Title: Coherent Extractive Multi-Document Summarization
Speaker: Gagan Bansal
Abstract: The popularity of Multi-Document Summarization continues to rise in the
era of information overload. The task involves generating a quality
summary of a collection of related documents. The domain of application
is huge, from summarizing new articles to automatically generating
Wikipedia articles. I present two systems that are aimed at solving this
complex problem for summarizing news articles. The first is coherent
summarizer for short document collections and the latter produces
hierarchical summaries for large collections. Both the systems use a
joint model of sentence selection and reordering which results in more
coherent summaries. Manual experiments done on a crowdsourcing platform
(Amazon Mechanical Turk) show that summaries generated by these systems
are prefered more by the users than the previous state-of-the-art.
Relevant papers:
NAACL'13
and ACL'14