Efficient Skyline Frequent-Utility Itemset Mining Algorithm on Massive Data (Extended abstract)

Itemset mining is a crucial technology for extracting interesting patterns that meet predefined thresholds from transaction databases, such as frequent itemset mining (FIM) and high-utility itemset mining (HUIM). Despite their significance, few studies have explored the simultaneous consideration of both support and utility. This gap arises from the theoretical complexity of integrating these dimensions and the practical difficulty of setting appropriate thresholds for both. To overcome this limitation, we introduce Skyline frequent-utility itemset mining (SFUIM), a method that examines frequent and high-utility itemsets without requiring predefined thresholds. Nevertheless, SFUIM faces significant challenges due to its expansive search space and intensive computational requirements. In this paper, we propose a PSI algorithm and its enhanced version PSI*, which confines calculations to specific partitions by prefix-based partitioning. Experiments demonstrate that PSI^* outperforms state-of-the-art methods, especially on large-scale datasets.