Graph-based summarization without redundancy

Abstract

In this paper we present a new text summarization method based on graph to generate concise summaries for highly redundant documents. By mapping the source documents into a textual graph, we turn the summarization into a new problem of finding the key paths composed by essential information. Unlike the extraction of original sentences, our method regenerates sentences by word nodes in the textual graph. In order to avoid the selection of unreasonable paths with grammatical or semantical problems, some syntax rules are defined to guide the path selecting process, and we merge the common paths shared by different sentences to reduce content redundancy. Evaluation results show that our method can get concise summaries with a higher content accuracy. © 2014 Springer International Publishing Switzerland.