Evaluating Retrieval-Augmented Generation for Reducing Hallucinations in Large Language Models

Ruthwi Bohra; Tejashri Krishnakumar; Sivakumar V

Evaluating Retrieval-Augmented Generation for Reducing Hallucinations in Large Language Models

Author(s)	Ms. Ruthwi Bohra, Ms. Tejashri Krishnakumar, Sivakumar V
Country	India
Abstract	Large Language Models (LLMs) frequently generate responses that lack experimental support or are factually mistaken, particularly when addressing domain-specific or unindexed queries. Retrieval-Augmented Generation (RAG) diminishes this vulnerability by fetching relevant passages from an external compilation and conditioning generation on the retrieved evidence. This experimental investigation evaluates whether RAG enhances generation accuracy and suppresses hallucinations relative to an unaugmented baseline LLM. We tested both systems across 50 benchmark queries (25 answerable, 25 unanswerable) extracted from the SQuAD 2.0 dataset. While the baseline LLM relied strictly on internal parameters, the RAG architecture indexed external contexts via all-MiniLM-L6-v2 embeddings and FAISS nearest-neighbour search before answer generation with Gemini 1.5 Flash. Comparative evaluation employed Exact Match, token F1 Score, Semantic Similarity, and Hallucination Rate, with RAG additionally appraised for Faithfulness and Retrieval Relevance. Experimental outcomes indicate that RAG increased F1 Score from 3.34% to 33.16% and Semantic Similarity from 38.19% to 57.46%, while Exact Match rose from 0.00% to 8.00%. More importantly, RAG reduced the hallucination rate on unanswerable queries from 100.00% down to 36.00%. Average Faithfulness and Retrieval Relevance scores reached 48.34% and 53.00%, respectively, highlighting that retrieval accuracy remains the primary constraint in grounded synthesis.
Keywords	Retrieval-Augmented Generation, Large Language Models, Hallucination Reduction, Natural Language Processing, Question Answering.
Field	Computer > Artificial Intelligence / Simulation / Virtual Reality
Published In	Volume 8, Issue 5, September-October 2026
Published On	2026-10-16

View / Download PDF File

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Evaluating Retrieval-Augmented Generation for Reducing Hallucinations in Large Language Models

Share this