Multi-granularity context semantic fusion model for Chinese event detection

Event detection is a key task in the field of information extraction, which is widely used in knowledge mapping, intelligent question answering, reading comprehension and other fields. Event detection model based on deep learning commomly regards event detection as a word classification task, but in Chinese, a language without natural separators, the error of word segmentation will lead to the mismatch between word and event trigger. In addition, the polysemy of Chinese words can make a word ambiguous in different contexts. In this paper, we propose a Chinese event detection model based on multi granularity context semantic fusion. Firstly, the semantic information in different word segmentation results is obtained through the Character-Word fusion gate mechanism to solve the problem of mismatch between Chinese words and event triggers. Then, the Character-Sentence fusion gate is designed to learn the semantic information of the whole sequence, and the self-attention mechanism is used to integrate the contextual semantic information to eliminate the ambiguity of Chinese words. Experiments on ACE2005 dataset show that our method can achieve better experimental results than the current mainstream Chinese event detection model.