REAP: Enhancing RAG with Recursive Evaluation and Adaptive Planning for Multi-Hop Question Answering
REAP enhances Retrieval-Augmented Generation for multi-hop questions by using Recursive Evaluation and Adaptive Planning. It replaces linear reasoning with a dynamic loop between a Sub-task Planner (SP) and a Fact Extractor (FE). The SP maintains a global plan, while the FE extracts structured facts from retrieved content. This iterative process allows the system to adapt its strategy based on retrieved information, significantly improving accuracy in complex reasoning tasks.