This is an unedited manuscript accepted for publication and provided as an Article in Press for early access at the author’s request. The article will undergo copyediting, typesetting, and galley proof review before final publication. Please be aware that errors may be identified during production that could affect the content. All legal disclaimers of the journal apply.
Preeti Kumari,
- Assistant Professor, Dr. Akhlesh Das Gupta Institute of Professional Studies, Delhi, India
Abstract
The rapid proliferation of online educational content has created an unprecedented information overload challenge for learners seeking structured, relevant, and personalized guidance. This paper presents RecomGPT, a lightweight yet robust Retrieval-Augmented Generation (RAG)-based educational recommendation system designed to provide highly personalized learning resource suggestions across three major modalities: online courses, educational videos, and PDF documents. RecomGPT integrates a carefully curated domain-specific knowledge base with a hybrid retrieval engine that combines semantic and keyword-based matching to identify resources that closely align with user queries and learning requirements. The system further incorporates multi-stage prompt engineering and the Google Gemini large language model (LLM) to generate exactly three ranked recommendations for each user query. To improve recommendation relevance and practical usability, the framework supports pricing-aware course filtering, allowing users to distinguish between free and paid learning resources. In addition, semantic query matching enables the system to understand contextual and conceptual relationships beyond simple keyword overlap, while automated output-format validation ensures consistency and reliability in generated recommendations. Experimental evaluation against established baseline approaches demonstrates that RecomGPT achieves a Precision@3 of 0.91, significantly outperforming conventional keyword-based and content-filtering recommendation methods. The complete architecture is implemented and deployed as a Flask-based web application with a responsive and user-friendly frontend, enabling learners to interact with the recommendation system conveniently across different devices. The proposed system demonstrates the potential of combining RAG, hybrid information retrieval, and generative AI to address educational information overload and deliver efficient, personalized, and accessible learning recommendations for diverse user needs.
Keywords: Abstract— The rapid proliferation of online educational content has created an unprecedented information overload challenge for learners seeking structured, relevant, and personalized guidance. This paper presents RecomGPT, a lightweight yet robust Retrieval-Augmented Generation (RAG)-based educational recommendation system designed to provide highly personalized learning resource suggestions across three major modalities: online courses, educational videos, and PDF documents. RecomGPT integrates a carefully curated domain-specific knowledge base with a hybrid retrieval engine that combines semantic and keyword-based matching to identify resources that closely align with user queries and learning requirements. The system further incorporates multi-stage prompt engineering and the Google Gemini large language model (LLM) to generate exactly three ranked recommendations for each user query. To improve recommendation relevance and practical usability, the framework supports pricing-aware course filtering, allowing users to distinguish between free and paid learning resources. In addition, semantic query matching enables the system to understand contextual and conceptual relationships beyond simple keyword overlap, while automated output-format validation ensures consistency and reliability in generated recommendations. Experimental evaluation against established baseline approaches demonstrates that RecomGPT achieves a Precision@3 of 0.91, significantly outperforming conventional keyword-based and content-filtering recommendation methods. The complete architecture is implemented and deployed as a Flask-based web application with a responsive and user-friendly frontend, enabling learners to interact with the recommendation system conveniently across different devices. The proposed system demonstrates the potential of combining RAG, hybrid information retrieval, and generative AI to address educational information overload and deliver efficient, personalized, and accessible learning recommendations for diverse user needs.
References
1] C. Romero and S. Ventura, “Educational data mining: A survey from 1995 to 2005,” Expert Systems with Applications, vol. 33, no. 1, pp. 135–146, 2007.
[2] F. Ricci, L. Rokach, and B. Shapira, Recommender Systems Handbook, 2nd ed. New York, NY, USA: Springer, 2015.
[3] M. D. Ekstrand, J. T. Riedl, and J. A. Konstan, “Collaborative filtering recommender systems,” Foundations and Trends in Human–Computer Interaction, vol. 4, no. 2, pp. 81–173, 2011.
[4] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Proc. NeurIPS, 2020, pp. 9459–9474.
[5] H. Drachsler, H. G. K. Hummel, and R. Koper, “Personal recommender systems for learners in lifelong learning networks,” Int. J. Learning Technology, vol. 3, no. 4, pp. 404–423, 2008.
[6] L. Gao, Z. Ma, J. Lin, and J. Callan, “Precise zero-shot dense retrieval without relevance labels,” in Proc. ACL, 2023, pp. 1762–1777.
[7] Y. Hou et al., “Large language models are zero-shot rankers for recommender systems,” in Proc. ECIR, 2024, pp. 364–381.
[8] S. E. Robertson and H. Zaragoza, “The probabilistic relevance framework: BM25 and beyond,” Foundations and Trends in Information Retrieval, vol. 3, no. 4, pp. 333–389, 2009.
[9] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence embeddings using Siamese BERT-networks,” in Proc. EMNLP, 2019, pp. 3982–3992.
Emerging Trends in Symmetry
| Volume | 02 | |
| 02 | ||
| Received | 03/07/2026 | |
| Accepted | 26/09/2026 | |
| Published | 08/10/2026 | |
| Publication Time | 97 Days |