Research map / Large language models / Language, safety and interpretability
Exploring Explainable Selection to Control 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 ESCA-BERT instead of the strongest listed baseline, BERTSUMabs | abstractive summarization · CNN/DailyMail | R-2: 19.39$^\ast$ -> 19.52 | Helped |
| Use ESCA-BERT instead of the strongest listed baseline, BERTSUMextabs | abstractive summarization · NYT50 | R-2: 31.02$^\ast$ -> 32.22 | Helped |
| Use pair-wise rather than point-wise ranking (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-1: 32.68 -> 36.41 | Helped |
| Use pair-wise rather than point-wise ranking (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-2: 15.41 -> 17.19 | Helped |
| Use pair-wise rather than point-wise ranking (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-L: 30.33 -> 33.68 | Helped |
| Replace 2-layer self-attention with the interaction matrix (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-1: 42.8 -> 42.7 | No clear effect |
| Replace 2-layer self-attention with the interaction matrix (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-2: 20.1 -> 20.0 | No clear effect |
| Replace 2-layer self-attention with the interaction matrix (ablation) | ESCA-BERT extractor · CNN/DailyMail extractive summaries | R-L: 39.2 -> 39.2 | No clear effect |
| Set novelty threshold to ε_n=0.3 rather than ε_n=0 (ablation) | ESCA controllable inference · CNN/DailyMail synthetic novelty | R-1: 44.78 -> 45.66 | Helped |
| Set novelty threshold to ε_n=0.3 rather than ε_n=0 (ablation) | ESCA controllable inference · CNN/DailyMail synthetic novelty | R-2: 35.39 -> 36.28 | Helped |
| Set relevance threshold to ε_r=0.5 rather than ε_r=0 (ablation) | ESCA controllable inference · CNN/DailyMail synthetic relevance | R-1: 41.35 -> 41.52 | Helped |
| Set relevance threshold to ε_r=0.5 rather than ε_r=0 (ablation) | ESCA controllable inference · CNN/DailyMail synthetic relevance | R-L: 38.57 -> 38.55 | Hurt |
Closest papers by meaning
- Controlling the Amount of Verbatim Copying in Abstractive Summarization (AAAI 2020 · similarity 0.90)
- Summarization Programs: Interpretable Abstractive Summarization with Neural Modular Trees (ICLR 2023 · similarity 0.89)
- MeanSum: A Neural Model for Unsupervised Multi-Document Abstractive Summarization (ICML 2019 · similarity 0.88)
- Learning to Summarize by Learning to Quiz: Adversarial Agentic Collaboration for Long Document Summarization (ICLR 2026 · similarity 0.87)
- Unsupervised Extractive Summarization with Learnable Length Control Strategies (AAAI 2024 · similarity 0.86)
- Language Models of Code Are Few-Shot Planners and Reasoners for Multi-Document Summarization with Attribution (AAAI 2025 · 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.