Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robustness in Fairness against Edge-level Perturbations in GNN-based Recommendation

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

TL;DR

The paper addresses robustness in fairness for graph-based recommender systems under edge-level perturbations in a white-box setting. It formalizes consumer and provider fairness via demographic parity and introduces an edge perturbation mechanism that uses a trainable score to select edges, optimizing a differentiable fairness objective at inference. Experiments on three datasets and three GNN models reveal that consumer fairness is more sensitive to perturbations than provider fairness, with NGCF often exhibiting stronger robustness across settings. The findings expose gaps in current robustness evaluation for fairness and point to future work involving more models, attack types, and grey/black-box scenarios, with code made publicly available for reproducibility.

Abstract

Efforts in the recommendation community are shifting from the sole emphasis on utility to considering beyond-utility factors, such as fairness and robustness. Robustness of recommendation models is typically linked to their ability to maintain the original utility when subjected to attacks. Limited research has explored the robustness of a recommendation model in terms of fairness, e.g., the parity in performance across groups, under attack scenarios. In this paper, we aim to assess the robustness of graph-based recommender systems concerning fairness, when exposed to attacks based on edge-level perturbations. To this end, we considered four different fairness operationalizations, including both consumer and provider perspectives. Experiments on three datasets shed light on the impact of perturbations on the targeted fairness notion, uncovering key shortcomings in existing evaluation protocols for robustness. As an example, we observed perturbations affect consumer fairness on a higher extent than provider fairness, with alarming unfairness for the former. Source code: https://github.com/jackmedda/CPFairRobust

Robustness in Fairness against Edge-level Perturbations in GNN-based Recommendation

TL;DR

The paper addresses robustness in fairness for graph-based recommender systems under edge-level perturbations in a white-box setting. It formalizes consumer and provider fairness via demographic parity and introduces an edge perturbation mechanism that uses a trainable score to select edges, optimizing a differentiable fairness objective at inference. Experiments on three datasets and three GNN models reveal that consumer fairness is more sensitive to perturbations than provider fairness, with NGCF often exhibiting stronger robustness across settings. The findings expose gaps in current robustness evaluation for fairness and point to future work involving more models, attack types, and grey/black-box scenarios, with code made publicly available for reproducibility.

Abstract

Efforts in the recommendation community are shifting from the sole emphasis on utility to considering beyond-utility factors, such as fairness and robustness. Robustness of recommendation models is typically linked to their ability to maintain the original utility when subjected to attacks. Limited research has explored the robustness of a recommendation model in terms of fairness, e.g., the parity in performance across groups, under attack scenarios. In this paper, we aim to assess the robustness of graph-based recommender systems concerning fairness, when exposed to attacks based on edge-level perturbations. To this end, we considered four different fairness operationalizations, including both consumer and provider perspectives. Experiments on three datasets shed light on the impact of perturbations on the targeted fairness notion, uncovering key shortcomings in existing evaluation protocols for robustness. As an example, we observed perturbations affect consumer fairness on a higher extent than provider fairness, with alarming unfairness for the former. Source code: https://github.com/jackmedda/CPFairRobust
Paper Structure (24 sections, 5 equations, 3 figures, 1 table)

This paper contains 24 sections, 5 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Attacks and Robustness in Recommendation
  4. Fairness in Recommendation
  5. Robustness and Beyond-accuracy Aspects
  6. Methodology
  7. Perturbation Task in Graph-based Recommendation
  8. Recommendation task.
  9. Perturbation task.
  10. Graph Perturbation Mechanism
  11. Fairness Notion and Operationalization
  12. Fairness Notion.
  13. Consumer and Provider Fairness Operationalization.
  14. Experimental Evaluation
  15. Evaluation Setting
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Impact of edges addition ($\dotplus$ Add) and deletion ($\dotplus$ Del) on the robustness in fairness, reported as the relative difference in $M$ between the non-perturbed and perturbed model, i.e. $\Delta/M(f(A, W), A)$ (\ref{['eq:fair_robustness']}). A stands for Age, G for Gender.
  • Figure 2: Trend of the operationalizations of DP across different stages of the perturbation process. Each stage reflects a fraction of perturbed edges w.r.t. $\tilde{E}$, depicted as points gradually larger as the amount of perturbed edges increases. The horizontal dashed line labeled as Orig denotes the DP of the recommendations generated by the non-perturbed system. In the consumer-side results, A stands for Age, G for Gender.
  • Figure 3: Relationship between $\Delta EI$ (y-axis), i.e. the disparity in edge perturbations distribution between the advantaged and the disadvantaged group, and $\Delta$ (x-axis), i.e. the disparity in fairness level before and after the attack. Successful attacks ($\Delta > 0$) are caused by targeting more the advantaged group if $\Delta EI > 0$, otherwise the disadvantaged group was targeted. A stands for Age, G for Gender. Some points overlap.