Research map / Reinforcement learning / Visual and representation RL
Copy or Rewrite: Hybrid Summarization with Hierarchical Reinforcement Learning
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 |
|---|---|---|---|
| copy selected sentences only instead of rewriting all (ablation) | Copy/Rewrite + EXT-RL · CNN/DailyMail test | ROUGE-1: 40.73 -> 41.48 | Helped |
| add copy/rewrite switching with unbalanced marker (ablation) | Copy + EXT-RL · CNN/DailyMail test | ROUGE-1: 41.48 -> 42.10 | Helped |
| add copy/rewrite switching with unbalanced marker (ablation) | Copy + EXT-RL · CNN/DailyMail test | ROUGE-L: 37.79 -> 38.87 | Helped |
| history mechanism instead of unbalanced marker (ablation) | Copy/Rewrite + EXT-RL · CNN/DailyMail test | ROUGE-1: 42.10 -> 41.98 | No clear effect |
| reinforce rewriter independently of extractor (ablation) | Copy/Rewrite + History · CNN/DailyMail test | ROUGE-1: 41.98 -> 42.21 | Helped |
| replace independent RL with hierarchical RL (HRL) (ablation) | Copy/Rewrite + History · CNN/DailyMail test | ROUGE-1: 42.21 -> 42.46 | Helped |
| replace independent RL with hierarchical RL (HRL) (ablation) | Copy/Rewrite + History · CNN/DailyMail test | ROUGE-L: 38.94 -> 39.19 | Helped |
| add hierarchical BERT representation (ablation) | Copy/Rewrite + History + HRL · CNN/DailyMail test | ROUGE-2: 19.10 -> 19.43 | Helped |
| HYSUM + HRL vs best previous work | HYSUM · CNN/DailyMail test | ROUGE-1: 41.5 -> 42.46 | Helped |
| HYSUM vs Fast-RL in human evaluation | HYSUM · Human eval 50 samples | conciseness (1-5): 3.1 -> 3.83 | Helped |
| HYSUM + HRL vs DCA on ROUGE-2 | HYSUM · CNN/DailyMail test | ROUGE-2: 19.21 -> 19.10 | No clear effect |
Closest papers by meaning
- A Deep Reinforced Model for Abstractive Summarization (ICLR 2018 · similarity 0.89)
- Controlling the Amount of Verbatim Copying in Abstractive Summarization (AAAI 2020 · similarity 0.89)
- Learning to summarize with human feedback (NeurIPS 2020 · similarity 0.88)
- Keywords-Guided Abstractive Sentence Summarization (AAAI 2020 · similarity 0.88)
- CAR-Transformer: Cross-Attention Reinforcement Transformer for Cross-Lingual Summarization (AAAI 2024 · similarity 0.85)
- Learning to Summarize by Learning to Quiz: Adversarial Agentic Collaboration for Long Document Summarization (ICLR 2026 · 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.