Learning from Graph: Mitigating Label Noise on Graph through Topological Feature Reconstruction
Conference proceedings talk, The 34th ACM International Conference on Information and Knowledge Management (CIKM 2025) , Seoul, Republic of Korea
This is a presentation of our paper Learning from Graph: Mitigating Label Noise on Graph through Topological Feature Reconstruction at CIKM 2025. You can find the paper here.
FAQ: Feature noise and Structure Noise are common in graph learning scenarios. Is TFR equally robust to these types of noise?
Answer: No, TFR is specifically optimized for label noise and has not been optimized for Feature noise or Structure Noise. However, we must admit that feature, structure, and label noise are all common issues in real-world scenarios. We believe designing a model that is robust to all of these types of noise is a crucial direction for future work. Of course, when every part of the graph data contains severe noise, a more efficient approach is to address the problem from a data perspective (e.g., re-cleaning the data to improve data quality)
