Mining with Rarity for Web Intelligence

Abstract

Mining with rarity makes sense to take advantage of data mining for Web intelligence. In some scenarios, the rare patterns are meaningful in data intelligent systems. Interesting pattern discovery plays an important role in real-world applications. In this field, a great deal of work has been done. In general, a high-utility pattern may include frequent items and also rare items. Rare pattern discovery emerges gradually and helps policy-makers making related marketing strategies. However, the existing Apriori-like methods for discovering high-utility rare itemsets (HURIs) are not efficient. In this paper, we address the problem of mining with rarity and propose an efficient algorithm, named HURI-Miner, which uses the data structure called revised utility-list to find HURIs from a transaction database. Furthermore, we utilize several powerful pruning strategies to prune the search space and save the computational complexity. In the process of rare pattern mining, the HURIs are directly generated without the generate-and-test method. Finally, a series of experimental results show that this proposed method has superior effectiveness and efficiency.