Improving event extraction using online learning strategy

The traditional mention-oriented model for event extraction cannot capture information of multiple event types in a sentence. To deal with this problem, we present a novel intensive model for event extraction task. Firstly, the model can filter out non-event sentences automatically by introducing some meaningful language features. Then, the model adopts an online learning strategy for event type ranking, which provides an alternative view of event type identification as multi-label classification so that we can achieve a predicted set of relevant event types for each sentence. The experimental evaluation verified that the model can improve the performance of the event extraction task.