OptiRAG: A Feedback-Driven Approach to Improving Retrieval-Augmented Generation

View at publisher Summarise with VIRAT
Authors: A. Akhil, Sasi Rekha Sankar
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by integrating external knowledge sources to mitigate hallucinations. However, current RAG systems often struggle to align retrieved content with contextual needs, leading to irrelevant or redundant data. This paper proposes OptiRAG, a dynamically optimized RAG system that improves retrieval precision and contextual understanding for more coherent responses. OptiRAG includes a Contextualized Awareness Layer to classify question intent, a Rapid Pre- Ranking Mechanism for efficient relevance scoring, a Dynamic Retrieval Pipeline with feedback-driven refinement, and a Selective Memory Integration module to prioritize high-utility knowledge. Additionally, a Feedback-Driven Optimizer continuously updates retrieval strategies based on performance feedback. Experimental results on four QA benchmarks—NaturalQuestions, TriviaQA, HotpotQA, and WebQuestionsSP—show that OptiRAG outperforms both traditional RAG models and recent state-of-the-art methods. It delivers up to a 20% improvement in accuracy, reduces irrelevant retrieval by 40%, and improves iterative learning efficiency by 25%. These findings underscore the value of dynamic optimization and strategic integration in RAG systems, offering a pathway toward more reliable and context-aware LLMs.

Fields of Study & Subjects

Topic Modeling Information Retrieval and Search Behavior Multimodal Machine Learning Applications

Publication details

Journal / source
CRC Press eBooks
Pages
46–50
Publication date
31 March 2026
Language
EN
Document type
Book chapter