Reproduce “Reasoning in Trees: Improving Retrieval-Augmented Generation for Multi-Hop Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuling Shi, Maolin Sun, Zijun Liu, Mo Yang, Yixiong Fang, Tianran Sun, Xiaodong Gu
The RT-RAG framework enhances multi-hop question answering in Retrieval-Augmented Generation by creating a hierarchical reasoning tree structure prior to retrieval. This approach improved average F1 scores by 7.0% and Exact Match scores by 6.0% over leading baselines on challenging benchmarks using GPT-4o-mini.
Key insights
- Pre-retrieval hierarchical tree-structured reasoning, validated by consensus-based selection, effectively mitigates query decomposition inaccuracies and error propagation in multi-hop QA.
- An adaptive bottom-up traversal strategy significantly improves robustness when encountering retrieval failures or inconsistent evidence by dynamically restructuring reasoning paths.
- Core components like query rewriting, rejection sampling, and consensus-based tree selection are crucial, each contributing measurably to overall system performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.