A Comparative Evaluation of Dense, Sparse, and Hybrid Retrieval for Improving Retrieval-Augmented Generation Reliability
Abstract
Abstract - Grounding large language model (LLM) outputs in external knowledge, and thereby curbing hallucination, is commonly achieved through Retrieval-Augmented Generation (RAG). Yet the retrieval stage of a RAG pipeline is itself often a weak link: dense retrievers capture semantic similarity but can miss exact lexical matches; sparse retrievers capture exact lexical matches but can miss semantic similarity; and hybrid retrievers combine both, though their effects on the reliability of generated outputs have received comparatively less evaluation. In this paper, we propose a comprehensive evaluation of dense, sparse and hybrid retrieval strategies within a unified retrieval and generation (RAG) pipeline, with a specific focus on how these retrieval methods affect generation reliability (faithfulness, hallucination rate, and answer relevance). We build a retriever-swappable RAG pipeline backed by a Llama-3-8B-Instruct generator and evaluate it on 300 held-out queries drawn from a technical support and product-documentation corpus. We compare four retrieval configurations—dense-only, sparse-only (BM25), static hybrid, and dynamic hybrid—using retrieval accuracy and generation-reliability metrics. Our results demonstrate that static hybrid retrieval achieves an 11-percentage-point increase in faithfulness over dense-only retrieval, and a 7-percentage-point decrease in the hallucination rate over sparse-only retrieval. An ablation study further examines the contributions of…
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.