Incorporating Background Knowledge for Subjective Rule Evaluation

Association rule mining is the task of finding interesting relationships hidden in large transaction databases. Despite the significant progress made in this field, one of the fundamental challenges that remain unresolved is the rule evaluation problem. Most notably, it is difficult to discriminate rules that are known to the domain experts from those that are unexpected. In this paper, we propose a framework called MIR that incorporates background knowledge acquired from an authoritative source into the rule evaluation task. We illustrate the advantages of using the framework in the medical informatics domain, where the rules are extracted from an electronic medical records (EMR) database while the domain knowledge is automatically acquired from the MEDLINE repository of biomedical citations.