Research on a personalized English reading recommendation system combining expert mixture model and NoPE

Abstract

Currently, personalized English reading recommendation systems generally face challenges, including difficulty in accurately capturing users’ implicit preferences, limited recommendation effectiveness due to data sparsity, insufficient alignment between recommended content and learning needs, and difficulty in adapting to dynamic preference changes. There is an urgent need for more intelligent and refined recommendation mechanisms. The expert mixture model makes professional recommendation decisions by having multiple expert sub models collaboratively handle multidimensional tasks such as reading difficulty, theme content, discourse structure, and learning goals. The gating network’s attention-enhancement mechanism improves expert selection accuracy, addressing the problems of expert redundancy and ineffectiveness in traditional MoE (Mixture of Experts). NoPE (Neural Preference Embedding) uses deep learning to vectorize implicit associations in user behavior sequences, mining real preferences from weak feedback scenarios, and effectively addressing the distortion in preference expression caused by collaborative filtering based on explicit ratings. Further analysis shows that the system’s recall rate for highly relevant content is 81.23%, and the average recommendation response time is only 0.9 s, a significant improvement over the 56.7% recall rate of traditional collaborative filtering models. Users read an average of 44.89 recommended articles per day, and 72.1% of users reported a significant improvement in reading efficiency.