Researchers from HKUST and Southern University of Science and Technology have proposed a multimodal dataset called GITQA which combines textual and visual information for graph reasoning tasks in the development of Large Language Models (LLMs).