A Survey of Large Graph Sampling Techniques

Abstract

As a common method for simplifying network graphs, large graph sampling can reduce the size of large graph data significantly. In this paper, related works are summarized from the following perspectives: random graph sampling techniques, feature-driven large graph sampling techniques, evaluation metrics of large graph sampling and applications of large graph sampling technique. Firstly, random graph sampling is categorized into three types, including random node, random edge, and random walk graph sampling. Secondly, the feature-driven large graph sampling techniques are discussed, including topology-preserving, community structure-preserving, dynamic network association and semantic association feature-driven large graph sampling. Thirdly, the evaluation metrics of large graph sampling techniques are introduced, including topological metrics, visual perception metrics and feature-driven metrics. Finally, the applications of large graph sampling technique in social networks, geographic traffic, biomedical and deep learning are summarized, and the development of large graph sampling method is prospected.