Research On Joint Extraction Method Of Biomedical Events Based On Hybrid Neural Network

In recent years,New Coronavirus（COVID-19）epidemic is spreading all over the world.The combination of big data and AI technology has greatly controlled the epidemic through rapid tracking of cases.At the same time,in order to make hospitals and relevant medical research institutes better study and consult the relevant literature of medical knowledge,information extraction technology has received unprecedented attention.As an important branch of information extraction,event extraction has attracted extensive attention of researchers in the field of natural language processing.To ensure the accuracy of recognition,the challenges of biomedical event extraction include:（1）The problem of long-distance dependence caused by the particularity of biomedical text;（2）Due to the different complexity of events,cascading errors are caused.In order to solve the above problems,a hybrid neural network combining spatial sliding window and attention mechanism is proposed to identify biomedical event triggers,and then a hybrid neural network combining dynamic path planning strategy and attention mechanism method is proposed to jointly extract biomedical events.On the premise of ensuring the robustness of the model,the F-value of the two tasks has also been significantly improved.The main work of this thsis is as follows:（1）Biomedical events are an important part of biomedical text mining.Traditional biomedical event extraction methods rely too much on natural language processing tools in the process of feature extraction,resulting in labor cost.In addition,due to the particularity of biomedical literature-there are many long text sentences,which may lead to the problem of long-distance dependence.In order to solve these problems,this thesis proposes a hybrid structure SWACG,which is composed of Re CNN-Bi GRU（Residual Convolution Neural Network and Bidirectional Gate Recurrent Unit）hybrid neural network and MUH-attention mechanism.The model uses Re CNN to extract word level features and Bi GRU to extract contextual semantic information.In addition,the spatial domain sliding window divides long sentences into equal length short sentences without destroying the context information,so as to avoid long-distance dependence.The experimental results show that the method proposed in this thesis has achieved good performance on commonly used biomedical event extraction corpus MLEE（Multi level extraction）.（2）Biomedical event detection is a pivotal information extraction task in molecular biology and biomedical research,which provides inspiration for the medical search,disease prevention,and new drug development.The existing methods usually detect simple biomedical events and complex events with the same model,and the performance of the complex biomedical event extraction is relatively low.In this thesis,we build different neural networks for simple and complex events respectively.To avoid redundant information,we design dynamic path planning strategy for argument detection.In addition,the proposed model reduces cascading errors by joint extraction.Experimental results demonstrate our approach achieves good performance on the biomedical benchmark MLEE dataset and outperforms the recent state-of-the-art methods.