Network Representation Learning: From Preprocessing, Feature Extraction to Node Embedding

Network representation learning (NRL) advances the conventional graph mining of social networks, knowledge graphs, and complex biomedical physics information networks. Dozens NRL algorithms have been reported in literature. Most them focus on node embeddings for homogeneous but they differ specific encoding schemes types semantics captured used embedding. This article reviews design principles different embedding techniques over To facilitate comparison algorithms, we introduce a unified reference framework to divide generalize process given network into preprocessing steps, feature extraction model training an task such as link prediction clustering. With this unifying framework, highlight representative methods, models, at stages process. survey not only helps researchers practitioners gain in-depth understanding also provides practical guidelines designing developing next generation systems.