Utility mining is one of the most important and contemporary research topics in data mining. It addresses the limitation of frequent itemset mining by considering nonbinary frequency values of items in transactions and different profit values for each item. Here, the term “utility” refers to the importance or profitability of items to the users. In this paper, the authors focus on high-utility itemset (HUI) mining that aims to find all itemsets with a utility value that is no smaller than a user-specified threshold. Mining HUIs is extremely important in the context of different business decisions such as revising revenue, adjusting inventory, and determining purchase orders, because it helps extract highly valuable information from a database by measuring how useful the item is.

Since a high-utility itemset may consist of some low-utility subsets, many approaches in the literature use a costly level-wise candidate generation and test strategy. The authors in this paper have proposed a concurrent algorithm, called CHUI-Mine, that avoids this costly task. This is accomplished along two different dimensions, as follows. First, the authors have proposed a new data structure, called CHUI-Tree, to maintain the information of HUIs. This data structure in fact “exploits a pattern growth approach to avoid the problem of the level-wise candidate generation-and-test strategy.” Second, the authors have introduced two concurrent processes to generate the potential high-utility itemsets: the first process constructs and subsequently prunes the tree dynamically and then places the conditional trees into a buffer, and the second one reads the conditional tree from the buffer and mines HUIs using a pattern growth approach.

Furthermore, the authors have theoretically proved that the result of concurrent HUI mining is in fact the same as mining the whole tree structure, which is the main rationale behind employing the dynamic tree pruning strategy. In the sequel, they have conducted extensive experiments on both synthetic and real datasets to show the efficiency and scalability of their strategy.