Tag
TAROT proposes a GNN-based framework that leverages LLMs to construct and refine task-adaptive semantic graphs for few-shot tabular learning, achieving state-of-the-art performance.