Research map / Architectures and efficiency / Attention and transformers
Global-aware Beam Search for Neural Abstractive Summarization
Ye Ma, Zixun Lan, Lu Zong, Kaizhu Huang
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 |
|---|---|---|---|
| replace beam search with global-aware inference | PEGASUS · BillSum | ROUGE-1 F1: 57.00 -> 58.66 | Improved |
| replace beam search with global-aware inference | PEGASUS · NewsRoom | ROUGE-1 F1: 42.05 -> 44.68 | Improved |
| replace beam search with global-aware inference | PEGASUS · XSUM | ROUGE-1 F1: 47.05 -> 47.33 | Improved |
| replace beam search with global-aware inference | BART · CNN/DM | ROUGE-1 F1: 44.12 -> 45.13 | Improved |
| add predicted global attention coverage penalty to beam search | BART beam search · XSUM | ROUGE-1 F1: 45.38 -> 44.54 | Hurt |
| add CTRL repetition penalty to beam search | BART beam search · CNN/DM | ROUGE-1 F1: 44.12 -> 44.11 | No clear effect |
| add attention head masking to beam search | BART · CNN/DM | ROUGE-1 F1: 44.12 -> 45.54 | Improved |
| add attention head masking to beam search | BART · XSUM | ROUGE-1 F1: 45.38 -> 45.35 | No clear effect |
| use true global attention (ORACLE) instead of predicted attention | global-aware · CNN/DM | ROUGE-1 F1: 45.13 -> 51.85 | Improved |
| remove length reward (ablation) | global-aware · CNN/DM | ROUGE-1 F1: 45.13 -> 44.39 | Hurt |
| remove attention score (ablation) | global-aware · CNN/DM | ROUGE-1 F1: 45.13 -> 44.12 | Hurt |
| remove length reward (ablation) | global-aware · XSUM | ROUGE-1 F1: 45.44 -> 45.57 | Improved |
Closest papers by meaning
- Faithful to the Original: Fact Aware Neural Abstractive Summarization (AAAI 2018 · similarity 0.84)
- Empirical Analysis of Beam Search Performance Degradation in Neural Sequence Models (ICML 2019 · similarity 0.84)
- Summarization Programs: Interpretable Abstractive Summarization with Neural Modular Trees (ICLR 2023 · similarity 0.84)
- Controlling the Amount of Verbatim Copying in Abstractive Summarization (AAAI 2020 · similarity 0.84)
- Exploring Explainable Selection to Control Abstractive Summarization (AAAI 2021 · similarity 0.83)
- Text summarization via global structure awareness (ICLR 2026 · similarity 0.83)
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.
