OpenGU is a comprehensive evaluation dataset for graph unlearning (GU) released by Beijing Institute of Technology in 2025. Related research papers include... OpenGU: A Comprehensive Benchmark for Graph UnlearningIt has been selected for NeurIPS 2025 Datasets and Benchmarks, aiming to provide a unified evaluation framework, multi-domain data resources and standardized experimental settings for forgetting methods in graph neural networks.

This dataset brings together 37 multi-domain graph datasets, covering different application scenarios from node-level, edge-level to graph-level tasks. It encompasses various real-world and heterogeneous graph scenarios, including social networks, reference networks, commodity relationship networks, biochemical networks, and visual structures, providing a rich and challenging evaluation foundation for deletable learning methods.

This dataset can be divided into two categories:

• Node/edge level task dataset: 19
  Used to evaluate changes in model behavior in tasks such as node classification and link prediction after removing nodes, edges, or local structures.
• Graph-level task datasets: 18
  Used to evaluate the model's performance on graph classification tasks after the entire graph or its substructures have been removed.