Relational Text Mining and Visualization

Discovering hidden patterns in distributed heterogeneous textual databases and unstructured data is a new challenge in data mining. Traditional data mining often assumes that preprocessing is already done -homogeneous data are available on the needed level. For distributed heterogeneous textual data this is not the case. Complex relations between items/entities (e.g., relations between people in a fraud detection task) should be discovered and generalized. This paper offers a new hierarchical relational clustering method (Φ-method) based on the Φ-equivalence concept. The method permits to process: (i) incomplete relations: (ii) relations without converting them to attributes of individual entities, and (iii) relations presented in distributed heterogeneous databases using XML tags. Clustering produced by the method is invariant, has a clear meaning and natural visualization.