Personalized Content Generation Using Bayesian User Modeling and Graph-Enhanced Retrieval-Augmented Generation

This paper presented a comprehensive system for real-time personalized learning video generation that integrates Bayesian user modeling, knowledge graph embeddings, and multi- modal retrieval-augmented generation. Our approach addresses critical limitations in existing educational AI systems: lack of deep personalization, inefficient generation pipelines, and inadequate pedagogical grounding. Experimental results with 150 learners demonstrate signif- icant improvements in learning outcomes (39.4% gain), en- gagement (87.3% completion rate), and user satisfaction (4.6/5 rating). The system achieves 2.65x real-time generation speed while maintaining 92.3% content quality, making personalized video content practical for on-demand learning. The Bayesian modeling framework successfully captures evolving learner preferences, converging to 96.8% accuracy within 50 interactions. TransE-based knowledge graph em- beddings enable semantic reasoning over educational con- tent, ensuring prerequisite coverage and conceptual coherence. Multi-modal RAG adapts content retrieval to individual input modality preferences. Comparative analysis reveals 40% improvement in personal- ization depth and 35% improvement in content relevance over existing platforms. The system’s comprehensive integration of user modeling, knowledge representation, and content gener- ation represents a significant advance in intelligent tutoring systems. Future work will expand interaction modalities, enhance metacognitive scaffolding, leverage multimodal LLMs, and conduct longitudinal evaluation studies. As AI capabilities continue advancing, personalized educational content gener- ation promises to democratize access to high-quality, adaptive learning experiences tailored to individual needs. The implications extend beyond formal education to cor- porate training, skills development, and lifelong learning. By bridging the gap between AI generative capabilities and pedagogically sound instructional design, this work contributes to the ongoing transformation of education in the digital age.

[1]     D. Zagulova, R. Prokopyev, and V. Ivanov, “Personalization of distance learning: Using PLS-SEM for evaluating FSLSM model,” Procedia Computer Science, vol. 225, pp. 1459–1468, 2023. [2]     M. Khaled, “Learning styles, personalization, and learning management systems: Towards a student-centred LMS approach,” Master’s thesis, Uppsala University, Sweden, 2021. [3]     M. Murtaza et al., “AI-based personalized e-learning systems: Issues, challenges, and solutions,” IEEE Access, vol. 10, pp. 81323–81342, 2022. [4]     K. I. Vorobyeva and A. Martinez, “Personalized learning through AI: Pedagogical approaches and critical insights,” Contemporary Educa- tional Technology, vol. 17, no. 1, 2025. [5]     S. S. Jahagirdar, J. Saha, and C. V. Jawahar, “Exploring prompts for understanding long-form multimodal videos,” in Proc. CVPR Workshop on Multimodal Learning, 2025. [6]     R. Sharma and K. Patel, “AI-powered personalized video tutoring systems for K-12 education: A comprehensive survey,” International Journal of Creative Research Thoughts, vol. 13, no. 1, pp. 201–215, 2025. [7]     M. E. ElAlami, S. M. Khater, and M. El. R. Rehan, “AI-based system for transforming text and sound to educational videos,” Fusion: Practice and Applications, vol. 21, no. 1, pp. 201–213, 2024. [8]     M. Abdelmagied et al., “Leveraging graph retrieval-augmented gen- eration to support learners’ understanding of knowledge concepts in MOOCs,” in Proc. EMOOCs, 2025. [9]     S. Chen and J. Wang, “AcademicRAG: Knowledge graph enhanced retrieval-augmented generation for academic contexts,” Master’s thesis, KTH Royal Institute of Technology, Sweden, 2025. [10]    R. Rim, M. A. Maalej, and A. Mahfoudhi, “Bayesian learning for user modeling,” Research in Computing Science, vol. 152, no. 12, pp. 153– 163, 2023.