Mining weighted closed itemsets directly for association rules generation under weighted support framework

Closed itemset mining avoids many duplicate itemsets generation, which derives the whole set of frequent itemsets exactly but is orders of magnitude smaller than the latter. But generally traditional methods assume every two items have same significance in database, which is unreasonable in many real applications. This paper addresses the issues of mining concise association rules with different significance, which can lead to reasonable but concise result. We find that weighted itemset search space is enumerable through exploiting the weighted support-significant framework. By adopting specific technique, duplicate search space can be pruned early with little cost. All the weighted closed itemsets are derived directly while enumerating them without many duplicate candidates generation. Then concise association rules based on weights can be generated. As illustrated in experiments, the proposed method leads to good results and achieves good performance.