Cohesion-based Sentence Ordering for Multi-document Summarization

Abstract

Sentence ordering is a difficult but an important task for multi-document summarization. A wrong order of sentences not only makes it difficult to be understood but also convey entirely different ideas to the reader. In this paper, we describe two naive ordering algorithms and give the reasons why they do not perform well. We propose a new sentence ordering algorithm, which combines constraints from majority order of themes and cohesion. This algorithm ensures: if the cohesion between two themes is bigger than the threshold value, then the sentences of the two themes would be placed to the adjacent location in the final summary. Experimental results show that summary generated by our improved sentence ordering algorithm is more coherent and more readable than the two naive algorithms.