基于翻译模型的异质重边信息网络链路预测研究

Research of link prediction based on translation model in heterogeneous multi⁃edge information network

在异质信息网络中,异质节点对象之间具有多元关系,形成异质重边信息网络.知识图谱表示旨在将实体和关系在低维的向量空间进行嵌入,可以用来学习异质重边信息网络中实体间的多元关系.首先通过注意力机制对异质重边信息网络中的多元关系进行融合表示,进而将异质节点的类型信息进行多元关系融合空间的映射,在多元关系融合空间上提出基于翻译的异质重边嵌入模型,用以学习异质节点之间的链路关系.最后,在MovieLens100k电影数据集上进行了异质节点多元关系的链路预测实验.实验结果表明,在异质重边信息网络中,基于改进的翻译模型在实体间链路预测性能方面要优于传统的知识表示方法,可以有效地提升链路预测的精度.

In heterogeneous information network,heterogeneous nodes have multiple relations which can form heterogeneous multi‐edge information network.Knowledge graph‐based representation aims to embed object entities and relations into a lowdimensional vector space which can be used to learn the multiple relations between entities in heterogeneous multi‐edge information network.In this paper,we first leverage fused representation of multiple relations for heterogeneous multi‐edge information network in terms of attention mechanism.Then,projected matrices are adopted to map the types of heterogeneous nodes into fused spaces of multiple relations.More,in the fusion representation space of multiple relations,translation‐based heterogeneous multi‐edge embedding model is proposed to learn the link relations among heterogeneous nodes.Finally,link prediction experiments of heterogeneous multi‐edge relations are carried out on MovieLens100k dataset.The experimental results demonstrate that the novel translation model is superior to traditional knowledge representation methods at the aspect of link prediction performance,which can effectively improve the accuracy of link prediction.