Extractive and Abstractive Event Summarization over Streaming Web Text / 4002
Chris Kedzie, Kathleen McKeown
During crises, information is critical for responders and victims. When the event is significant, as in the case of hurricane Sandy, the amount of content produced by traditional news outlets, relief organizations, and social media vastly overwhelms those trying to monitor the situation. An emerging task in this space is to monitor an event as it unfolds over time by processing an associated stream of documents to produce a rolling update summary containing the most salient information with respect to the event. In this thesis, we develop two extractive summarization systems for streaming text data. Both systems explicitly predict the salience of input stream text to create a rolling summary. Finally, we discuss our proposed work for combining these systems with an abstractive text generation model.