University major information recommendation based on federated learning

Abstract

This paper focuses on the research of applying federated learning to recommendation systems and proposes a university major recommendation method based on federated learning. Furthermore, an improved knowledge distillation architecture is implemented for university major recommendations. In the collaborative structure of the system based on federated learning, knowledge distillation is used to optimize the recommendation performance. The federated learning algorithm, FedDyn, is employed to aggregate model parameters through weighted averaging, enabling a training mode where only local training data and local models are uploaded to the central server. After reading and studying other papers that apply federated learning to recommendation systems, this paper conducts further speculation and research, aiming to apply relevant knowledge and techniques to establish a system that can recommend specific content to a targeted audience, such as students after the college entrance examination. This includes providing major-related information, predicting students major preferences, and delivering the latest industry news related to specific majors.This paper also categorizes the technologies used in the creation of recommendation systems and compares them into three categories. The study suggests that the recommendation system utilizing knowledge distillation will be more efficient.