Preceding Document Clustering by Graph Mining Based Maximal Frequent Termsets Preservation

This paper presents an approach to cluster documents. It introduces a novel graph mining based algorithm to find frequent termsets present in a document set. The document set is initially mapped onto a bipartite graph. Based on the results of our algorithm, the document set is modified to reduce its dimensionality. Then, Bisecting K-means algorithm is executed over the modified document set to obtain a set of very meaningful clusters. It has been shown that the proposed approach, Clustering preceded by Graph Mining based Maximal Frequent Termsets Preservation (CGFTP), produces better quality clusters than produced by some classical document clustering algorithm(s). It has also been shown that the produced clusters are easily interpretable. The quality of clusters has been measured in terms of their F-measure.