Explore domain-specific reflective tokens for Self-BioRAG

Investigate alternative reflective tokens tailored to biomedical and clinical domains for use within the Self-BioRAG framework, beyond the RET (retrieval), REL (relevance), SUP (supportiveness), and USE (utility) tokens adapted from Self-RAG, in order to enhance domain-specific retrieval decisions, evidence assessment, and self-reflection during generation.

Background

Self-BioRAG adopts four reflective tokens (RET, REL, SUP, USE) from Self-RAG to enable on-demand retrieval and self-reflection during biomedical generation. The authors note that while these tokens have been effective, they may not be fully optimal for all domain-specific settings.

They explicitly defer the design and examination of other reflective tokens suited to specific domains, indicating a concrete direction to extend the reflective control mechanisms in Self-BioRAG.

References

Exploring other reflective tokens suitable for specific domains is left for future work.

— Improving Medical Reasoning through Retrieval and Self-Reflection with Retrieval-Augmented Large Language Models  (2401.15269 - Jeong et al., 2024) in Self-Reflection Language Model (Critic Language Model), Data Construction of Critic LM C