基于BERT多知识图融合嵌入的中文NER模型

A Chinese NER Model Based on BERT with Multi Knowledge Graph Fusion and Embedding

针对目前特定领域知识图谱构建效率低、领域已有知识图谱利用率不足、传统模型提取领域语义专业性强实体困难的问题,提出了基于BERT多知识图融合嵌入的中文NER模型(BERT-FKG),实现了对多个知识图通过融合语义进行实体间属性共享,丰富了句子嵌入的知识。该模型在开放域和医疗领域的中文NER任务中,表现出了更好的性能。实验结果表明,多个领域知识图通过计算语义相似度进行相似实体的属性共享,能够使模型吸纳更多的领域知识,提高在NER任务中的准确率。

Aiming at the problems of low efficiency in the construction of knowledge graph in specific fields,insufficient utilization of existing knowledge graph in the field,and difficulty in extracting domain semantic professional entities from traditional models,a Chinese named entity recognition(NER)model based on Bert(bidirectional encoder representations from transformers)multi knowledge graph fusion and embedding(BERTFKG)is proposed in this paper.It realizes the attribute sharing among entities through semantic fusion for multiple knowledge graphs and enriches the knowledge of sentence embedding.The proposed model shows better performance in Chinese NER tasks in open domain and medical field.The experimental results show that multiple domain knowledge graphs share the attributes of similar entities by calculating semantic similarity,which can make the model absorb more domain knowledge and improve the accuracy in NER tasks.