Uma Comparação Sistemática de Diferentes Abordagens para a Sumarização Automática Extrativa de Textos em Português

Abstract

Automatic document summarization is the task of automatically generating condensed versions of source texts, presenting itself as one of the fundamental problems in the areas of Information Retrieval and Natural Language Processing. In this paper, different extractive approaches are compared in the task of summarizing individual documents corresponding to journalistic texts written in Portuguese. Through the use of the ROUGE package for measuring the quality of the produced summaries, we report on results for two different experimental domains, involving (i) the generation of headlines for news articles written in European Portuguese, and (ii) the generation of summaries for news articles written in Brazilian Portuguese. The results demonstrate that methods based on the selection of the first sentences have the best results  when building extractive news headlines in terms of several ROUGE metrics. Regarding the generation of summaries with more than one sentence, the method that achieved the best results was the LSA Squared algorithm, for the various ROUGE metrics.