Sentence Ordering based on Cluster Adjacency in Multi-Document Summarization

In this paper, we propose a cluster-adjacency based method to order sentences for multi-document summarization tasks. Given a group of sentences to be organized into a summary, each sentence was mapped to a theme in source documents by a semi-supervised classification method, and adjacency of pairs of sentences is learned from source documents based on adjacency of clusters they belong to. Then the ordering of the summary sentences can be derived with the first sentence determined. Experiments and evaluations on DUC04 data show that this method gets better performance than other existing sentence ordering methods.