Complex Graph Analysis and Representation Learning: Problems, Techniques, and Applications

Graph representation learning (GRL) has become a new learning paradigm, supporting a wide range of tasks such as node classification, link prediction, and graph classification. However, the effectiveness of graph analysis heavily depends on the quality of data representation. While existing GRL methods have made significant progress in learning from simple graphs, addressing the challenges posed by complex graph structures remains an active area of research. In many real-world scenarios, graph data usually exhibits characteristics such as complexity, heterogeneity, and dynamicity, where objects and their interactions may be multi-type, multi-modal, and even multi-dimensional, posing challenges to graph-related analysis. To tackle these challenges, GRL has been developed and widely used to model more complex and powerful graphs. In this survey, we provide a comprehensive and structured analysis of the existing literature on GRL from two clear points of view of simple graph and complex graph. We begin by providing a detailed and thorough analysis of state-of-the-art GRL techniques and classify them according to their underlying learning mechanisms. Furthermore, we systematically investigate GRL from the perspective of complex graphs to address the challenges posed by graph complexity. We emphasize the need for specialized GNN models that can handle the complexity of such systems. Finally, we highlight several promising directions for future research.