基于递归投影的结构性网络嵌入

Structural network embedding based on recurrent projection

近些年来,在网络嵌入(network embedding)领域的大多数研究都着眼于基于网络节点邻接关系的社区身份,如node2vec和DeepWalk;而基于网络拓扑结构的结构身份研究则十分匮乏,前沿方法如struc2vec等,通常效率很低。提出了递归结构性网络嵌入(recurrent structural network embedding,RSNE),一种新颖而高效的结构特征学习方法。RSNE递归式地把节点的结构身份定义为其邻居结构身份的非线性投影。为了避免退化为基于邻接关系的聚类,采用了一种有效而鲁棒的初始化方法。理论分析显示RSNE在时间复杂度上显著优于现有的结构性网络嵌入方法,可视化与量化实验结果也表明RSNE在分类准确性和鲁棒性上达到了最新方法相同或更好的效果,同时消耗的计算时间与空间消耗也远远更少。

In recent years,most researches in network embedding,such as node2vec and DeepWalk,are focused on the community identity defined by nodes’adjacency,instead of the structural identity defined by topology structure.And state-of-the-art methods of the latter,like struc2vec,are usually time-inefficient.This paper proposed RSNE(recurrent structural network embedding),a novel and efficient method to learn node representation from structural identity.RSNE defined a node’s structural identity as the non-linear projection of its neighbors’structural identities in a recurrent manner.In order to avoid degradation into clustering with nodes’adjacency,it applied an accurate and robust initialization method based on degrees.Theoretical analysis shows that the proposed method is significantly better than the existing methods in terms of time complexity,and can also effectively use hard disk space for memory optimization.Numerical visualized and quantified experiment results suggest that RSNE has equal or better performance than state-of-the-art methods in classification accuracy and robustness while consuming much less computation and time.