Research map / Large language models / Retrieval and knowledge
A Unified Pretraining Framework for Passage Ranking and Expansion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use unified UED for passage expansion and re-ranking rather than DeepCT + TFR-BERT | MS MARCO passage retrieval · MS MARCO passage retrieval leaderboard | MRR@10: 40.1† / 42.1 -> 42.4 / 43.6 | Helped |
| Use unified UED for passage expansion and re-ranking rather than BERT Large + T5 | TREC 2019 Deep Learning Track · TREC 2019 Deep Learning Track | NDCG@10: 74.9† -> 77.2 | Helped |
| Use unified UED for passage expansion and re-ranking rather than BERTter Indexing (+RM3) | TREC 2019 Deep Learning Track · TREC 2019 Deep Learning Track | MAP: 50.5† -> 52.6 | Helped |
| Add document expansion (ablation) | UED full ranking framework · MS MARCO | Recall@1000: 86.2 -> 94.0 | Helped |
| Add document expansion (ablation) | UED full ranking framework · MS MARCO | MRR@10: 39.5 -> 42.4 | Helped |
| Add final re-ranking (ablation) | UED full ranking framework · MS MARCO | MRR@10: 28.7 -> 42.4 | Helped |
| Use two-stage UED pretraining instead of a BERT Large encoder (ablation) | UED full ranking framework · MS MARCO | MRR@10: 39.4 -> 42.4 | Helped |
| Use two-stage UED pretraining instead of a BERT Large encoder (ablation) | UED full ranking framework · MS MARCO | Recall@1000: 91.2 -> 94.0 | Helped |
| Jointly fine-tune ranking and query-generation tasks (ablation) | UED full ranking framework · MS MARCO | MRR@10: 40.9 -> 42.4 | Helped |
| Jointly fine-tune ranking and query-generation tasks (ablation) | UED passage expansion · MS MARCO development set | MRR@10: 27.5 -> 28.7 | Helped |
| Add decoder pre-training (ablation) | UED full ranking framework · MS MARCO | Recall@1000: 91.8 -> 94.0 | Helped |
| Pretrain the encoder with SRP rather than NSP (ablation) | UED full ranking framework · MS MARCO | MRR@10: 40.8 -> 42.4 | Helped |
Closest papers by meaning
- Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankers (ICLR 2025 · similarity 0.88)
- Diversifying Query: Region-Guided Transformer for Temporal Sentence Grounding (AAAI 2025 · similarity 0.87)
- Making Retrieval-Augmented Language Models Robust to Irrelevant Context (ICLR 2024 · similarity 0.86)
- Analyzing Similarity Metrics for Data Selection for Language Model Pretraining (NeurIPS 2025 · similarity 0.86)
- USTAD: Unified Single-model Training Achieving Diverse Scores for Information Retrieval (ICML 2024 · similarity 0.86)
- InstructRetro: Instruction Tuning post Retrieval-Augmented Pretraining (ICML 2024 · similarity 0.86)
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.