Information Technology Reference
In-Depth Information
Furthermore, most business users evaluate the usefulness of a rule by examining
the practical significance of consequent(s), and the use of interestingness measure
(such as support and confidence) is sometimes of secondary importance. This is
logical because a rule with a significant consequent (e.g., the occurrence of a life-
threatening disease) is more important than a rule with trivial consequent (e.g., the
occurrence of minor cuts and bruises), even though the former rule may have lower
interestingness values than the latter one. As far as we know, most previous studies in
rule summarization have not taken this point into consideration.
The above observation motivates us to develop a new method called Consequent-
based Association Rules Summarization (CARS). Previous scientific approaches for
rule summarization, though theoretically interesting, are practically limited in
reducing the number of rules. CARS is different. It is an applied approach designed
for three pragmatic reasons. Firstly, CARS summarizes a large number of association
rules into a few rule summaries, where each rule summary has the same
consequent(s). This allows end-users to evaluate a set of rules based on consequent
importance, an important angle for evaluating rules. Secondly, each rule summary
ranks the antecedents that determine the consequent. Such information is useful for
assessing the plausibility of the rules against business knowledge. Thirdly, each rule
summary provides ranges of interestingness values to allow user to refine the search
for rules of particular consequent(s). This is important because rules with different
consequents may have different ranges of interestingness values. So CARS effectively
compartmentalizes the search for rules of certain consequent(s).
The rest of this paper is organized as follows. Section 2 provides some preliminary
information about association rule mining and discusses issues in association rule
mining and review existing solutions to the issues. Section 3 describes the proposed
method for summarizing association rules. Section 4 presents the results and
demonstrates an application of CARS. Section 5 concludes this paper.
2
Related Work
This section first provides a brief introduction to association rule mining. Then, it
discusses some common issues and solutions related to association mining solution.
2.1
The Association Rule Mining Process
Figure 1 shows the process of association rule mining. First, data is prepared and then
read in by the association rule mining algorithm. Second, the algorithm parameter
settings (such as the support and confidence thresholds) are defined and rules are
generated from the input data. Third, the rules are (sometimes) sent to a post-
processing stage to improve the presentation of rule mining results. This work focuses
on improving the final stage.
2.2
Two Issues of Association Rule Mining
A longstanding issue of association rule mining is that the approach tends to generate
excessively large number of rules. A common way to address this issue is to reduce
Search WWH ::




Custom Search