Data Network Security Analysis of Vocational Undergraduate English Teaching AI Platform Based on Federated Learning

Abstract

This paper aims to solve the data privacy protection and network security issues in the vocational undergraduate English teaching AI (Artificial Intelligence) platform. Combining federated learning and differential privacy technology can achieve course recommendations while ensuring the security and confidentiality of student data in a distributed environment. This paper introduces a federated learning framework, which keeps student data on local devices and uses local model training and parameter-sharing mechanisms for distributed learning. On this basis, differential privacy technology is introduced, using Laplace noise to protect model parameters to prevent the model from being reversed and leaking students' sensitive information. To improve the accuracy of personalized recommendations, this paper adopts the Transformer model to perform course recommendations by deep learning the characteristics of students' learning behaviors. Experimental results show that the average accuracy and average F1 value of Transformer course recommendations are 97.2% and 97.3% respectively. The average attack recovery rate of the Laplace noise differential privacy is 29.6%, which can effectively reduce the risk of privacy leakage. Under 10 attack methods, the average attack success rates of federated learning, federated learning without differential privacy, and centralized models in this paper are 7.1%, 38.3%, and 49.0%, respectively. The personalized recommendation model based on federated learning and differential privacy in this paper has made significant progress in privacy protection and recommendation accuracy, has high application value, and can meet the application needs of data network security of vocational undergraduate English teaching AI platforms.