Recommender systems are often judged on two axes: how relevant their recommendations are and how fair they are to users or items. Those aspects are commonly measured with separate metrics, but they can also be considered jointly. A new arXiv preprint (2610.27850) announces a robustness analysis of Pareto-based joint evaluation of fairness and relevance.
The abstract notes that recommender systems can be evaluated for relevance and fairness either in isolation, using a measure for each aspect, or jointly with a combined approach. The preprint applies Pareto-based evaluation, a framework that treats trade-offs between conflicting objectives. The announced work examines how robust such Pareto-based joint evaluation is under different conditions.
Because the abstract excerpt is brief, the full methodology and results are not yet available from the listing. The announcement signals continued research into how to reliably combine fairness and accuracy in recommender system evaluation.