Research map / Label-efficient and robust learning / Recognition and long-tail
CAWA: An Attention-Network for Credit Attribution
Saurav Manchanda, George Karypis
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 CAWA instead of MLTM for sentence attribution | Credit attribution · Ohsumed, TMC2007, Patents, Delicious | PPPA: average performance gain of 6.2% | Improved |
| Use CAWA instead of SEG-REFINE for sentence attribution | Credit attribution · Ohsumed, TMC2007, Patents, Delicious | PPPA: average performance gain of 9.8% | Improved |
| Use CAWA instead of MLTM | Multilabel classification · Five datasets | F1: average performance gain of 4.1% | Improved |
| Use CAWA instead of SEG-REFINE | Multilabel classification · Five datasets | F1: average performance gain of 1.6% | Improved |
| Add attention to the DNN-A architecture (ablation) | DNN-A · All five datasets | SOV and PPPA | Improved |
| Add average pooling before softmax (ablation) | CAWA · All five datasets; varying β | SOV | Improved |
| Increase β too far, making attention outweigh document probabilities (ablation) | CAWA · Credit attribution; large β | SOV | Hurt |
| Increase α, reducing the relative contribution of attention loss (ablation) | CAWA · All five datasets | AUC | Improved |
| Add attention loss that penalizes attention on absent classes (ablation) | CAWA · Multilabel classification | AUC | Hurt |
| Model each sentence as a distribution over classes rather than a single class | CAWA · Ohsumed, Patents, Delicious similar-class subsets | F1: 0.64 -> 0.68; 0.48 -> 0.53; 0.26 -> 0.39; 0.33 -> 0.41; 0.47 -> 0.50; 0.58 -> 0.62 | Improved |
Closest papers by meaning
- Retrospective In-Context Learning for Temporal Credit Assignment with Large Language Models (NeurIPS 2025 · similarity 0.81)
- De-biased Attention Supervision for Text Classification with Causality (AAAI 2024 · similarity 0.80)
- Language Models of Code Are Few-Shot Planners and Reasoners for Multi-Document Summarization with Attribution (AAAI 2025 · similarity 0.80)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · similarity 0.80)
- Scalable In-context Ranking with Generative Models (NeurIPS 2025 · similarity 0.80)
- Don't Pay Attention, PLANT It: Pretraining Attention via Learning-to-Rank (NeurIPS 2026 · 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.
