Abstract
Association rule mining is an important tool for finding hidden knowledge
in basket datasets. However, a drawback of association rule mining is
that, while some of the mined association rules may contain new knowledge,
many other rules are trivial facts or redundant, and thus are
"uninteresting" to users.
To avoid uninteresting rules, different "interestingness" measures of
mined association rules are proposed. Such measures may be "objective",
i.e., mined rules are ranked by a predefined system, or "subjective", that
users are required to specify whether a rule is interesting. Only the
association rules that are classified "interesting" under selected
interesting measures are presented to users.
In this talk, we will discuss different interestingness measures for
association rules. Besides, we will discuss some potential research
directions on interestingness of association rules.
