Weighted graph model based sentence clustering and ranking for document summarization

This paper proposes a sentence ranking and clustering based summarization method that extracts essential sentences from a document. To discover central sentences, a weighted undirected graph that takes sentence similarities and the discourse relationship between sentences as the weights of edges is constructed for the given document. A graph-ranking algorithm is implemented to calculate the scores of sentences. We also build a matrix for the document, and an algorithm based on Sparse Non-negative Matrix Factorization is introduced to cluster the sentences in the document. High ranked sentences of each cluster are selected to comprise the summarization of the document. The experimental results on the Document Understanding Conference (DUC) 2001 data set demonstrate the effectiveness of the document summarization algorithm.