DuetRAG: Collaborative Retrieval-Augmented Generation
Dian Jiao, Li Cai, Jingsheng Huang, Wenqiao Zhang, Siliang Tang, Yueting Zhuang
TL;DR
DuetRAG tackles the problem of irrelevant or low-quality knowledge retrieval in domain-specific QA by bootstrapping domain knowledge finetuning with retrieval-augmented generation. It introduces Reciter (internal domain knowledge), Discoverer (external knowledge retrieval), and Arbiter (referee) to generate and selectively synthesize answers, with LoRA-based fine-tuning guiding both internal and external components. Evaluation on HotPotQA with LLama-7B shows DuetRAG outperforms 0-shot, RAG, and domain-finetuned baselines, demonstrating robust multi-hop reasoning through internal/external knowledge fusion. The work highlights practical impact for knowledge-intensive QA in specialized domains and points to broader applicability and transferability in future studies.
Abstract
Retrieval-Augmented Generation (RAG) methods augment the input of Large Language Models (LLMs) with relevant retrieved passages, reducing factual errors in knowledge-intensive tasks. However, contemporary RAG approaches suffer from irrelevant knowledge retrieval issues in complex domain questions (e.g., HotPot QA) due to the lack of corresponding domain knowledge, leading to low-quality generations. To address this issue, we propose a novel Collaborative Retrieval-Augmented Generation framework, DuetRAG. Our bootstrapping philosophy is to simultaneously integrate the domain fintuning and RAG models to improve the knowledge retrieval quality, thereby enhancing generation quality. Finally, we demonstrate DuetRAG' s matches with expert human researchers on HotPot QA.