Privacy Preserved Federated Learning for Online Ranking System (OLTR) for 6G Internet Technology

Abstract

With the development of 6G internet technology, advanced and efficient connectivity and data speeds will be possible, opening the way to more advanced and customised online services. But this progress also raises serious privacy issues, especially regarding Online Learning to Rank (OLTR) systems, which are essential for customising user experiences. To overcome the problems, this research presents a new approach called Federated Pairwise Differentiable Gradient Descent (FPDGD), along with an attention mechanism (AM) designed for OLTR systems in the context of 6G technology. FPDGD preserves user privacy by using the decentralised nature of federated learning to facilitate the cooperative training of OLTR models across numerous devices without centralising sensitive user data. The effective involvement of the attention mechanism in the model further improves its capacity to dynamically rank the relative importance of various elements according to user interactions, resulting in more precise and individualised content positioning. Using a noise-adding clipping method based on differential privacy theory, we establish strong privacy guarantees without sacrificing the model’s performance. Our method’s improved ability across different privacy guarantee levels is seen from empirical evaluations, which show a considerable improvement over current federated OLTR techniques. The proposed method is an acceptable choice for real-life applications in the upcoming 6G internet age when user privacy and personalized experiences are critical due to its efficiency and capacity. This work highlights the ability of federated learning combined with attention techniques to improve privacy and performance in the era of 6G technology, in addition to addressing the urgent need for privacy-preserving technologies in OLTR systems.