Toward Scalable Multi-Level and Dynamic Representation Learning for Multiplex Graphs
Multiplex graphs provide an expressive modeling framework for real-world systems in which entities interact through multiple types of relations. However, representation learning on multiplex graphs remains challenging due to their high dimensionality, large scale, and dynamic nature. In our journal work published in IEEE Transactions on Knowledge and Data Engineering (DOI: 10.1109/TKDE.2023.3305809), we introduced HMGE, a hierarchical aggregation framework designed to capture complex interactions across multiple dimensions in static multiplex graphs. While effective, HMGE leaves several fundamental challenges open, including scalability in large-scale and high-dimensional settings, representation learning across multiple levels of abstraction, and the modeling of time-evolving multiplex structures. This Journal-Conference paper outlines research directions that address these challenges by investigating scalable learning strategies that preserve graph completeness, geometry-aware multi-level representation learning that accounts for latent manifold structure, and dynamic modeling approaches for evolving multiplex graphs. By addressing these issues, this work advances graph representation learning toward principled and scalable solutions capable of exploiting large-scale, highdimensional, and time-evolving multiplex graph data in modern interconnected systems.