基于BERT的关系抽取方法研究

Research on Relation Extraction Method Based on BERT

基于转换器的双向编码表征(Bidirectional Encoder Representation from Transformers,BERT)是由谷歌在2018发布的预训练模型,该模型在11项NLP任务中取得了最优的结果。基于此,笔者提出了基于BERT的关系抽取模型,将头节点和尾节点的先验知识融合到BERT模型中,并采用SemEval公开数据集进行实验。实验结果表明,此融合方法提高了原模型的召回率和F1值。

BERT is a pre-trained model released by Google in 2018. This model has achieved state of the art results in 11 NLP tasks. Based on this, This paper proposes a relationship extraction model based on BERT, which merges the prior knowledge of the head node and the tail node into the BERT model, and uses SemEval public data set for experiments. Experiments results show that this fusion method improves the recall rate and F1 value of the original model.