Authors: An Zhang, Wenchang Ma, Pengbo Wei, Leheng Sheng, Xiang Wang

Published on: February 21, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2402.13769

Summary

• What is new: Introduction of Adversarial Graph Dropout (AdvDrop) for unbiased representation learning in graph neural networks used for recommender systems.
• Why this is important: Current graph neural networks aggregate user-item interactions without differentiating between biased and unbiased data, leading to distorted recommendations.
• What the research proposes: AdvDrop framework employs adversarial learning to separate biased and unbiased interactions, ensuring invariant bias-mitigated and bias-aware representations.
• Results: Significant improvements on five datasets, demonstrating the ability to achieve unbiased recommendations and meaningful separation of subgraphs.

Technical Details

Technological frameworks used: Adversarial Graph Dropout (AdvDrop)

Models used: Graph Neural Networks (GNNs) for collaborative filtering

Data used: Five public datasets covering general and specific biases

Potential Impact

Online recommendation platforms, e-commerce websites, streaming services

Want to implement this idea in a business?

We have generated a startup concept here: FairRecs.