Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning

Researchers propose LEMP4HG, a new language model-enhanced approach for improving graph neural networks on heterophilic graphs where connected nodes have different characteristics. The method leverages language models to better understand semantic relationships between text-attributed nodes, outperforming existing methods while maintaining efficiency through selective message enhancement.

• →Traditional graph neural networks struggle with heterophilic graphs where connected nodes are dissimilar, limiting their effectiveness.
• →LEMP4HG uses language models to explicitly model semantic relationships between nodes using their text attributes for better message passing.
• →The approach introduces MVRD heuristic with active learning to selectively enhance messages for computational efficiency.
• →Experimental results show consistent outperformance on heterophilic graphs while maintaining robust performance on homophilic graphs.
• →The method addresses limitations of existing approaches that rely on suboptimal message representation and overlook semantic potential of node text.