Towards CST-Enhanced Summarization

摘要

In this paper, we propose to enhance the process of automatic extractive multi-document text summarization by taking into account cross-document structural relationships as posited in Cross-document Structure Theory (CST). An arbitrary multi-document extract can be CST-enhanced by replacing low-salience sentences with other sentences that increase the total number of CST relationships included in the summary. We show that CST-enhanced summaries outperform their unmodified counterparts using the relative utility evaluation metric. We also show that the effect of a CST relationship on an extract depends on its type.