Abstract: The developing automated construction technology and the growing amount of information have introduced noises and conflicts into knowledge graph.For this reason,the evaluation of knowledge credibility is required to better support the application of knowledge graph.Based on knowledge representation learning,this paper designs a knowledge credibility evaluation model,PTCA,which calculates the knowledge credibility based on the strength of the correlation between entities,entity type information,and multi-step path information.The model has been tested with multiple tasks,including the triplet classification task,the noise detection task for knowledge graph,and the knowledge graph completion task.Experimental results show that the model can detect the noises and conflicts existing in the knowledge graph,as well as effectively calculate the credibility of triplets.It demonstrates better evaluation performance on noisy datasets than CKRL and the PTransE benchmark model.

Key words: knowledge graph, knowledge representation learning, knowledge credibility evaluation, triplet credibility evaluation, noise detection

摘要： 知识图谱自动化构建技术的发展以及信息量的增加导致知识图谱中引入了噪声和冲突，为了有效应用知识图谱，需要对知识的可信度进行评估。建立一种基于知识表示学习的知识可信度评估模型PTCA，利用实体之间的关联强度、实体类型信息以及多步路径信息对知识的可信度进行计算。通过三元组分类、知识图谱噪声检测以及知识图谱补全等3个任务对模型性能进行测试，结果表明，PTCA模型可以检测知识图谱内部存在的噪声和冲突，对三元组知识的可信度进行有效计算，且在有噪声干扰的数据集上的评估性能优于CKRL和PTransE模型。

关键词: 知识图谱, 知识表示学习, 知识可信度评估, 三元组可信度评估, 噪声检测

CLC Number: 

• TP182