Research map / Reinforcement learning / Visual and representation RL
A Deep Reinforced Model for Abstractive Summarization
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 RL instead of maximum-likelihood training with intra-attention | Abstractive summarization model · CNN/Daily Mail test | ROUGE-1: 38.30 -> 41.16 | Helped |
| Use mixed ML+RL training instead of RL alone | Abstractive summarization model · CNN/Daily Mail human evaluation | Readability: 4.18 -> 7.04 | Helped |
| Use mixed ML+RL training instead of RL alone | Abstractive summarization model · CNN/Daily Mail human evaluation | Relevance: 6.32 -> 7.45 | Helped |
| Use mixed ML+RL training instead of RL alone | Abstractive summarization model with intra-attention · CNN/Daily Mail test | ROUGE-1: 41.16 -> 39.87 | Hurt |
| Add intra-decoder attention (ablation) | Maximum-likelihood summarization model · CNN/Daily Mail test | ROUGE-1: 37.86 -> 38.30 | Helped |
| Add intra-decoder attention (ablation) | Maximum-likelihood summarization model · New York Times test | ROUGE-1: 44.26 -> 43.86 | Hurt |
| Add intra-decoder attention (ablation) | Maximum-likelihood summarization model · CNN/Daily Mail test, long reference summaries | ROUGE-1 | Helped |
| Add intra-decoder attention (ablation) | Maximum-likelihood summarization model · CNN/Daily Mail test, short reference summaries | ROUGE-1 | Hurt |
| Remove the maximum-likelihood term from mixed ML+RL training (ablation) | Abstractive summarization model · CNN/Daily Mail human evaluation | Readability: 7.04 -> 4.18 | Hurt |
| Optimize ROUGE-L with RL rather than ROUGE-2 | Policy-trained summarization model · Summarization output quality | Sentence completeness | Helped |
| Use limited-length ML+RL model rather than the full extractive model | New York Times dataset split from prior work · New York Times test, limited-length recall | ROUGE-1 recall: 42.2 -> 42.94 | Helped |
Closest papers by meaning
- Towards Improving Faithfulness in Abstractive Summarization (NeurIPS 2022 · similarity 0.90)
- Copy or Rewrite: Hybrid Summarization with Hierarchical Reinforcement Learning (AAAI 2020 · similarity 0.89)
- MeanSum: A Neural Model for Unsupervised Multi-Document Abstractive Summarization (ICML 2019 · similarity 0.89)
- Learning to summarize with human feedback (NeurIPS 2020 · similarity 0.89)
- Learning to Summarize by Learning to Quiz: Adversarial Agentic Collaboration for Long Document Summarization (ICLR 2026 · similarity 0.86)
- CAR-Transformer: Cross-Attention Reinforcement Transformer for Cross-Lingual Summarization (AAAI 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.