基于多视图知识表示和神经网络的旅游领域实体对齐方法

Entity alignment method for tourism domain based on multi-view knowledge representation and neural network

针对目前旅游领域实体对齐任务中的长尾实体过多和现有知识以及标注数据稀缺的问题,提出一种基于多视图知识表示和神经网络相结合的实体对齐方法。采用预训练模型完成多视图的知识表示学习,获得了实体的结构嵌入、关系嵌入和描述信息嵌入,然后利用卷积神经网络对结合了三种视图嵌入的实体综合嵌入进行相似度计算。实验精准率达到91.4%、召回率达到87.9%、综合指标F1值达到89.6%。结果表明,该方法有效地完成了旅游领域的实体对齐任务。

Aiming at the problems of too many long-tail entities and existing knowledge and the scarcity of labeled data in the current entity alignment tasks in the tourism domain,this paper proposed an entity alignment method based on the combination of multi-view knowledge representation and neural network.It used a pre-trained model to complete the knowledge representation learning of multiple views,obtained the structural embedding,relationship embedding and description information embedding of the entity,and then used the convolutional neural network to calculate the similarity of the comprehensive embedding of the entities combined with the three views.The experimental accuracy rate reached 91.4%,the recall rate reached 87.9%,and the comprehensive index F 1 value reached 89.6%.The results show that the method effectively completes the entity alignment task in the tourism domain.