College of Digital Technology and Engineering, Ningbo University of Finance and Economics, Ningbo, Zhejiang, China.
PLoS One. 2021 Mar 12;16(3):e0248349. doi: 10.1371/journal.pone.0248349. eCollection 2021.
In recent years, high utility itemsets (HUIs) mining has been an active research topic in data mining. In this study, we propose two efficient pattern-growth based HUI mining algorithms, called High Utility Itemset based on Length and Tail-Node tree (HUIL-TN) and High Utility Itemset based on Tail-Node tree (HUI-TN). These two algorithms avoid the time-consuming candidate generation stage and the need of scanning the original dataset multiple times for exact utility values. A novel tree structure, named tail-node tree (TN-tree) is proposed as a key element of our algorithms to maintain complete utililty-information of existing itemsets of a dataset. The performance of HUIL-TN and HUI-TN was evaluated against state-of-the-art reference methods on various datasets. Experimental results showed that our algorithms exceed or close to the best performance on all datasets in terms of running time, while other algorithms can only excel in certain types of dataset. Scalability tests were also performed and our algorithms obtained the flattest curves among all competitors.
近年来，高实用项目集（HUI）挖掘已成为数据挖掘领域的一个活跃研究课题。在这项研究中，我们提出了两种基于模式增长的高效 HUI 挖掘算法，分别称为基于长度和尾节点树的高实用项目集（HUIL-TN）和基于尾节点树的高实用项目集（HUI-TN）。这两种算法避免了耗时的候选生成阶段，以及多次扫描原始数据集以获取精确效用值的需求。我们提出了一种新的树结构，称为尾节点树（TN-tree），作为我们算法的关键元素，用于维护数据集现有项目集的完整效用信息。HUIL-TN 和 HUI-TN 的性能在各种数据集上与最先进的参考方法进行了评估。实验结果表明，在运行时间方面，我们的算法在所有数据集上都超过或接近最佳性能，而其他算法只能在某些类型的数据集上表现出色。还进行了可扩展性测试，我们的算法在所有竞争对手中获得了最平坦的曲线。
