Since their introduction in 1993, association rules have been successfully applied to the description and summarization of discovered relations between attributes in a large collection of objects. However, most of the research works in this area have focused on mining simple objects, usually represented as a set of binary variables. The proposed work presents a framework for mining complex objects, whose attributes can be of any data type (single and multi-valued). The mining process is guided by the semantics associated to each object feature, which is stated by users by providing both a comparison criterion and a similarity function over the object subdescriptions. Experimental results show the usefulness of the proposal.
展开▼
