Research of Occupancy-Based Skyline Pattern Mining
Kai Zhang, Kun Hu · 2021
The DOFRA algorithm uses user-specified minimum support threshold and occupancy threshold to find frequent and complete patterns. This process requires constant modification of threshold to adapt to different datasets. From a macro point of view, there is a negative correlation between the pattern support and the pattern length (occupancy), and user-specified minimum threshold mining has a large deviation to the result. In this article, by designing an efficient Pattern-extension-list (PEL) structure to replace the occupancy upper bound calculation in the DOFRA algorithm, a skyline mining algorithm based on the occupancy upper bound calculation (SOPM-occubound) is proposed. Two types of Skyline algorithms based on the PEL structure, SOPM (candidate set) and SOPM-occumax (without candidate itemsets), are used to mine occupancy-based skyline patterns (OBSPs). The advantage of SOPM is that the unreasonable setting of the threshold results in the problem of extremely different data excavated, and it can effectively find the most representative modes in the two dimensions of support and occupancy, and dominate other modes. Finally, the experimental results on several real and simulated datasets show that SOPM algorithm based on PEL structure is superior to SOPM-occubound in terms of running time and memory usage, and SOPM-occumax algorithm does not generate candidate item sets, thus it is the most efficient.