An AMR-based extractive summarization method for cohesive summaries

Abstract

The main goal of automatic text summarization is condensing the original text into a shorter version, preserving the information content and general meaning. The extractive summarization, one of the main approaches for automatic text summarization, consists to select the most relevant sentences of a document, and generate a summary. This paper proposes a new mono-document extractive summarization method using a semantic representation of the sentence of a document expressed in AMR (Abstract Meaning Representation). In this method, AMR semantic representation is used to capture the most important concepts of each sentence (in core semantic terms), and a concept-based Integer Linear Programming (ILP) approach to select the most informative sentences improving both relevance and text cohesion of the summary. Two datasets proposed by DUC (2001 and 2002) were used to evaluate the effectiveness of our method on extrative summarirazion and commparing it with other state-of-the-art summary systems.