RAG Questions Need Parsing Too: Turn the User’s String Into Briefs for Retrieval and Generation

To improve the accuracy of RAG systems, a user's question must be parsed into structured data instead of being used as a single raw string. This process breaks down the query into distinct components, identifying the core topic, negative cues, and expected answer format. The parsed information is then strategically split into two separate briefs, one for the retrieval system and another for the generation model. This separation allows the retrieval component to focus solely on finding relevant text, while the generation model receives specific instructions for reasoning and formulating a precise answer.