面向异质变分超图自动编码器的超边链接预测模型

Exploring heterogeneous variational hypergraph autoencoder for hyper-edge link prediction

如何采用超边建模网络数据中的多元关联关系,实现潜在超边链接关系的预测具有重要的现实意义。现有方法主要集中于研究具有成对关系的网络数据,然而,直接将现有的链接预测方法用于超图网络中的超边链接预测具有一定的局限性。因此,提出基于异质变分超图自动编码器的超边链接预测模型(heterogeneous variational hypergraph autoencoder,HVGAE)。首先,利用超图卷积实现变分超图自动编码器,将超图网络数据转换成一种低维空间表示;其次,加入节点近邻度函数,最大程度地保留其结构信息,从而构建异质超图网络超边链接预测模型。针对三种不同类型的超图网络进行实验,结果表明相比其他的基准方法,HVGAE模型获得了较好的预测结果,说明其能够较好地解决超图网络中的超边链接预测问题。

How to use hyper-edge to model the multiple association relationship in network data and realize the prediction of potential hyper-edge link relationship has important practical significance.Existing link prediction methods mainly focus on networks with pairwise relationships.However,directly applying existing link prediction methods to hyper-edge link prediction in hypergraph networks has certain limitations.Therefore,this paper proposed a hyper-edge link prediction model,called HVGAE(heterogeneous variational hypergraph autoencoder)based on heterogeneous variational hypergraph autoencoder.Firstly,this method used hypergraph convolution to realize variational hypergraph autoencoder,and converted the hypergraph network data into a low-dimension representation.Then it added nodes near-neighbor similarity function to retain the structural information to the largest degree,so as to construct heterogeneous hyper-edge link prediction model.Experiments on three diffe-rent types of hypergraph networks,the results show that HVGAE model has gained better prediction result compared with that of other baseline methods,indicating that it can better solve the problem of hyper-edge link prediction in the hypergraph network.