TOPIC: Top-k High-Utility Itemset Discovering

Abstract

Utility-driven itemset mining is widely applied in many real-world scenarios. However, most algorithms do not work for itemsets with negative utilities. Several efficient algorithms for high-utility itemset (HUI) mining with negative utilities have been proposed. These algorithms can find complete HUIs with or without negative utilities. However, the major problem with these algorithms is how to select an appropriate minimum utility (minUtil) threshold. To address this issue, some efficient algorithms for extracting top-k HUIs have been proposed, where parameter k is the quantity of HUIs to be discovered. However, all of these algorithms can solve only one part of the above problem. In this paper, we present a method for TOP-k high-utility Itemset disCovering (TOPIC) with positive and negative utility values, which utilizes the advantages of the above algorithms. TOPIC adopts transaction merging and database projection techniques to reduce the database scanning cost, and utilizes minUtil threshold raising strategies. It also uses an array-based utility technique, which calculates the utility of itemsets and upper bounds in linear time. We conducted extensive experiments on several real and synthetic datasets, and the results showed that TOPIC outperforms state-of-the-art algorithm in terms of runtime, memory costs, and scalability.

Towards Target High-Utility Itemsets

“…With these efficient techniques, EFIM is in general two to three orders of magnitudes faster than previous work. TOPIC [21] can handle itemsets with negative utility and find top-k patterns to cope with the problem of setting a suitable minUtil value. Up to now, there are many other advanced studies of HUIM, such as [42]- [44].…”

“…And if oriOrder equals patOrder, it means that items x.item and T [index] can be matched, and then the pointer index is moved back one position (lines [5][6][7][8][9][10][11]. For the items to be matched, if patOrder < oriOrder, the traversed itemset is a mismatch with T according to the pruning strategy, and the current recursion terminates (lines [12][13][14] [15][16][17][18][19][20][21][22][23][24]. Algorithm 3 shows how to construct a utility-list.…”

“…For the second issue, many researchers have designed improved utility data mining algorithms. Many utility mining studies [19], [21]- [24] apply the TWU to facilitate the mining process and use efficient data structures and pruning strategies to improve the performance of algorithms. There are already several algorithms in the high-utility itemset mining (HUIM) literature, such as list-based (i.e., HUI-Miner [23], FHM [25], FHN [22], and HUOPM [26]), tree-based (i.e., IHUP [27], UP-Growth [19] and MU-Growth [28]), and projection-based (i.e., EFIM [24]) algorithms.…”

FUIM: Fuzzy Utility Itemset Mining

“…If TWU of an itemset is less than the user-specified minUtil, it would be an unpromising itemset; otherwise, it maybe a real HUI which needs to be confirmed. After that, there are many studies on improving performance of mining HUIs with different data structures, such as list-based algorithms (e.g., HUI-Miner [11] and FHM [12]), tree-based algorithms (e.g., UP-Growth [13] and MU-Growth [14]) and projection-based algorithms (e.g., EFIM [15], and TOPIC [16]).…”