Interesting Association Rules in Multiple Taxonomies

In this paper we study association rules in order to understand customer behaviour. We examine the case where many customers may choose from a long list of products. Suppose that several taxonomies for these products are given: the products are grouped in different ways, e.g., by colour, by price, by brand and so on. Then a rule is called interesting if its support, i.e., the number of customers satisfying the rule, deviates substantially from the predictions that are generated through one or more taxonomies. Such a prediction is found by replacing any product in a rule with its parent in the taxonomy at hand, and then estimating the support of the original rule through the support of the parent rules and the conditional probabilities of the “lifted” products. This notion of interestingness is easy to handle and adheres to the intuition.

[1]  Jiawei Han,et al.  Mining Multiple-Level Association Rules in Large Databases , 1999, IEEE Trans. Knowl. Data Eng..

[2]  Heikki Mannila,et al.  Finding interesting rules from large sets of discovered association rules , 1994, CIKM '94.

[3]  Gregory Piatetsky-Shapiro,et al.  Discovery, Analysis, and Presentation of Strong Rules , 1991, Knowledge Discovery in Databases.

[4]  Heikki Mannila,et al.  Fast Discovery of Association Rules , 1996, Advances in Knowledge Discovery and Data Mining.

[5]  Ramakrishnan Srikant,et al.  Mining generalized association rules , 1995, Future Gener. Comput. Syst..

[6]  Roberto J. Bayardo,et al.  Mining the most interesting rules , 1999, KDD '99.

[7]  Dimitrios Gunopulos,et al.  Constraint-Based Rule Mining in Large, Dense Databases , 1999, Proceedings 15th International Conference on Data Engineering (Cat. No.99CB36337).

[8]  Bart Goethals,et al.  A priori versus a posteriori filtering of association rules , 1999, 1999 ACM SIGMOD Workshop on Research Issues in Data Mining and Knowledge Discovery.