Yoo, J.*, Lee, M. C.*, Shekhar, S., & Faloutsos, C. (2023, August). Less is More: SlimG for Accurate, Robust, and Interpretable Graph Mining. 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023.
https://dl.acm.org/doi/10.1145/3580305.3599413
Please cite the paper as:
@inproceedings{yoo2023less,
title={Less is More: SlimG for Accurate, Robust, and Interpretable Graph Mining},
author={Yoo, Jaemin and Lee, Meng-Chieh and Shekhar, Shubhranshu and Faloutsos, Christos},
booktitle={Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages={3128--3139},
year={2023}
}
How can we solve semi-supervised node classification in various types of graphs possibly with noisy features and structures?
We propose SlimG, which exhibits the folloiwing desirable properties:
- C1.1 - Accurate, on both real-world and synthetic datasets, almost always winning or tying in the first place.
- C1.2 - Robust, being able to handle numerous real settings such as homophily, heterophily, no network effects, useless features.
- C1.3 - Fast and Scalable, using carefully chosen features, it takes only 32 seconds on million-scale real-world graphs (ogbn-Products) on a stock server.
- C1.4 - Interpretable, learning the largest weights on informative features, ignoring noisy ones, based on the linear decision function.
We also explain the reasons for its success, thanks to our three additional contributions:
- C2 - Explanation: We propose GNNExp, a framework for the systematic linearization of a GNN.
- C3 - Sanity Checks: We propose seven possible scenarios of graphs (homophily, heterophily, no network effects, etc.), which reveal the strong and weak points of each GNN.
- C4 - Experiments: We conduct extensive experiments to better understand the success of SlimG even with its simplicity.
The code is written in Python 3.10.8 and built on a number of packages.
Please see "requirements.txt" for the package details.
To run the code, you can simply use the following comment:
python src/main.py
The datasets will be automatically downloaded when running for the first time.