TDGBench_v1.0

TDGBench v1.0 is a biomedical benchmark for evaluating text-derived knowledge graph construction methods and graph neural networks through semi-supervised node classification. This archive groups the GT2KG and KGGen graphs derived from the MedMentions corpus, including their raw and processed variants, together with the shared evaluation nodes and predefined data splits. Each variant is provided separately to support reproducible comparisons. The benchmark also uses a curated UMLS-NCI reference graph. Because UMLS content is subject to licensing restrictions, the reference graph is not redistributed in this archive. The accompanying documentation describes how authorized users can reconstruct it. The README specifies the contents, formats, preprocessing steps, licenses, and evaluation protocol for this release.