Multi-View Sentence Ranking for Query-Biased Summarization

Abstract

This paper proposes a novel sentence ranking approach to query-biased summarization where ranking performance can be boosted by encouraging biased information richness in a multi-view framework. To investigate how the final ranking result can benefit from diverse local ranking's combination, the proposed approach firstly constructs two base rankers to rank all the sentences in a document set from two independent but complementary views (i.e. query-dependent view and query-independent view), and then aggregates them into a consensus one, which can overcome each base ranker's local preference and improve global robustness of the overall ranking result. Experimental results on the DUC dataset illustrate the effectiveness of the proposed method.