Sentence Ordering for Coherent Multi-document Summary Generation

摘要

Web queries often give rise to a lot of documents and the user is overwhelmed by the information. Query-specific extractive summarization of a selected set of retrieved documents helps the user to get a gist of the information. The current extractive summary generation systems focus on extracting query-relevant sentences from the documents. However, the selected sentences are presented either in the order in which the documents were considered or in the order in which they were selected. This approach does not guarantee a coherent summary. In this paper, we propose incremental integrated graph to represent the sentences in a collection of documents. Sentences from the documents are merged into a master sequence to improve coherence and flow. The same ordering is used for sequencing the sentences in the extracted summary. User evaluations indicate that the proposed technique markedly improves the user satisfaction with regard to coherence in the summary.