Graph-based Methods for Multidocument Summarization: Exploring Relationship Maps, Complex Networks and Discourse Information

Abstract. In this work we investigate the use of graphs for multi-document summarization. We adapt the traditional Relationship Map approach to the multi-document scenario and, in a hybrid approach, we consider adding CST (Cross-document Structure Theory) relations to this adapted model. We also investigate some measures derived from graphs and complex networks for sentence selection. We show that the superficial graph-based methods are promising for the task. More importantly, some of them perform almost as good as a deep approach