A Level-wise Clustering Algorithm on Structured Documents

Abstract

Document clustering is the process of applying clustering technique for document management [4][5]. Similar documents are grouped together so that both managing and searching the documents is efficient. However, since traditional document clustering algorithms do not take the structure information of documents into consideration, the clustering results can not reflect the characteristics of the documents fully. As the result, we represent each document as a tree structure and propose a level-wise clustering algorithm to solve this issue. The clustering process applies the level property of the tree and run level by level by the concept generalization operation. In order to store the clustering results and search interesting clusters efficiently, a multistage graph called Level-wise Document Clustering Graph (LDC-Graph) is proposed. Based on LDC-Graph, three search strategies are provided to meet the different requirements for uses. Finally, the experimental results show that the similarity search is efficient and the accuracy of the search is acceptable