基于Bert和卷积神经网络的人物关系抽取研究

Research on character relationship extraction based on Bert and convolutional neural network

通过构造人物关系数据集,将人物关系定义为14类,提出了基于Bert-BiGRU-CNN的人物关系抽取网络模型.该模型首先通过Bert预训练模型获取上下文语义信息的词向量,利用双向门限循环单元网络(BiGRU)进一步获取相关的文本特征,然后加入卷积神经网络(CNN)获取局部文本特征,最后通过全连接层加Softmax分类器进行关系分类.在构造的人物关系数据集中进行了实验,结果表明,本文模型相较于其他4种模型进一步提高了人物关系抽取的精确率和召回率.

As one of the important research fields of natural language processing,relationship extraction has been widely used in many aspects,but there is currently no in-depth research in the field of character relationship extraction.In this paper,by constructing the character relationship data set,the character relationship is defined as 14 categories,and a network model of character relationship extraction based on Bert-BiGRU-CNN is proposed.The model obtains the word vector of context semantic information through the Bert pre-training model,and uses the bidirectional threshold recurrent unit network(BiGRU)to further obtain relevant text features,then joins the convolutional neural network(CNN)to obtain local text features,and finally through the fully connected layer Add Softmax classifier for relationship classification.By conducting experiments on the constructed character relationship data set,the experimental results show that the network model based on Bert-BiGRU-CNN proposed in this paper can further improve the accuracy and recall rate of character relationship extraction compared to the other four models.