2011 Presentations (Communicative Events)

Identifying Event Descriptions using Co-training with Online News Summaries

McKeown, Kathleen; Thadani, Kapil; Wang, William

Systems that distill information about events from large corpora generally extract sentences that are relevant to a short event query. We present a novel co-training strategy for this task that employs a multidocument news summary corpus featuring 2.5 million unlabeled sentences, thus obviating the need for extensive manual annotation. Our experiments indicate that this technique significantly outperforms standard classification approaches with linear feature combination on this task. An analysis of our approach under various settings reveals how classifier and parameter choice can be used to control runtime overhead while contributing to an absolute increase of 22% in recall.