Retrieval-Augmented Generation for Improving Factual Reliability in AI-Based Question Answering Systems

Noah Anderson

Retrieval-Augmented Generation for Improving Factual Reliability in AI-Based Question Answering Systems

Author(s)	Noah Anderson
Country	United States
Abstract	Artificial intelligence-based question answering systems can generate fluent responses across broad knowledge domains, but linguistic fluency does not necessarily guarantee factual reliability. Generative language models may produce unsupported statements, rely on outdated parametric knowledge, or respond confidently when relevant evidence is unavailable. Retrieval-Augmented Generation provides a methodological approach for addressing this limitation by retrieving external documents and conditioning answer generation on selected evidence. The effectiveness of this approach, however, depends on retrieval quality, evidence relevance, and the ability of the generator to use retrieved material faithfully. This hypothetical methodological study evaluates Retrieval-Augmented Generation for improving factual reliability in knowledge-intensive question answering. The proposed framework compares three conceptual configurations: a parametric-only generative question-answering model, a sparse-retrieval RAG system, and a dense-retrieval RAG system. A fictional corpus of 20,000 knowledge passages supports a hypothetical set of 1,000 question-answer pairs divided into training, validation, and testing partitions. The illustrative analysis evaluates exact-match accuracy, token-level F1 score, evidence-supported answer rate, and unsupported-claim rate. Under the assigned hypothetical values, the dense-retrieval RAG configuration achieves the strongest overall question-answering and factual-support performance. A second analysis examines performance across factoid, multi-evidence, temporally sensitive, and ambiguous questions. The assigned results show that factual reliability varies according to question structure and retrieval difficulty rather than depending solely on the underlying generator. The study concludes that retrieval augmentation provides a useful framework for grounding generated answers in external evidence, but retrieval does not automatically eliminate hallucination or factual error. Incorrect, incomplete, outdated, or weakly matched passages can themselves become sources of unreliable generation. Future empirical research should therefore evaluate retrieval recall, evidence attribution, answer factuality, source quality, and uncertainty jointly rather than relying exclusively on conventional answer-accuracy metrics.
Keywords	Retrieval-Augmented Generation, Question Answering, Factual Reliability, Knowledge Retrieval, Dense Passage Retrieval, Large Language Models, Hallucination Reduction, Evidence Grounding, Natural Language Processing, Artificial Intelligence.
Field	Engineering
Published In	Volume 2, Issue 1, January-February 2026
Published On	2026-01-11

View / Download PDF File

E-ISSN: 3067-5073

doi

CrossRef DOI is assigned to each research paper published in our journal.

IJMLAIDSE DOI prefix is
10.00000/IJMLAIDSE

Downloads

Research Paper Format Copyright Permission and Undertaking Form

All research papers published on this website are licensed under Creative Commons Attribution-ShareAlike 4.0 International License, and all rights belong to their respective authors/researchers.

CC-BY-SA

About IJMLAIDSE 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 Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us	+1-872-302-3851	editor@uscse.com

International Journal of Machine Learning, AI & Data Science Evolution

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Monthly Scholarly International Journal

Retrieval-Augmented Generation for Improving Factual Reliability in AI-Based Question Answering Systems

Share this