A thorough survey and detailed account of methods for association rule mining, this paper outlines a new approach to calculate interestingness by clustering 61 interestingness measures.
The purpose of the proposed research is to explain the specifics and general methods of an interestingness measures specification and to experiment with clustering the measures in order to investigate their similarity. It also examines the basic algorithms that allow this calculation. It adopts the position that the data resources for association rule mining are sets of attribute-value pairs from which the algorithms extract rules that relate the presence of certain feature values with that of others.
The experiments carried out to cluster the interestingness measures use 110 datasets from the life science domain. The distance between the different interestingness measures is computed through a random selection of association rules and consequent ranking of their results.
The paper presents a detailed discussion of the results of the experiment and explains the advantages of the approach, emphasizing the importance of looking closely at the ranking behavior of the clustering.
A very thoroughly written paper with a great deal of technical insight and an exhaustive survey of interestingness measures and other related techniques, it can be used as a valuable resource by scholars and students at the introductory level of association rule calculation, clustering, and ranking evaluation. It is also good reading for those scientists and engineers interested in big data and reasoning.