A taxonomy-based approach to determining generic interestingness of association rules

Abstract

Items are related to each other either because of the generic category to which they belong or due to their usage contexts. We describe four notions of item-relatedness based on relationships existing between them using a taxonomy. We combine two of them to get a new measure of item relatedness. We compare and contrast this measure with a traditional taxonomy-based similarity measure. Interestingness of an association rule is then inversely proportional to the least-related item pair in the rule.