Research map / Large language models / Language, safety and interpretability
Compress-then-Rank: Faster and Better Listwise Reranking with Large Language Models via Ranking-Aware Passage Compression
Zhewei Zhi, Yingyi Zhang, Yizhen Jing, Xianneng Li, Jianing Liu, Huajie Liu, Yongliang Ding
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 |
|---|---|---|---|
| compress passages to 8 vectors instead of full text | RankMistral full-text · TREC DL19 | nDCG@10: 0.7196 -> 0.7905 | Improved |
| compress passages to 8 vectors instead of full text | RankMistral full-text · TREC DL19 | processed tokens: 9635.1 -> 800.0 | Improved |
| use C2R vs ListT5-3B | BEIR per-dataset · Covid | nDCG@10: 0.8470 -> 0.8217 | Mixed |
| add compressor pre-training (ablation) | C2R · SciFact | nDCG@10: 0.6529 -> 0.7276 | Improved |
| add SFT stage 1 (coarse alignment) (ablation) | C2R · TREC DL20 | nDCG@10: 0.5516 -> 0.7791 | Improved |
| add SFT stage 2 (fine refinement) (ablation) | C2R · TREC DL20 | nDCG@10: 0.5973 -> 0.7791 | Improved |
| remove SFT entirely (ablation) | C2R · DL20, News, Robust04, SciFact | nDCG@10 | Hurt |
| reduce compressed length from 32 to 8 (ablation) | C2R · TREC DL19 | nDCG@10: 0.7382 -> 0.7905 | Improved |
| reduce compressed length from 32 to 8 (ablation) | C2R · News | nDCG@10: 0.4990 -> 0.5623 | Improved |
Closest papers by meaning
- AcuRank: Uncertainty-Aware Adaptive Computation for Listwise Reranking (NeurIPS 2025 · similarity 0.91)
- Contrastive Retrieval Heads for Improved Attention-Based Reranking (NeurIPS 2026 · similarity 0.90)
- Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankers (ICLR 2025 · similarity 0.90)
- Very Efficient Listwise Multimodal Reranking for Long Documents (ICML 2026 · similarity 0.89)
- UniRank: Unified List-wise Reranking via Confidence-Ordered Denoising (NeurIPS 2026 · similarity 0.89)
- Language Ranker: A Lightweight Ranking framework for LLM Decoding (NeurIPS 2025 · similarity 0.87)
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.
