Unsupervised document summarization using clusters of dependency graph nodes

Abstract

In this paper, we investigate the problem of extractive single document summarization. We propose an unsupervised summarization method that is based on extracting and scoring keywords in a document and using them to find the sentences that best represent its content. Keywords are extracted and scored using clustering and dependency graphs of sentences. We test our method using different corpora including news, events and email corpora. We evaluate our method in the context of news summarization and email summarization tasks and compare the results with previously published ones.