Graph Representation Learning with Missing or Incomplete Information

This thesis addresses key challenges in Graph Representation Learning (GRL) by developing innovative methods to handle incomplete information, improve efficiency, and enhance contextual learning. It explores unsupervised and semi-supervised approaches, introducing Graph Contrastive Adjusted Zooming (G-Zoom) for multi-scale learning, Unifying Graph Contrastive Learning (UGCL) for flexible contextual adaptation, Graph Group Discrimination (GGD) for computational efficiency, and Graph Complementary Learning (GOAL) to address incomplete topology. These methods significantly outperform existing baselines, demonstrating robust scalability and applicability across diverse datasets. Limitations in dynamic and ultra-large-scale graphs are noted, offering future research directions to advance GRL in complex, real-world scenarios.