Share-Inherit: A novel approach for mining frequent patterns
Xiaoyong Lin, Qunxiong Zhu · 2010
Mining frequent patterns has attracted considerable attention in the data mining field. Most of the current studies adopt the pattern growth approach of divide-and-conquer. However, as the mining process is completely split into parts, all relevant algorithms still encounter some performance bottlenecks. In this study, we propose a new data structure, Share-struct, which is derived but obviously different from FP-tree. Then we developed an efficient algorithm, Share-Inherit, for mining all frequent patterns. Based on the Share-struct, Share-Inherit provides a way to share most of the results from the previous mining process instead of separating them distinctively, thereby dramatically reducing the cost of traversing FP-tree and significantly improving the performance of algorithms based on pattern growth. Furthermore, various optimization techniques used in Share-Inherit sufficiently improve the algorithm. The experimental results show that Share-Inherit algorithm not only performs better than the existing algorithms using pattern growth, but also improves them by incorporating Share-Inherit strategy.