Semi-extractive multi-document summarization via submodular functions

Abstract

In this paper, we propose an Extractive Maximum Coverage KnaPsack (MCKP) based model for query-based multi document summarization which integrates three monotone and submodular measures to detect importance of a sentence including Coverage, Relevance, and Compression. We apply an efficient scalable greedy algorithm to generate a summary which has a near optimal solution when its scoring functions are monotone nondecreasing and submodular. We use DUC 2007 dataset to evaluate our proposed method and the result shows improvement over two closely related works.