基于实体相似性的知识表示学习方法

Entity similarity based knowledge graph embedding

知识表示学习旨在将知识图谱中的实体和关系表示成低维稠密实值向量,能有效缓解知识图谱的数据稀疏性和显著提升计算效率。然而,现有大多数知识表示学习方法仅将实体视为三元组的一个组成部分,没有考虑实体自身具有的特质,如实体相似性。为了加强嵌入向量的语义表达,提出基于实体相似性的表示学习方法SimE。该方法首先利用实体的结构邻域度量实体的相似性,再将实体的相似性和拉普拉斯特征映射结合作为基于三元组事实的表示学习方法的约束,形成联合表示。实验结果表明,该方法在链接预测和三元组分类等任务上与目前最好的方法性能接近。

Knowledge representation learning aims to represent the entities and relationships in the knowledge graph as low-dimensional dense real-valued vectors,which can effectively alleviate the data sparsity of the knowledge graph and significantly improve the calculation efficiency.However,most existing knowledge representation learning methods only treat entities as an integral part of triples,and do not consider the characteristics of entities themselves,such as entity similarity.In order to strengthen the semantic expression of embedded vectors,this paper proposed a representation learning method SimE based on entity similarity.The method first used the structural domain of the entity to measure the similarity of the entity,and then combined the similarity of the entity and the Laplace feature map as a constraint of the representation learning method based on the fact of triples to form a joint representation.Experimental results show that the method is close to the best method currently in tasks such as link prediction and triple classification.