After Retrieval, Before Generation: Enhancing the Trustworthiness of Large Language Models in Retrieval-Augmented Generation

TL;DR

This paper introduces BRIDGE, a framework that enhances the trustworthiness of retrieval-augmented generation in large language models by adaptively balancing internal and external knowledge sources, improving response accuracy and reliability.

Contribution

It proposes a unified, dynamic approach with soft bias and decision trees to handle conflicting or unreliable knowledge in RAG, addressing a gap in existing methods.

Findings

BRIDGE outperforms baselines by 5-15% in accuracy.

It maintains balanced performance across various real-world scenarios.

The Trustworthiness Response Dataset (TRD) enables comprehensive analysis.

Abstract

Retrieval-augmented generation (RAG) is a promising paradigm, yet its trustworthiness remains a critical concern. A major vulnerability arises prior to generation: models often fail to balance parametric (internal) and retrieved (external) knowledge, particularly when the two sources conflict or are unreliable. To analyze these scenarios comprehensively, we construct the Trustworthiness Response Dataset (TRD) with 36,266 questions spanning four RAG settings. We reveal that existing approaches address isolated scenarios-prioritizing one knowledge source, naively merging both, or refusing answers-but lack a unified framework to handle different real-world conditions simultaneously. Therefore, we propose the BRIDGE framework, which dynamically determines a comprehensive response strategy of large language models (LLMs). BRIDGE leverages an adaptive weighting mechanism named soft bias to guide knowledge collection, followed by a Maximum Soft-bias Decision Tree to evaluate knowledge and select optimal response strategies (trust internal/external knowledge, or refuse). Experiments show BRIDGE outperforms baselines by 5-15% in accuracy while maintaining balanced performance across all scenarios. Our work provides an effective solution for LLMs' trustworthy responses in real-world RAG applications.