding all FWIs to increase their usability in practical applications. As the second contribution, three baseline algorithms for mining top-rank-k FWIs, namely TFWIT, TFWID and TFWIN that use state-of-the-art data structures, namely tidset, diffset and WN-list structures, are developed. Next, this study proposes the threshold raising strategy and the early pruning strategy supported by a new theorem to effectively mine top-rank-k FWIs. An improved version of TFWIN named TFWIN+ employs these strategies to improve the performance of mining top-rank-k FWIs and is more efficient when compared to the original version. Finally, the empirical evaluations in terms of processing time and memory usage among these algorithms were conducted to show the effectiveness of TFWIN+. The experimental results show that TFWIN+ outperforms TFWIT, TFWID and TFWIN for mining top-rank-k FWIs.