Personalised knowledge recommendation system for English teaching based on MoE-RAG algorithm

This paper proposes a personalised English teaching knowledge recommendation system based on the MoE-RAG algorithm.The system integrates a mixture of experts (MoE) architecture with eight specialised sub-models and a sparse gated network to dynamically select the most relevant experts for each query.Combined with a retrieval-augmented generation (RAG) module, it retrieves relevant knowledge from multiple bases and fuses it with expert output via a transformer-based generator to produce personalised recommendations.This approach effectively addresses cold-start issues and enhances interpretability.Experiments on 10,000 interaction records from 500 students show that MoE-RAG significantly outperforms traditional models (e.g., collaborative filtering), achieving 87.5% accuracy, 90.2% precision, 84.1% recall, and an 87.0%F1-score.Through a real-time feedback and reinforcement learning mechanism, the system dynamically adjusts resources and optimises learning paths, demonstrating strong adaptability across different learning stages and improving student engagement, time optimisation, and satisfaction.This system promotes the intelligent, personalised development of English education.