Kojaku et al., "Residual2Vec: Debiasing graph embedding with random graphs" (2021)

2021-10-14 → 2026-08-14

Sadamori Kojaku, Jisung Yoon, Isabel Constantino, and Yong-Yeol Ahn, Advances in Neural Information Processing Systems 34, 24150–24163 (2021)
Link | arXiv | Code

@inproceedings{kojaku2021residual2vec,
    author = {Sadamori Kojaku and Jisung Yoon and Isabel Constantino and Yong-Yeol Ahn},
    title = {Residual2Vec: Debiasing graph embedding with random graphs},
    booktitle = {Advances in Neural Information Processing Systems},
    volume = {34},
    pages = {24150--24163},
    publisher = {Curran Associates, Inc.},
    url = {https://proceedings.neurips.cc/paper/2021/hash/ca9541826e97c4530b07dda2eba0e013-Abstract.html},
    archivePrefix = {arXiv},
    eprint = {2110.07654},
    primaryClass = {cs.LG},
    year = {2021},
}

Shows how random-walk sampling can transmit structural biases, especially degree bias, into graph embeddings. Residual2Vec uses random-graph baselines to remove selected structural effects, improving link prediction and clustering while allowing salient structure to be modeled explicitly.

Graph embedding, Negative sampling, Random graph

Receive my updates