Research and development of network representation learning

Abstract

Network representation learning is a bridge between network raw data and network application tasks which aims to map nodes in the network to vectors in the low-dimensional space. These vectors can be used as input to the machine learning model for social network application tasks such as node classification, community discovery, and link prediction. The traditional network representation learning methods are based on homogeneous information network. In the real world, the network is often heterogeneous with multiple types of nodes and edges. Moreover, from the perspective of time, the network is constantly changing. Therefore, the research method of network representation learning is continuously optimized with the complexity of network data. Different kinds of network representation learning methods based on different networks were introduced and the application scenarios of network representation learning were expounded.

Keywords

• large-scale information network
• network representation learning
• network embedding
• deep learning