Reproduce “SymbioticRAG: Enhancing Document Intelligence Through Human-LLM Symbiotic Collaboration” on OpenResearch

About this paper

Qiang Sun, Tingting Bi, Sirui Li, Eun-Jung Holden, Paul Duuring, Kai Niu, Wei Liu

SymbioticRAG introduces a two-tier framework for Retrieval-Augmented Generation (RAG) that integrates human judgment in determining content relevance and supports users in refining their information needs. The system demonstrated a reduction in Human-Retriever Distance by up to 33% and increased user satisfaction by up to 70% compared to traditional RAG methods across diverse scenarios.

Key insights

• Human judgment is critical for defining relevance, particularly in complex, domain-specific tasks, and integrating user selection leads to better alignment between retrieved content and user intent.
• Providing direct access to source documents and enabling interactive content curation helps users transition from 'unconscious incompetence' to more precise query formulation, facilitating exploratory learning.
• Bidirectional human-machine interaction, where user actions inform and refine the system's retrieval capabilities, can create adaptive and personalized RAG experiences over time.

Summary by alphaXiv.