GDGB comprises eight carefully selected and rigorously processed DyTAG datasets covering various domains. The nodes and edges in all datasets are endowed with rich semantic textual information to subsequently support challenging DyTAG generation tasks.

GDGB introduces two novel DyTAG generation tasks: Transductive Dynamic Graph Generation (TDGG) and Inductive Dynamic Graph Generation (IDGG). Additionally, to ensure holistic evaluation, we design multifaceted metrics that jointly consider topological patterns, temporal dynamics, and semantic quality.

We propose GAG-General, an LLM-based multi-agent framework tailored for DyTAG generation tasks. Furthermore, we implement both TDGG and IDGG tasks within GAG-General, integrating the proposed holistic evaluation metrics to ensure reproducible DyTAG generation and robust benchmarking for future baselines.