Download PDFDownload Chat
The paper "Retrieval Augmented Retrieval" introduces an innovative approach to enhance information retrieval in Large Language Models through recursive retrieval augmentation. The authors present a novel framework where retrieved documents are used to find additional relevant in-context examples, creating a more comprehensive context for query processing. The system employs a two-stage retrieval process: first retrieving relevant documents, then using these to guide the selection of appropriate in-context examples. This recursive approach significantly improves retrieval accuracy by providing richer contextual information to the LLM. The research demonstrates substantial performance improvements across various tasks, including question answering, fact verification, and information synthesis. Results show that this method achieves up to 25% better accuracy compared to traditional retrieval-augmented generation approaches. The paper also explores the impact of different retrieval strategies and the optimal number of recursive steps for various applications.
