Sentence extraction using time features in multi-document summarization

Abstract

In multi-document summarization (MDS), especially for time-dependent documents, humans tend to select sentences in time sequence. Based on this insight, we use time features to separate documents and assign scores to sentences to determine the most important sentences. We implemented and compared two different systems, one using time features and the other not. In the evaluation of 29 news article document sets, our test method using time features turned out to be more effective and precise than the control system. © Springer-Verlag Berlin Heidelberg 2005.