Performance Evaluation of Query Fusion and Multi-Query in Retrieval-Augmented Generation Systems
Abstract
Retrieval-augmented generation (RAG) systems often suffer from limited retrieval robustness due to the inherent limitations of single-query similarity search. To address this challenge, this paper systematically evaluates two advanced RAG retrieval methods—Query Fusion (leveraging Reciprocal Rank Fusion) and Multi-Query—across multiple datasets, hyperparameter configurations, and question difficulty levels. Experimental results show that both methods outperform the baseline RAG approach, with distinct performance trade-offs: Query Fusion delivers consistent retrieval efficiency, low sensitivity to hyperparameters, and retrieval accuracy across all datasets and question difficulty tiers. In contrast, Multi-Query exhibits higher computational overhead, greater hyperparameter sensitivity, and variable accuracy, though it retains recall potential in specific contexts. Evaluation across difficulty levels reveals that both methods enhance RAG robustness against increasing cognitive complexity, with Query Fusion achieving the most consistent improvements. This work provides actionable insights for designing efficient, robust RAG systems tailored to diverse application requirements.
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.