Fusion of knowledge-enhanced ERNIE and bidirectional RNN for Chinese medical relationship extraction
Deep learning-based methods in relation extraction usually focus only on the representation of fine-grained text units,resulting in insufficient learned text features.A combination of ERNIE model fused with knowledge enhancement and neural net-work was proposed to perform relationship extraction.The method was divided into two parts:firstly,the text was vectorized through knowledge enhancement,specifically,the fine-grained text unit and the coarse-grained text unit they were operated by weighted average to achieve the effect of knowledge enhancement,and then the unit was predicted to be the result of RTD to de-termine whether there was any alternative word generated.Finally,the text feature vectors were fed into the BiLSTM network to get the contextual semantic information of the words,and the sentence sequence was scored and the one with the highest score was selected.The experimental results showed that the method obtained an accuracy rate of 95%,a precision rate of 91%,a re-call rate of 92%,and an f1-score of 92%for relationship extraction,which were all improved by more than5%when compared with the existing methods,so the method proposed in this paper was effective.
knowledge-enhancerelationship extractionneural networknatural language processing
NETL
NSTL
万方数据
维普
