An important area of business data mining is the determination of association rules via the analysis of large databases of transaction records that contain item sets, for example from sales. In this fine paper, the authors propel the research frontier a little more by studying generalized association rules in the presence of taxonomies, like, for example, the hierarchy between printer and impact printer or nonimpact printer.

The problem that emerges when taxonomy information is incorporated into rule mining is how to effectively compute the occurrences of an item set in the transaction database. More than this, a uniform minimum rule support assumption, which is the current usual approach, can hide the discovery of interesting rules, because different parts (levels) of the hierarchy must be treated differently (to allow, that is, differentiations on the minimum support).

The authors successfully attack these problems with two algorithms, and also present convincing numerical experiments that show good linear scale-up behavior. The algorithms, by using the notion of rule-lift, can differentiate an association rule from a mere correlation over multiple minimum support assumptions, avoid combinatorial explosion during the rule search, and discover less-supported but intuitively existing rules in the transaction database.

Data mining researchers and people from business analysis and marketing departments can benefit from the approach and algorithms that the authors present.