Clustering and refinement of hierarchical concept from categorical databases based on rough sets

Discovering knowledge from large databases is a challenge in many applications. The implicit meanings of knowledge can be repressed by different knowledge representations. A concept hierarchy is a concise and general form of knowledge representation. Hierarchical concept description can organize relationships of data and express knowledge embedded in databases explicitly. We propose a new scheme based on rough sets to cluster and refine the concept hierarchy automatically for a given data set with nominal attributes. The proposed scheme consists of two algorithms: the concept clustering algorithm and the concept refinement algorithm. The experimental results show that the concept hierarchy mined by the proposed scheme contains meaningful concept in comparison with the previous approaches. The analyses of the algorithms also show that the proposed scheme is efficient and scaleable for large databases. It can also be extended to mining meaningful rules from databases.