Semantic Pivoting Model for Effective Event Detection

Event Detection, which aims to identify and classify mentions of event instances from unstructured articles, is an important task in Natural Language Processing (NLP). Existing techniques for detection only use homogeneous one-hot vectors represent the type classes, ignoring fact that semantic meaning types task. Such approach inefficient prone overfitting. In this paper, we propose Semantic Pivoting Model Effective Detection (SPEED), explicitly incorporates prior information during training captures more semantically meaningful correlation between input events. Experimental results show our proposed model achieves state-of-the-art performance outperforms baselines multiple settings without using any external resources.