Design and implementation of an AI-enabled English teaching support system
Abstract
Traditional English teaching system has some problems, such as insufficient semantic understanding, low efficiency of personalized recommendation and high real-time interaction delay. This paper proposes an adaptive English teaching system based on AI. The system adopts a six layer microservice architecture, integrating rule parsing and deep learning hybrid syntax analysis engine, dynamic knowledge graph (KG), and dual tower neural network recommendation model, combined with Flink event time processing and CUDA parallel computing to optimize end-to-end latency to 10ms level. Experiments show that the F1 value of the hybrid parsing engine on FCE (First Certificate in English) data set is 0.857, which is 57.8% higher than that of the baseline model. Dynamic KG realizes non-monotonic updating of edge weights through error pattern capture, and RMSE (root mean square error) decreases by 55.2%. The cold start hit rate of the recommended model of twin towers is 0.547, which is better than that of LightGCN. The system supports 5500 concurrent requests, the delay of p99 is less than 200ms, and the error rate of non-native speech recognition words is reduced to 9.8%. The research results provide an engineering solution for the high concurrency and low delay requirements of online English teaching scenarios.
