An unsupervised approach to generating generic summaries of documents

摘要: We model document summarization as a quadratic Boolean programming problem.We create modified differential evolution to solve the optimization problem.Experimental study shows that improves results. present an optimization-based unsupervised approach automatic summarization. In proposed approach, text is modeled problem. This generally attempts optimize three properties, namely, (1) relevance: summary should contain informative textual units are relevant user; (2) redundancy: summaries not multiple convey same information; and (3) length: bounded in length. The this paper applicable both tasks: single- multi-document tasks, documents split into sentences preprocessing. select some salient from document(s) generate summary. Finally, generated by threading all selected order they appear original document(s). implemented our on task. When comparing methods several existing open DUC2005 DUC2007 data sets, we found method results significantly. because, first, when extracting sentences, only focuses relevance scores of whole sentence collection, but also topic representative sentences. Second, generating summary, deals with problem repetition information. were evaluated using ROUGE-1, ROUGE-2 ROUGE-SU4 metrics. paper, demonstrate result depends similarity measure. Results experiment showed combination symmetric asymmetric measures yields better than their use separately.