Extractive single-document summarization based on global-best harmony search and a greedy local optimizer

Abstract

Due to the great amount of documents available on the Web, end users need to be able to access information in summary form–keeping the most important information in the document. The methods employed for automatic text summarization generally allocate a score to each sentence in the document, taking into account certain features. The most relevant sentences are then selected, according to the score obtained for each sentence. In this paper, the extractive single document summarization task is treated as a binary optimization problem and, based on the Global-best Harmony Search metaheuristic and a greedy local search procedure, a new algorithm called ESDS-GHS-GLO is proposed. This algorithm optimizes an objective function, which is a lineal normalized combination of the position of the sentence in the document, sentence length, and coverage of the selected sentences in the summary. The proposed method was compared with the state of the art methods MA-SingleDocSum, DE, FEOM, UnifiedRank, Net-Sum, QCS, CRF, SVM, and Manifold Ranking, using ROUGE measures on the DUC2001 and DUC2002 datasets. The results showed that ESDS-GHS-GLO outperforms most of the state-of-the-art methods except MA-SingleDocSum. ESDS-GHS-GLO obtains promissory results using a fitness function less complex than MA-SingleDocSum, therefore requiring less execution time.