Mitigating sensitive data exposure with adversarial learning for fairness recommendation systems