Multi-document summarization using weighted similarity between topic and clustering-based non-negative semantic feature

Abstract

This paper presents a new multi-document summarization method using weighted similarity between topic and non-negative semantic features to extract meaningful sentences relevant to a given topic. The proposed method decomposes a sentence into the linear combination of sparse non-negative semantic features so that it can represent a sentence as the sum of a few semantic features that are comprehensible intuitively. It can avoid extracting the sentences whose similarities with topic are high but are meaningless by using the weighted similarity measure between the topic and the semantic features. Clustering sentences remove noises so that it can avoid the biased semantics of the documents to be reflected in summaries. Besides, it can enhance the coherence of document summaries by arranging extracted sentences in the order of their rank. The experimental results using DUC data show that the proposed method achieves better performance than the other methods. © Springer-Verlag Berlin Heidelberg 2007.