CGM: A biomedical text categorization approach using concept graph mining

Text Categorization is used to organize and manage biomedical text databases that are growing at an exponential rate. Feature representations for documents are a crucial factor for the performance of text categorization. Most of the successful existing techniques use a vector representation based on key entities extracted from the text. In this paper we investigate a new direction where we represent a document as a graph. In this representation we identify high level concepts and build a rich graph structure that contains additional concepts and relationships. We then use graph kernel techniques to perform text categorization. The results show a significant improvement in accuracy when compared to categorization based on only the extracted concepts.