The recent advancements in the field demonstrate a significant shift towards leveraging self-supervised learning (SSL) and semi-supervised learning techniques for handling complex, graph-structured data in various domains. In healthcare, SSL is being increasingly adopted to optimize representations from unlabeled data, enhancing tasks such as disease prediction, medical image analysis, and drug discovery. Meanwhile, in cybersecurity, the application of Positive Unlabeled (PU) and Negative Unlabeled (NU) learning is being explored to address challenges in intrusion detection, malware detection, and threat intelligence, particularly in scenarios with imbalanced or limited labeled data. Additionally, the telecom industry is witnessing innovations in fine-grained graph representation learning, with frameworks like the Data-and-Model Driven Graph Structure Learning (DMGSL) being proposed to automate the refinement of wireless data knowledge graphs, enabling more effective network analytics and management. These developments collectively highlight a trend towards more automated, data-driven approaches that can handle the complexity and heterogeneity of modern datasets, promising to advance a wide range of applications.