Complex networks have been widely adopted to model diverse real-world systems. However, the increasing complexity of such networks, characterized by heterogeneity, higher-order interactions, and temporal and dynamic learning settings, poses critical challenges to traditional machine learning methods. To address these challenges, graph representation learning (GRL) has emerged as a fundamental paradigm for analyzing complex networks by learning low-dimensional representations that preserve structural and semantic information in non-Euclidean spaces. With the rapid development of deep learning, graph neural networks (GNNs) have become central to GRL, enabling end-to-end learning on graph-structured data. In this survey, we provide a systematic and comprehensive overview of graph representation learning in complex networks. We review representation learning techniques at multiple granularities, including node-level, edge-level, and graph-level embeddings. In addition, we summarize key aspects of advanced GRL research, encompassing GNN architectural design, structural complexity modeling, learning and optimization strategies, as well as representative real-world applications. Finally, we highlight emerging topics and open challenges to outline promising directions for future research.
Li et al. (Sat,) studied this question.