PrimeKG-CL: Benchmark for Evolving Biomedical Knowledge Graphs

Researchers introduced PrimeKG-CL, a continual graph learning benchmark built from nine authoritative biomedical databases with over 129,000 nodes and 8.1 million edges. It uses two genuine temporal snapshots from June 2021 and July 2023, capturing 5.83 million added edges, 889,000 removed, and 7.21 million persistent edges. The benchmark includes 10 entity-type-grouped tasks, multimodal node features, and per-task test stratification. It evaluates three tasks: biomedical relationship prediction, entity classification, and knowledge graph question answering. This addresses the gap in existing continual graph learning studies that rely on synthetic splits of static, generic knowledge graphs, which cannot replicate the asynchronous, structured evolution of real biomedical KGs.