DRAGIN: Dynamic Retrieval Augmented Generation based on the Information Needs of Large Language Models

DRAGIN (Dynamic Retrieval Augmented Generation based on Information Needs) enhances traditional Retrieval Augmented Generation methods by enabling multi-step and context-aware information retrieval for large language models. It addresses issues like hallucinations, outdated information, and the need for proprietary knowledge by dynamically retrieving data based on real-time information needs. This dynamic framework involves a two-step process: RIND (Real-time Information Needs Detection) to determine when to retrieve information, and QFS (Query Formulation based on Self-Attention) to determine what to retrieve, resulting in more accurate and contextually relevant responses.