基于伪孪生网络的政务实体链接模型

Government affairs entity linking model based on pseudo-siamese network

为了提高政务领域实体链接任务的准确率,降低响应时间,提出了一种基于伪孪生网络的实体链接模型.模型通过伪孪生网络框架解耦问句和候选实体的特征提取过程,并预先计算候选实体的向量表示,显著地提高了模型在大规模数据集上的性能.同时,通过引入候选实体在知识图谱中的上下文信息,增强实体链接模型的语义匹配能力,从而提高链接准确率.实验结果表明:相比现有基于统计学的政务实体链接模型,所提模型在准确性和响应速度上具有综合优势,可满足交互式政务问答应用场景需求.

To improve the accuracy of the entity linking task in government affairs field and reduce the response time,an entity linking model based on the pseudo-siamese network is proposed.In the model,the feature extraction process between user queries and candidate entities is decoupled by pseudo-siamese network framework,and the vector representation of the candidate entities is pre-calculated,which significantly improves the performance of the model on large-scale dataset.At the same time,by introducing the context information of the candidate entity in the knowledge graph,the semantic matching ability of the entity linking model is enhanced,thereby the link accuracy is improved.The experiment results showed that compared with the existing statistics-based government entity linking model,the proposed entity linking model has comprehensive advantages in accuracy and response speed,and can meet the needs of interactive government affairs question and answer application scenarios.