Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fair Augmentation for Graph Collaborative Filtering

Ludovico Boratto, Francesco Fabbri, Gianni Fenu, Mirko Marras, Giacomo Medda

TL;DR

This paper serves as a solid response to recent research highlighting unfairness issues in graph collaborative filtering by reproducing one of the latest mitigation methods and reveals that fair graph augmentation is consistently effective on high-utility models and large datasets.

Abstract

Recent developments in recommendation have harnessed the collaborative power of graph neural networks (GNNs) in learning users' preferences from user-item networks. Despite emerging regulations addressing fairness of automated systems, unfairness issues in graph collaborative filtering remain underexplored, especially from the consumer's perspective. Despite numerous contributions on consumer unfairness, only a few of these works have delved into GNNs. A notable gap exists in the formalization of the latest mitigation algorithms, as well as in their effectiveness and reliability on cutting-edge models. This paper serves as a solid response to recent research highlighting unfairness issues in graph collaborative filtering by reproducing one of the latest mitigation methods. The reproduced technique adjusts the system fairness level by learning a fair graph augmentation. Under an experimental setup based on 11 GNNs, 5 non-GNN models, and 5 real-world networks across diverse domains, our investigation reveals that fair graph augmentation is consistently effective on high-utility models and large datasets. Experiments on the transferability of the fair augmented graph open new issues for future recommendation studies. Source code: https://github.com/jackmedda/FA4GCF.

Fair Augmentation for Graph Collaborative Filtering

TL;DR

This paper serves as a solid response to recent research highlighting unfairness issues in graph collaborative filtering by reproducing one of the latest mitigation methods and reveals that fair graph augmentation is consistently effective on high-utility models and large datasets.

Abstract

Recent developments in recommendation have harnessed the collaborative power of graph neural networks (GNNs) in learning users' preferences from user-item networks. Despite emerging regulations addressing fairness of automated systems, unfairness issues in graph collaborative filtering remain underexplored, especially from the consumer's perspective. Despite numerous contributions on consumer unfairness, only a few of these works have delved into GNNs. A notable gap exists in the formalization of the latest mitigation algorithms, as well as in their effectiveness and reliability on cutting-edge models. This paper serves as a solid response to recent research highlighting unfairness issues in graph collaborative filtering by reproducing one of the latest mitigation methods. The reproduced technique adjusts the system fairness level by learning a fair graph augmentation. Under an experimental setup based on 11 GNNs, 5 non-GNN models, and 5 real-world networks across diverse domains, our investigation reveals that fair graph augmentation is consistently effective on high-utility models and large datasets. Experiments on the transferability of the fair augmented graph open new issues for future recommendation studies. Source code: https://github.com/jackmedda/FA4GCF.
Paper Structure (21 sections, 13 equations, 3 figures, 4 tables)

This paper contains 21 sections, 13 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Reproducibility Study
  3. GNN-based Recommendation Task
  4. Reproduced Approach
  5. Graph Augmentation Pipeline
  6. Sampling Policies
  7. Experimental Setup
  8. Datasets
  9. Models
  10. Experimental Settings
  11. FA4GCF
  12. Sampling Policy Formalization
  13. Augmentation Formalization
  14. Extended Sampling Policies Set
  15. Sampling Policies Overlap
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Jaccard similarity between sampling policies across distinct models, due to the policies being model-dependant.
  • Figure 2: Comparison among sampling policies in terms of unfairness mitigation ($\Delta$ percentage) across gender groups. The $\Delta$ of the base model is labeled as Base. The first column (row) of each subplot pertains to the User-sampling (Item-sampling) settings. The other cells pertain to the User-Item sampling settings. Smaller values, i.e. darker cells, are better.
  • Figure 3: Impact of $\Psi_{\mathcal{U}}$ and $\Psi_{\mathcal{I}}$ on fairness ($\Delta$) and utility (NDCG) levels on the settings identified by each subplot title.