Construction of a distributed English proficiency assessment framework based on federated learning

This paper proposes a distributed English proficiency assessment framework based on federated learning, aiming to achieve efficient English proficiency assessment under data privacy protection. The framework integrates English learning data of multiple clients into local training through distributed architecture design, uses differential privacy and homomorphic encryption technology to ensure data security, and realizes global model aggregation through the federated average algorithm. Experimental results show that the framework achieves an evaluation accuracy of 0.82, which is only about 3% lower than centralized training. At the same time, the privacy parameter ε is 1.0, which meets the security requirements of distributed scenarios. Training efficiency analysis shows that as the number of clients increases from 2 to 12, the average training time per client drops to 13.75 seconds, reflecting good scalability. The study verifies the comprehensive performance of the framework in privacy protection, evaluation accuracy and distributed efficiency, and provides a feasible solution for personalized English proficiency assessment in the field of education.