COPF: An Online Framework for Deployment-Stable Counterfactual Fairness in Evolving Graphs
Not on the map. Why?
Abstract (arXiv)
Online link recommendation on evolving graphs is performative: by choosing which candidate links to show users, the system changes which links form and what feedback it later observes. Consequently, fairness estimates from logged outcomes can be misleading and may drift after deployment when the recommendation policy is updated. We introduce COPF (Counterfactual Online Performative Fairness), a decision-layer framework for deployment-stable fairness monitoring and control in online link recommendation. COPF (i) defines group-level opportunity gaps over exposure (shown vs. not shown) counterfactuals, (ii) makes them estimable by explicit exploration and by logging the probability (propensity) that each candidate is shown, and (iii) audits and controls fairness using residual outcome indistinguishability (OI) over a configurable auditor family with graph-aware doubly robust (GA-DR) estimators. We provide a noisy transfer theorem showing that Residual-OI on estimated GA-DR residuals implies bounds on exposure-counterfactual group gaps under temporal mixing and bounded local interference, and we instantiate an online multicalibration auditor together with a primal-dual controller. Experiments on two TGB streams and a controlled synthetic bipartite stream show that COPF reduces worst-case spikes in exposure-counterfactual group disparities with modest impact on ranking utility. Our code is available at https://github.com/lsnnnnnnnn/COPF.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 26 of its 43 references that are in the map. First posted on arXiv 30 May 2026.
Builds on
26 of 43 references.
References · 26
- 2020Temporal Graph Networks for Deep Learning on Dynamic Graphscited by 521
- 2020Performative Predictioncited by 318
- 2018Delayed Impact of Fair Machine Learningcited by 183
- 2020Say No to the Discrimination: Learning Fair Graph Neural Networks with Limited Sensitive Attribute Informationcited by 168
- 2023Temporal Graph Benchmark for Machine Learning on Temporal Graphscited by 137
- 2022Towards Better Evaluation for Dynamic Link Predictioncited by 120
- 2017How Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utilitycited by 116
- 2017Runaway Feedback Loops in Predictive Policingcited by 97
- 2020Stochastic Optimization for Performative Predictioncited by 92
- 2020Performative Prediction in a Stateful Worldcited by 90
- 2020Outcome Indistinguishabilitycited by 74
- 2021FairDrop: Biased Edge Dropout for Enhancing Fairness in Graph Representation Learningcited by 68
14 more
- 2019Counterfactual Risk Assessments, Evaluation, and Fairnesscited by 59
- 2022Learning Fair Node Representations with Graph Counterfactual Fairnesscited by 58
- 2023Performative Prediction: Past and Futurecited by 52
- 2021CrossWalk: Fairness-enhanced Node Representation Learningcited by 34
- 2015Counterfactual Risk Minimization: Learning from Logged Bandit Feedbackcited by 33
- 2022Fairness in Recommender Systems: Research Landscape and Future Directionscited by 30
- 2023Towards Fair Graph Neural Networks via Graph Counterfactualcited by 23
- 2024PageRank Bandits for Link Predictioncited by 14
- 2024From Fairness to Infinity: Outcome-Indistinguishable (Omni)Prediction in Evolving Graphscited by 13
- 2022Fair When Trained, Unfair When Deployed: Observable Fairness Measures are Unstable in Performative Prediction Settingscited by 11
- 2022Link recommendations: Their impact on network structure and minoritiescited by 6
- 2023Consistent End-to-End Estimation for Counterfactual Fairnesscited by 4
- 2024A New Paradigm for Counterfactual Reasoning in Fairness and Recoursecited by 3
- 2024Recommendation Fairness in Social Networks Over Timecited by 3
Built on by
Not cited yet.