Influence and conditional influence-new interestingness measures in association rule mining

Discusses the issues of interestingness in association rule mining. First, a rule is possibly redundant or misleading even if it possesses high degrees of confidence and support. Second, association rules do not reflect the effect of negatively influential facts. Such problems are related to confidence deviation. In the paper, therefore, two new measures of interestingness, namely influence and conditional influence, are introduced to represent the effect of the antecedent on the consequent. Furthermore, the mining algorithms are extended accordingly such that certain redundant rules can be eliminated and negatively influential rules may be discovered.