Tightness: A novel heuristic and a clustering mechanism to improve the interpretation of association rules

In this paper we present a clustering-based approach to mitigate the ‘rule immensity’ and the resulting ‘understandability’ problem in association rule (AR) mining. Clustering ‘similar’ rules facilitates exploration of connections among rules and the discovery of underlying structures. We first introduce the notion of ‘tightness’ of an AR. It reveals the strength of binding between various items present in an AR. We elaborate on its usefulness in the retail market-basket context and develop a distance-function on the basis of ‘tightness.’ Usage of this distance function is exemplified by clustering a small artificial set of ARs with the help of average-linkage method. Clusters thus obtained are compared with those obtained by running a standard method (from recent data mining literature) on the same data set.