PlanRAG: A Plan-then-Retrieval Augmented Generation for Generative Large Language Models as Decision Makers

PlanRAG (Plan-then-Retrieval Augmented Generation) extends traditional RAG techniques by incorporating planning and re-planning steps into the retrieval-augmented generation process. This allows for the generation of an initial plan describing the series of data analyses needed for decision-making, and the iterative refinement of the plan based on query results. PlanRAG's strength lies in its systematic planning and data retrieval, resulting in improved decision-making performance and lower rates of missed critical data analysis2.

Decision QA is a new task proposed by researchers to enable language models to make optimal decisions by analyzing structured data and business rules. It is a QA-style task that takes a database, business rules, and a decision-making question as input and generates the best decision as output. The purpose of Decision QA is to facilitate complex decision-making in real-world business scenarios by leveraging the capabilities of language models.