RELOOP: Recursive Retrieval with Multi-Hop Reasoner and Planners for Heterogeneous QA

Researchers introduce RELOOP, a new retrieval-augmented generation framework that improves multi-step question answering across text, tables, and knowledge graphs. The system uses hierarchical sequences and structure-aware iteration to achieve better accuracy while reducing computational costs compared to existing RAG methods.

• →RELOOP framework unifies handling of documents, tables, and knowledge graphs through a single hierarchical sequence format.
• →The system uses guided iteration with Head and Iteration Agents to collect just-enough evidence before generating answers.
• →Testing shows consistent improvements in exact match and F1 scores across multiple benchmark datasets including HotpotQA and HybridQA.
• →The framework reduces unnecessary computational overhead while maintaining accuracy through budget-aware iteration.
• →Evidence canonicalization feature improves answer consistency and makes the reasoning process more auditable.