Biomedical events,as an important part of biomedical text mining,play an important role in biomedical re-search and disease prevention.Trigger identification is the key and prerequisite step of biomedical event extraction,which aims to extract the key words describing event types.Traditional trigger identification methods rely too much on natural lan-guage processing tools in the process of feature extraction,consuming a lot of manual cost.In addition,due to the particular-ity of biomedical literature—there are many long text sentences,the problem of long-distance dependence is obvious.To solve these problems,we propose a hybrid structure,which is composed of residual convolution neural network and bidirec-tional long short term memory,hybrid neural network and multi head attention mechanism.The proposed model uses residu-al convolution neural network to extract vocabulary-level features and bidirectional long short term memory to obtain con-textual semantic information.Furthermore,spatial domain sliding windows divide long sentences into equal-length short sentences without damaging context information,which can avoid long-distance dependency without destroying the context information.The experimental results show that our method outperforms the state-of-the-art methods on the commonly used multi-level event extraction(MLEE)corpus,achieving 81.15%F-score.
维普
万方数据