<ul><li>Large language models (LLMs) have strong generalization ability and have been used in zero-shot learning problems.</li><li>Adopting LLMs in text-attributed graphs (TAGs) faces challenges of limited information on graph structure and unreliable responses.</li><li>This paper introduces Dynamic Text Bundling Supervision (DENSE) method to query LLMs with bundles of texts, obtain bundle-level labels, and supervise graph neural networks.</li><li>Experimental results across ten datasets validate the effectiveness of the proposed DENSE method.</li></ul>

Dynamic Text Bundling Supervision for Zero-Shot Inference on Text-Attributed Graphs

Discover more