Research map / Architectures and efficiency / Attention and transformers
Preserve Context Information for Extract-Generate Long-Input Summarization Framework
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 |
|---|---|---|---|
| Add local context prompts to DYLE (ablation) | DYLE · QMSum | ROUGE-1: 34.42 -> 36.50 | Helped |
| Add global context prompts to DYLE (ablation) | DYLE · QMSum | ROUGE-2: 9.71 -> 11.22 | Helped |
| Add both local and global context prompts to DYLE (ablation) | DYLE · QMSum | ROUGE-2: 9.71 -> 11.41 | Helped |
| Add local context prompts to DYLE (ablation) | DYLE · arXiv | ROUGE-1: 46.41 -> 46.69 | Helped |
| Add global context prompts to DYLE (ablation) | DYLE · arXiv | ROUGE-1: 46.41 -> 46.86 | Helped |
| Add both local and global context prompts rather than only global prompts (ablation) | DYLE · arXiv | ROUGE-1: 46.86 -> 46.81 | Hurt |
| Increase local context prompts from K=1 to K=4 (ablation) | CAEG · QMSum | ROUGE-1: 35.46 -> 36.50 | Helped |
| Increase global context prompts from K=3 to K=4 (ablation) | CAEG · QMSum | ROUGE-1: 36.11 -> 35.15 | Hurt |
| Increase global context prompts from K=2 to K=4 (ablation) | CAEG · arXiv | ROUGE-1: 46.86 -> 46.47 | Hurt |
| Use context prompts rather than adjacent snippets to transfer local context (ablation) | DYLE · QMSum | ROUGE-1: 35.74 -> 36.50 | Helped |
| Use context prompts rather than snippet embeddings to transfer local context (ablation) | DYLE · QMSum | ROUGE-1: 33.26 -> 36.50 | Helped |
Closest papers by meaning
- RECOMP: Improving Retrieval-Augmented LMs with Context Compression and Selective Augmentation (ICLR 2024 · similarity 0.86)
- BooookScore: A systematic exploration of book-length summarization in the era of LLMs (ICLR 2024 · similarity 0.86)
- Document Summarization with Conformal Importance Guarantees (NeurIPS 2025 · similarity 0.85)
- Unsupervised Extractive Summarization with Learnable Length Control Strategies (AAAI 2024 · similarity 0.85)
- Language Models of Code Are Few-Shot Planners and Reasoners for Multi-Document Summarization with Attribution (AAAI 2025 · similarity 0.85)
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.85)
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.