LambdaFair: a Fair and Effective LambdaMART

We present LambdaFair, an in-processing method that jointly optimizes NDCG (effectiveness) and rND (fairness) to enhance both effectiveness and fairness in ranking. We design three variants that prioritize fairness, effectiveness, or a balanced solution. Experiments on publicly available datasets show that LambdaFair improves statistical parity while preserving competitive ranking quality without compromising training efficiency.