Abstract:

Interestingness measures are intended for selecting patterns according to their potential interest to the user in association rules. The FPgrowth algorithm based on the FPtree structure is an efficient algorithm for mining association rules. This algorithm is not quite effective in the process of mining potentially valuable lowsupport patterns. To solve this problem, a novel type of interestingness measure called betweenitems positive correlation interestingness measure is presented. This measure has a good autimonotone, and the presence of an item in one transaction increases the presence of every other item in the same pattern. This paper also proposes an improved FP mining algorithm which creates a compact FP structure by the node linked list and uses a nonrecursive function to decrease the overhead of creating an extra data structure at each mining step. More importantly, this algorithm exploits an efficient pruning strategy which uses the interestingness measure to filter the nonpositive correlated long model and invalid itemsets. The range of the support threshold is expanded. The experimental results indicate the given algorithm is efficient and feasible.

Key words: association rules;interestingness measure;betweenitems positive correlation;pruning