Research map / Large language models / Retrieval and knowledge
GRO-RAG: Gradient-aware Re-rank Optimization for Multi-source Retrieval-Augmented Generation
Siyuan Chen, Hang Ding, Kangxiaoyu, Jiechao Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add gradient-aware re-ranking (ablation) | GRO-RAG (GLM-4) · HotpotQA | F1: 37.6 -> 42.8 | Improved |
| add gradient-aware re-ranking (ablation) | GRO-RAG (Llama3.1-8B) · MuSiQue | F1: 16.2 -> 18.6 | Improved |
| add source combination selection (ablation) | GRO-RAG (GLM-4) · 2WikiMultihopQA | F1: 28.6 -> 30.3 | Improved |
| add source combination selection (ablation) | GRO-RAG (Llama3.1-8B) · MuSiQue | EM: 10.2 -> 10.3 | No clear effect |
| use both local and web sources instead of local only | Vanilla RAG (Llama3.1-8B) · HotpotQA | F1: 34.2 -> 36.0 | Improved |
| use web source only instead of local | Vanilla RAG (Llama3.1-8B) · HotpotQA | F1: 34.2 -> 31.5 | Hurt |
| use GRO-RAG instead of CRAG | Llama3.1-8B · HotpotQA | F1: 34.2 -> 39.1 | Improved |
| use GRO-RAG instead of FLARE | Llama3.1-8B · 2WikiMultihopQA | EM: 23.0 -> 22.8 | Mixed |
| replace dense re-ranker with gradient re-ranking | GRO-RAG (Llama3.1-8B) vs BGE-M3 · MuSiQue | nDCG@10: 0.3922 -> 0.4039 | Improved |
| replace E5-base with gradient re-ranking | GRO-RAG (Llama3.1-8B) vs E5-base · HotpotQA | nDCG@10: 0.7013 -> 0.6442 | Hurt |
| replace BM25 with gradient re-ranking | GRO-RAG (Llama3.1-8B) vs BM25 · HotpotQA | nDCG@10: 0.6237 -> 0.6442 | Improved |
| add multi-step gradient optimization | GRO-RAG · HotpotQA | nDCG@10: total gain around 3-5% | Improved |
Closest papers by meaning
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.89)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.89)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.88)
- Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection (ICLR 2025 · similarity 0.88)
- DynamicRAG: Leveraging Outputs of Large Language Model as Feedback for Dynamic Reranking in Retrieval-Augmented Generation (NeurIPS 2025 · similarity 0.88)
- UDA: A Benchmark Suite for Retrieval Augmented Generation in Real-World Document Analysis (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
