Graph Purification

Graph purification aims to improve the quality and reliability of graph data by identifying and removing or correcting errors, noise, or malicious manipulations. Current research focuses on developing algorithms and models that leverage both local and global graph properties to detect and mitigate issues like mislabeled nodes, spurious edges, and adversarial attacks, often employing iterative refinement strategies or incorporating human-in-the-loop approaches. These advancements are crucial for enhancing the robustness and accuracy of graph-based machine learning models, particularly graph neural networks, and improving the reliability of knowledge graphs used in various applications.