This is the implementation of our paper 'Chain of Propagation Prompting for Node Classification'
In this paper, we propose using a simple pattern of message-passing, which is based on a weighted multi-hop graph, to prompt self-attention to capture a more complex message-passing pattern, resulting in the maximal receptive field and reduced reliance on label information. Also, a majority voting-like module is used to improve predictive confidence.
The paper can be viewed via https://dl.acm.org/doi/10.1145/3581783.3612431, and the framework of our CPP is listed in the following.
If you are interested in our work, please also cite our paper:
@inproceedings{zhu2023chain, title={Chain of Propagation Prompting for Node Classification}, author={Zhu, Yonghua and Deng, Zhenyun and Chen, Yang and Amor, Robert and Witbrock, Michael}, booktitle={Proceedings of the 31st ACM International Conference on Multimedia}, pages={3012--3020}, year={2023} }