Research on Relation Extraction Method Based on Multi-channel Convolution and BiLSTM Model

Deep learning methods have achieved good results in relation extraction research and have received widespread attention. However, the existing deep learning methods use a single word vector model, which cannot fully utilize the rich semantic information and syntactic structure in the corpus. The high parameter dimension causes information overload and cannot make full use of context information. Aiming at the problems of the current method, this paper proposes a multichannel relation extraction framework that uses multiple word vector models to map the corpus to form a multi-channel. Feature extraction is performed through the neural network model fused with convolutional neural network, BiLSTM and attention mechanism, and finally completes the relationship extraction task through the classifier. The experimental results on the SemEval 2010 task 8 data set show that this method can not only acquire rich semantic information in the corpus, but also better learn local features and use contextual information. Compared with other methods, the method in this paper achieves competitive performance.