Gangotri Ravinder Balla, Prea Binu Varghese, Rutuja Suresh Farkande, Ms.Yogeshwari Yawalkar- DYPIMCAM Akurdi Pune.

DOI: https://doi.org/10.5281/zenodo.23391270

Large Language Models (LLMs) have become widely used for question answering, education, research, summarization and information support. Despite their usefulness, LLMs may generate false, unsupported or inconsistent information, commonly described as hallucination. Retrieval-Augmented Generation (RAG) addresses this reliability problem by retrieving relevant information from external knowledge sources and supplying it to the language model as contextual evidence before generating a response. This paper studies the role of RAG in reducing factual hallucinations through a review of existing research and a 50 question pilot comparison of responses generated with and without retrieval support. The submitted report records 49 of 50 responses as correct without RAG and all 50 as correct with RAG; it separately presents a clearly labelled illustrative distribution of 46/3/1 versus 48/2/0 for demonstrating response categories. The distinction is retained to avoid presenting illustrative values as observed experimental results. The study also examines limitations including poor retrieval, unreliable sources, incomplete context and incorrect interpretation of retrieved evidence. The paper concludes that RAG is a practical hallucination-reduction approach, but it should be combined with high-quality retrieval, evidence verification and human oversight for critical applications.

Large Language Models; Hallucination; Retrieval-Augmented Generation; Factuality; Information Retrieval; Generative AI

Large Language Model hallucination is a significant challenge for reliable generative artificial intelligence because fluent responses may still contain unsupported or incorrect information. Retrieval-Augmented Generation provides a practical approach by connecting an LLM with external information and using retrieved evidence as context during generation. The reviewed literature demonstrates that retrieval can improve factual grounding, while newer research emphasizes adaptive retrieval, corrective retrieval, citation support and explicit verification. The submitted pilot records 49 correct responses without RAG and 50 correct responses with RAG, while its illustrative analysis demonstrates how a distribution of 46/3/1 versus 48/2/0 can be visualized. Because the dataset is small and relatively simple, neither presentation should be interpreted as proof that RAG eliminates hallucination. Overall, RAG is best understood as a practical hallucination-reduction technique whose effectiveness depends on source quality, retrieval relevance and correct use of evidence.

1. Guu, K., Lee, K., Tung, Z., Pasupat, P., & Chang, M. (2020). REALM: Retrieval-Augmented Language Model Pre-Training. Proceedings of ICML.
2. Lewis, P., Perez, E., Piktus, A., et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems.
3. Gao, T., Yen, H., Yu, J., & Chen, D. (2023). Enabling Large Language Models to Generate Text with Citations. Proceedings of EMNLP.
4. Es, S., James, J., Espinosa-Anke, L., & Schockaert, S. (2023). RAGAS: Automated Evaluation of Retrieval Augmented Generation. arXiv:2309.15217.
5. Gao, Y., Xiong, Y., Gao, X., et al. (2023). Retrieval-Augmented Generation for Large Language Models: A Survey. arXiv:2312.10997.
6. Asai, A., Wu, Z., Wang, Y., Sil, A., & Hajishirzi, H. (2024). Self RAG: Learning to Retrieve, Generate, and Critique through Self Reflection. ICLR 2024.
7. Niu, C., Wu, Y., Zhu, J., Xu, S., Shum, K., Zhong, R., Song, J., & Zhang, T. (2024). RAGTruth: A Hallucination Corpus for Developing Trustworthy Retrieval-Augmented Language Models. Proceedings of ACL 2024.
8. Vu, T., Iyyer, M., Wang, X., et al. (2024). FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation. Findings of ACL 2024

Gangotri Ravinder Balla, Prea Binu Varghese, Rutuja Suresh Farkande, Ms.Yogeshwari Yawalkar- DYPIMCAM Akurdi Pune. (2026). A Systematic Review of Hallucination Mitigation in Large Language Models Using Retrieval-Augmented Generation. International Journal of Computer Techniques, 13(5), 470–474. ISSN: 2394-2231. DOI: https://doi.org/10.5281/zenodo.23391270