No Prejudice! Fair Federated Graph Neural Networks for Personalized Recommendation

Authors: Nimesh Agrawal, Anuj Kumar Sirohi, Sandeep Kumar, Jayadeva

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental evaluation on three publicly available datasets showcases the efficacy of F2PGNN in mitigating group unfairness by 47% 99% compared to the state-of-the-art while preserving privacy and maintaining the utility. The results validate the significance of our framework in achieving equitable and personalized recommendations using GNN within the FL landscape. Extensive experiments on three publicly available datasets (one small and two large-scale) elucidate the effectiveness of F2PGNN. Detailed analysis and ablation study further validates the strength and efficacy of individual components proposed in F2PGNN.
Researcher Affiliation Academia 1 Department of Electrical Engineering, Indian Institute of Technology, Delhi, India 2 Yardi School of Artificial Intelligence, Indian Institute of Technology, Delhi, India
Pseudocode Yes The detailed pseudo-code of F2PGNN algorithm is given in Algorithm 1. ... the pseudocode for algorithm is given in (Appendix B of (Agrawal et al. 2024)).
Open Source Code Yes Source code is at: https://github.com/nimeshagrawal/F2PGNN-AAAI24
Open Datasets Yes To empirically evaluate our framework, we have used three publicly available real-world datasets, namely Movie Lens (ML-100K and ML-1M versions) (Harper and Konstan 2015), and Amazon-Movies (AM) ( 500K ratings) (Ni, Li, and Mc Auley 2019).
Dataset Splits Yes We then follow 80/10/10 train/validation/test split for each user history sorted according to rating timestamps.
Hardware Specification Yes All experiments are performed on a machine with AMD EPYC 7282 16-Core Processor @ 2.80GHz with 128GB RAM, 80GB A100 GPU on Linux Server.
Software Dependencies Yes We implemented F2PGNN in Python 3.9 using Tensor Flow 2.5.
Experiment Setup Yes We then follow 80/10/10 train/validation/test split for each user history sorted according to rating timestamps. For all datasets, we first filter 20-core data 1, which ensures that each user has rated at least 20 items and each item has been interacted by at least 20 users...hyperparameter settings are given in Appendix D.3 of (Agrawal et al. 2024). For F2PGNN, the parameter β controls how much weightage is be given to the Lfair for fairness adaptation in each communication round.