F-055 Contrasting Correlations Based on Double-Clique Search

概要

論文の詳細を見る
The studies on contrast set mining and correlation mining have been paid much attention in this decade. In this paper, we consider a problem of contrasting correlations measured by k-way mutual information. As the cost for enumerating item sets and computing their mutual information is high, we introduce a new graph theoretic technique to cut off useless itemsets. In fact, we make two graphs: one graph represents implicitly correlated pairs in one database, and the other shows mediately correlated pairs in another database. By using the graphs with the different types of edges, we experimentally show that the problem of contrasting correlations can be effectively solved.
2010-08-20