基于知识增强的中文命名实体识别

Chinese Named Entity Recognition Based on Knowledge Enhancement

基于字词联合的中文命名实体识别模型能够兼顾字符级别与词语级别的信息,但受未登录词影响较大且在小规模数据集上存在训练不充分等问题。在现有LR-CNN模型的基础上,提出一种结合知识增强的中文命名实体识别模型,采用相对位置编码的多头注意力机制提高模型上下文信息捕捉能力,通过实体词典融入先验知识降低未登录词的影响并增强模型学习能力。实验结果表明,该模型在保持较快解码速度和较低计算资源占用量的情况下,在MSRA、People Daily、Resume、Weibo数据集上相比SoftLexicon、FLAT等模型F1值均有明显提升,同时具有较强的鲁棒性和泛化能力。

Chinese Named Entity Recognition(CNER)models can capture both character-level and word-level information,but are vulnerable to the negative impact of Out-of-Vocabulary(OOV)words and insufficient training caused by small datasets.To address this problem,an additional knowledge enhanced CNER model is proposed based on the LR-CNN model.The model uses the multi-head attention mechanism with relative position embedding to improve the ability of the model to capture contextual information.Additionally,the entity dictionary is used to add prior knowledge to reduce the impact of OOV words,and to enhance the generalization ability of the model.Experimental results show that compared with SoftLexicon,FLAT and other models on the MSRA,People Daily,Resume,Weibo datasets,the F1 value has significantly improved.It displays excellent robustness and generalization ability.