A Survey of Large-scale Complex Information Network Representation Learning Methods

Abstract

With the increasing growth of data scale and the increasing complexity of network structure, the heterogeneity, high sparsity, heterogeneity and high dimensionality of large-scale complex networks have become increasingly prominent. How to represent network information reasonably and effectively so as to better serve subsequent network analysis tasks has become the key problem of network analysis. Network representation learning aims to represent the components (nodes, edges, subnets, etc.) in the network as low dimensional dense vectors. This vector can fully retain the original network structure information and other heterogeneous information, and has the advantages of improving computing efficiency, mitigating the impact of data sparsity, and effectively merging heterogeneous information. This research focuses on homogeneous networks and heterogeneous networks, summarizes and analyzes advantages and shortcomings of the network representation learning methods in recent years, and gives the possible research directions and contents in the future work.