Frequent pattern-growth approach for document organization

Abstract

In this paper, we propose a document clustering mechanism that depends on the appearance of frequent senses in the documents rather than on the co-occurrence of frequent keywords. Instead of representing each document as a collection of keywords, we use a document-graph which reflects a conceptual hierarchy of keywords related to that document. We incorporate a graph mining approach with one of the well-known association rule mining procedures, FP-growth, to discover the frequent subgraphs among the document-graphs. The similarity of the documents is measured in terms of the number of frequent subgraphs appearing in the corresponding document-graphs. We believe that our novel approach allows us to cluster the documents based more on their senses rather than the actual keywords.