| Summary: | 碩士 === 國立中央大學 === 資訊工程研究所 === 95 === Some recent works have showed the improved approaches which are certainly better than original Apriori-like algorithms for mining association rules. Mining frequent patterns (itemsets) plays an important role of discovering association rules. In the past, Apriori-like methods were adopted to mine frequent itemsets. But these approaches are inefficient to perform a mining task. This is a result from its repeatedly scans of database and iteratively checking a large set of candidates by pattern matching. A compact structure, called FP-tree, was developed to improve the disadvantages of Apriori-like algorithms. By FP-growth approach, proposed by J. Han, we can facilitate mining frequent itemsets. Although FP-growth is a relatively more efficient approach for mining frequent itemsets, the results deduced by FP-growth may be too detailed to satisfy managers or policymakers. We proposed that lower level frequent itemsets and those compressed data within a FP-tree can be generalized furthermore for mining higher level frequent itemsets. Our basic idea is employing the properties and structure of FP-tree according to an existed conceptual hierarchy on mined items. We then provide efficient evolution algorithms to modify the original FP-tree to a higher level FP-tree. In our approaches, the transformed FP-tree still retains the properties of primitive FP-tree. By these novel approaches, we can effectively achieve the goals of transforming from a lower level FP-tree to a higher one, providing more generalized information to managers. Our experimental results also show the effectiveness of the proposed methods.
|
|---|
