BIG-LSTM-CRF model integrating Global Semantic Information
The named entity recognition task is a kind of natural language processing task that per-forms sequence annotation for the input text sentence,and its purpose is to extract the subject entity and object entity in the text sentence.The extraction method based on deep neural network has ob-tained excellent performance,and BI-LSTM-CRF is one of the most effective and representative models.However,the influence of global semantic information on entity recognition accuracy is ignored in the training process.This paper improves the performance of BI-LSTM-CRF model for named entity recognition tasks by introducing global semantic information.Firstly,the high-dimensional semantic information of the input text sentence is extracted by adding a fully connected layer with activation operation,and then the high-dimensional semantic information is deeply fused with each character through a fully connected layer,and the vector representation of the sentence with the global semantic information is obtained,which is used for subsequent named entity recognition tasks.By using the improved model on the CLUENER2020 dataset,it is verified that adding the global semantic informa-tion fusion module can improve the accuracy of model named entity recognition.
BI-LSTM-CRFnatural language processingnamed entity recognitionneural network
万方数据
维普
