Network Security of AI-Assisted Teaching Data in Vocational Undergraduate English Education Based on Differential Privacy Protection

Abstract

With the widespread application of AI-assisted teaching in vocational undergraduate English education, the privacy protection and network security issues of teaching data are becoming increasingly prominent. Currently, traditional AI-assisted teaching systems mostly use encryption technology and data anonymization to protect teaching data, making it difficult to cope with complex network attacks. In addition, the protection of different teaching data lacks personalization, resulting in insufficient network security protection capabilities for teaching data. This paper combines a PDP protection mechanism with the DDPG algorithm to adaptively provide privacy budgets according to different teaching data protection requirements to protect AI-assisted teaching data. The study uses Laplace to construct a differential privacy protection mechanism, performs sensitivity calculations in query data, and uses a dynamic privacy allocation method based on multi-task optimization of data sensitivity to allocate privacy budgets. Then, the DDPG algorithm is applied to design the state, action, and reward functions, and the privacy budget in the differential privacy algorithm is adaptively adjusted according to the sensitivity and real-time privacy protection requirements. Finally, the DDPG algorithm and the differential privacy protection mechanism are integrated to form the PDP protection mechanism, which is deployed in the AI-assisted teaching system for experimental verification. The experiment is based on the data in the student management database of a vocational undergraduate college in Nanjing and the school's network security management center database, and conducts verification experiments on test records, interactive behavior data, etc., in terms of network security performance and data availability. Experimental results show that the PDP protection mechanism has a protection success rate of 96.43% on examination record types, an increase of 7.2% compared to DP, and the attack detection accuracy reaches 94.32%. Experimental results show that the PDP protection mechanism can dynamically adjust the privacy budget to ensure data availability, and significantly improve the network security protection performance of AI-assisted teaching data.