Multi-document extractive summarization using semantic graph

The automatic texts summarization consists in synthesizing in a short text the most relevant information contained in text documents, and allows to reduce the generated problems by the information overload. In this paper, an unsupervised method for extractive multi-document summarization is presented. In this proposal, the conceptualization and underlying semantics structure of the textual content is represented in a semantic graph using WordNet, and a concept clustering algorithm is applied to identifying the topics of the documents set, with which the relevance of the sentences is evaluated to build the summary. The method was evaluated with texts corpus from MultiLing 2015, and ROUGE metrics were used to measure the quality of the generated summaries. The obtained results were compared with those other participant systems in MultiLing 2015, evidencing improves in most of the cases.