LLM-Independent Adaptive RAG: Let the Question Speak for Itself

2025-05-07

AI Summary

To address the high computational overhead and false-positive retrieval issues caused by redundant retrieval in Retrieval-Augmented Generation (RAG), this paper proposes a lightweight, LLM-agnostic adaptive retrieval method. Instead of relying on large language model (LLM)-based uncertainty estimation, our approach dynamically decides whether to trigger retrieval based solely on shallow semantic and structural features of the input question—such as token length, entity density, and interrogative word patterns—combined with interpretable rules and a lightweight classifier. Through systematic analysis of 27 external features and their seven compositional categories, we establish the first LLM-free adaptive retrieval paradigm. Evaluated across six open-domain QA benchmarks, our method achieves question-answering accuracy comparable to state-of-the-art LLM-based adaptive approaches, while reducing inference latency by 62% and decreasing retrieval calls by 58%, thereby significantly enhancing both efficiency and practical deployability.

Technology Category

• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Large Language Models~(LLMs) are prone to hallucinations, and Retrieval-Augmented Generation (RAG) helps mitigate this, but at a high computational cost while risking misinformation. Adaptive retrieval aims to retrieve only when necessary, but existing approaches rely on LLM-based uncertainty estimation, which remain inefficient and impractical. In this study, we introduce lightweight LLM-independent adaptive retrieval methods based on external information. We investigated 27 features, organized into 7 groups, and their hybrid combinations. We evaluated these methods on 6 QA datasets, assessing the QA performance and efficiency. The results show that our approach matches the performance of complex LLM-based methods while achieving significant efficiency gains, demonstrating the potential of external information for adaptive retrieval.

Problem

• Reducing LLM hallucinations without high computational cost
• Eliminating reliance on LLM-based uncertainty estimation
• Improving adaptive retrieval efficiency using external information

Innovation

• Lightweight LLM-independent adaptive retrieval methods
• Utilizes 27 features organized into 7 groups
• Matches LLM-based performance with higher efficiency