Research map / Label-efficient and robust learning / Recognition and long-tail
DeepChannel: Salience Estimation by Contrastive Learning for Extractive Document Summarization
Jiaxin Shi, Chen Liang, Lei Hou, Juanzi Li, Zhiyuan Liu, Hanwang Zhang
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 |
|---|---|---|---|
| Use DeepChannel instead of the strongest listed baseline, SWAP-NET | CNN/Daily Mail summarization · CNN/Daily Mail test | ROUGE-1 F-1: 41.60 -> 41.50 | No clear effect |
| Use DeepChannel instead of rnn-ext + RL | CNN/Daily Mail summarization · CNN/Daily Mail test | ROUGE-2 F-1: 18.72 -> 17.77 | Hurt |
| Use DeepChannel instead of rnn-ext + RL | CNN/Daily Mail summarization · CNN/Daily Mail test | ROUGE-L F-1: 37.76 -> 37.62 | No clear effect |
| Train on CNN/Daily Mail and evaluate out of domain with DeepChannel rather than Refresh | Extractive summarization · DUC 2007, 75 bytes | ROUGE-1 recall: 18.39 -> 19.53 | Improved |
| Train on CNN/Daily Mail and evaluate out of domain with DeepChannel rather than Refresh | Extractive summarization · DUC 2007, 275 bytes | ROUGE-1 recall: 26.80 -> 28.85 | Improved |
| Train DeepChannel with 1/100 of the CNN/Daily Mail training data rather than Refresh | Extractive summarization · CNN/Daily Mail test, 1/100 training | ROUGE-1 F-1: 36.05 -> 39.41 | Improved |
| Add the attention penalization term (α=0.001) instead of removing it (ablation) | DeepChannel · CNN/Daily Mail test | ROUGE-1 F-1: 40.89 -> 41.50 | Improved |
| Add the attention penalization term (α=0.001) instead of removing it (ablation) | DeepChannel · CNN/Daily Mail test | ROUGE-2 F-1: 17.21 -> 17.77 | Improved |
| Add the attention penalization term (α=0.001) instead of removing it (ablation) | DeepChannel · CNN/Daily Mail test | ROUGE-L F-1: 37.08 -> 37.62 | Improved |
| Increase attention penalization weight from α=0.001 to α=0.01 (ablation) | DeepChannel · CNN/Daily Mail test | ROUGE-1 F-1: 41.50 -> 41.30 | Hurt |
| Increase attention penalization weight from α=0.001 to α=0.1 (ablation) | DeepChannel · CNN/Daily Mail test | ROUGE-1 F-1: 41.50 -> 40.49 | Hurt |
Closest papers by meaning
- Salience Estimation via Variational Auto-Encoders for Multi-Document Summarization (AAAI 2017 · similarity 0.88)
- A Deep Reinforced Model for Abstractive Summarization (ICLR 2018 · similarity 0.86)
- Unity in Diversity: Learning Distributed Heterogeneous Sentence Representation for Extractive Summarization (AAAI 2018 · similarity 0.86)
- SummaRuNNer: A Recurrent Neural Network Based Sequence Model for Extractive Summarization of Documents (AAAI 2017 · similarity 0.86)
- Learning to Summarize by Learning to Quiz: Adversarial Agentic Collaboration for Long Document Summarization (ICLR 2026 · similarity 0.83)
- Unsupervised Extractive Summarization with Learnable Length Control Strategies (AAAI 2024 · similarity 0.82)
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.
