Multi-Strategy Approach to Mining Interesting Rules
Cheng Ji · Chinese Journal of Computers · 2000
A large set of rules can be discovered from large database by using the data mining technologies, but most of them are of no interesting to the user. How to filter the interesting rules is the crucial step within the knowledge discovery in database. The methods to calculate the subjective interestingness and to measure the importance of rules by using of the domain knowledge are studied. A new objective interestingness function, which can measure the novelty and simplicity of rules, is given. A multi strategy approach, which combines with background knowledge, for selecting interesting rules is proposed in the paper. The process of filtering interesting rules includes the several sub processes: deleting the redundant rules; grouping the rules into sub groups; clustering each sub group into classes, and selecting the most interesting rule from each class; finally combining them into the set of interesting rules. Some concepts, such as the importance of attribute, the interestingness of rules, the distance between rules etc, are proposed in the paper. There is an example in the paper, which applying the multi strategy approach, for illustrating the process to select the interesting rules that discovered from CPICDB (chinese parasite infection census database). The example demonstrated the algorithm represented in the paper is practicable and effective.