The steady growth of online education in Chinese higher education, despite its drawbacks, underscores an increasing interest in digital learning platforms. This study proposes enhancements to the Paragraph Vector, Stacking classifiers, and ZEN semantic coders to construct accurate learner profiles, facilitating personalized online education. By applying these improved models, universities can adapt their online courses more effectively, as demonstrated by case study analyses. Performance data for a sample of college students shows a concentration of scores between 80 and 90, with an average of 84.01, suggesting moderate academic achievement. Engagement metrics, particularly in educational experience richness, scored highest at 3.7854, with all engagement scores averaging above 3, indicating a strong engagement in the learning process.