Research map / Architectures and efficiency / RNNs and normalization
A Neural Corpus Indexer for Document Retrieval
Yujing Wang, Yingyan Hou, Haonan Wang, Ziming Miao, Shibin Wu, Hao Sun, Qi Chen, Yuqing Xia, Chengmin Chi, Guoshuai Zhao, Zheng Liu, Xing Xie and 4 more
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 DocT5Query-generated training queries (ablation) | NCI · NQ320k | Recall@1: 60.23 -> 65.86 | Improved |
| add document-content queries (ablation) | NCI · NQ320k | Recall@1: 62.49 -> 65.86 | Improved |
| use the PAWA decoder instead of shared classifier weights (ablation) | NCI · NQ320k | Recall@1: 63.36 -> 65.86 | Improved |
| use hierarchical semantic identifiers instead of random identifiers (ablation) | NCI · NQ320k | Recall@1: 62.75 -> 65.86 | Improved |
| add consistency-based regularization (ablation) | NCI · NQ320k | Recall@1: 65.07 -> 65.86 | Improved |
| constrain beam search to valid identifiers (ablation) | NCI · NQ320k | Recall@1: 65.65 -> 65.86 | Improved |
| increase PAWA adapter depth from 0 to 4 layers (ablation) | NCI · NQ320k | Recall@1: 63.36 -> 65.86 | Improved |
| increase PAWA adapter depth from 4 to 8 layers (ablation) | NCI · NQ320k | Recall@1: 65.86 -> 63.60 | Hurt |
| set consistency-regularization weight α to 0.15 rather than 0 (ablation) | NCI · NQ320k | Recall@1: 65.07 -> 65.86 | Improved |
| increase consistency-regularization weight α from 0.15 to 0.2 (ablation) | NCI · NQ320k | Recall@1: 65.86 -> 65.55 | Hurt |
| use NCI w/ qg-ft ensemble instead of SEAL (Large) | NQ320k · NQ320k | Recall@1: 59.93 -> 72.78 | Improved |
| use NCI ensemble instead of SEAL (Large) | TriviaQA · TriviaQA | R-Precision: 69.2 -> 80.84 | Improved |
Closest papers by meaning
- Semi-Parametric Retrieval via Binary Bag-of-Tokens Index (ICLR 2025 · similarity 0.86)
- Adaptive Retrieval and Scalable Indexing for k-NN Search with Cross-Encoders (ICLR 2024 · similarity 0.86)
- Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankers (ICLR 2025 · similarity 0.84)
- Contrastive Retrieval Heads for Improved Attention-Based Reranking (NeurIPS 2026 · similarity 0.84)
- Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT (ICML 2024 · similarity 0.84)
- FreshStack: Building Realistic Benchmarks for Evaluating Retrieval on Technical Documents (NeurIPS 2025 · similarity 0.84)
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.
