Test of Complementarity on Sentence Extraction Methods

In this work three approaches to sentence extraction methods are analyzed. We try to find if the used methods show some complementary features. In order to accomplish this goal, the methods of sentence extraction were applied and combined, analyzing the results in the theoretical framework that we propose. We test three approaches: graph-based, keyword-based and representation-based. The methods were tested using the text collection DUC 2002, obtaining the best performance for a very simple method based on representation.index. Even though no complementary methods were found, the results allow to identify some relevant features of the methods.