Research map / Large language models / Language, safety and interpretability
Importance-Aware Learning for Neural Headline Editing
Qingyang Wu, Lei Li, Hao Zhou, Ying Zeng, Zhou Yu
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 headline-generation adaptation (ablation) | PAS · PHED test | BLEU-4: 19.23 -> 21.46 | Improved |
| Add headline-generation adaptation (ablation) | PAS · PHED test | ROUGE-L: 37.05 -> 39.47 | Improved |
| Add headline-generation adaptation (ablation) | PAS · PHED test | Sent-REP-4 (%): 14.61 -> 8.882 | Improved |
| Add headline-generation adaptation (ablation) | PAS · PHED test | Unique 4-grams: 15.58k -> 18.20k | Improved |
| Add SIA loss (ablation) | PAS · PHED test | BLEU-4: 21.13 -> 21.46 | Improved |
| Add SIA loss (ablation) | PAS · PHED test | ROUGE-L: 39.27 -> 39.47 | Improved |
| Add SIA loss (ablation) | PAS · PHED test | Token-REP-4 (%): 0.4004 -> 0.3722 | Improved |
| Add SIA loss (ablation) | PAS · PHED test | Sent-REP-4 (%): 9.673 -> 8.882 | Improved |
| Add SIA loss (ablation) | PAS · PHED test | Perplexity: 7.588 -> 7.616 | Hurt |
| Use SIA instead of MLE for fine-tuning (ablation) | PAS · PHED test | human preference (win rate): PAS vs. PAS - SIA: 7.56% win, 6.44% lose | Improved |
| Increase token-level weight parameter α (ablation) | SIA · PHED test | Token-REP-4 | Mixed |
| Increase sentence-level weight parameter β (ablation) | SIA · PHED test | Sent-REP-4 | Improved |
Closest papers by meaning
- Attractive or Faithful? Popularity-Reinforced Learning for Inspired Headline Generation (AAAI 2020 · similarity 0.86)
- Structure Learning for Headline Generation (AAAI 2020 · similarity 0.84)
- Detecting Incongruity between News Headline and Body Text via a Deep Hierarchical Encoder (AAAI 2019 · similarity 0.83)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.82)
- Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis (NeurIPS 2026 · similarity 0.81)
- Revealing and Mitigating Over-Attention in Knowledge Editing (ICLR 2025 · similarity 0.80)
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.
