基于BERT+Bi-LSTM+CRF的航天领域命名实体识别研究

Research on Named Entity Recognition in Aerospace Field Based on BERT+Bi-LSTM+CRF

针对互联网开放数据中文本表述模糊、实体边界不清等问题,构建航天语料库Space-Corpus,提出一种基于BERT+Bi-LSTM+CRF的航天领域命名实体识别模型。基于微调的多层双向Transformer编码器(bidirectional encoder representations from transformer,BERT)模型生成输入语料的向量化表示,结合双向长短期记忆网络(bi-directional long short-term memory,Bi-LSTM)获取上下文特征,通过条件随机场(conditional random field,CRF)层进行序列解码标注,输出得分最高的预测标签。实验结果表明,该模型在Space-Corpus语料库上较基于BERT模型、基于BERT+Bi-LSTM以及基于CNN+Bi-LSTM+CRF识别模型的准确率、召回率及F1值均有提升。

Aiming at the problems of fuzzy text expression and unclear entity boundary in Internet open data,this paper constructs Space-Corpus,and proposes a named entity recognition model based on BERT+Bi-LSTM+CRF.The bidirectional encoder representations from transformer(BERT)model based on two-way training Transformer generates the vectorized representation of the input corpus,combines with bi-directional long short-term memory(Bi-LSTM)to obtain the context features,decodes and annotates the sequence through conditional random field(CRF),and outputs the predicted label with the highest score.Experimental results show that the proposed model outperforms the BERT model,BERT+Bi-LSTM model and CNN+Bi-LSTM+CRF model in terms of accuracy,recall and F1 score on Space-Corpus corpus.