Research map / Label-efficient and robust learning / Recognition and long-tail
ProtoryNet - Interpretable Text Classification Via Prototype Trajectories
Dat Hong, Tong Wang, Stephen Baek
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 |
|---|---|---|---|
| allow fine-tuning of the sentence encoder (ablation) | ProtoryNet · IMDB, Amazon, Yelp, Rotten Tomatoes, Hotel, Steam | accuracy: IMDB 0.871 -> 0.914; Rotten Tomatoes 0.771 -> 0.881 | Improved |
| use ProtoryNet instead of ProSeNet on long texts | ProtoryNet (not fine-tuned) · Yelp Reviews long texts | accuracy: 0.931 -> 0.949 | Improved |
| use ProtoryNet instead of ProSeNet on short texts | ProtoryNet (not fine-tuned) · Yelp Reviews short texts | accuracy: 0.943 -> 0.863 | Hurt |
| prune prototypes with a 0.5% threshold from K=200 (ablation) | ProtoryNet · Amazon Reviews | accuracy: 0.918 -> 0.918 with 14 of 200 prototypes | No clear effect |
| replace LSTM with logistic regression (ablation) | ProtoryNet · IMDB | accuracy: 0.859 -> 0.914 | Improved |
| replace LSTM with a decision tree (ablation) | ProtoryNet · Rotten Tomatoes | accuracy: 0.796 -> 0.881 | Improved |
| replace LSTM with average of sentiments (ablation) | ProtoryNet · IMDB | accuracy: 0.602 -> 0.914 | Improved |
| apply sparsity transformation (one active prototype) instead of dense similarity (ablation) | ProtoryNet (fine-tuned) · Rotten Tomatoes | accuracy: 0.896 -> 0.881 | Hurt |
| apply sparsity transformation instead of dense similarity (ablation) | ProtoryNet (fine-tuned) · Yelp Reviews | accuracy: 0.956 -> 0.954 | Hurt |
| add diversity and prototypicality terms with small nonzero weights (ablation) | ProtoryNet · Hotel Reviews, K=100 | accuracy | Improved |
| increase initial number of prototypes K (ablation) | ProtoryNet · Hotel Reviews | accuracy | No clear effect |
| use ProtoryNet instead of ProSeNet on multi-class | ProtoryNet · Consumer complaints (4-class) | accuracy: 0.878 -> 0.927 | Improved |
Closest papers by meaning
- Language Model Meets Prototypes: Towards Interpretable Text Classification Models through Prototypical Networks (AAAI 2025 · similarity 0.84)
- A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations (AAAI 2025 · similarity 0.83)
- ProtoArgNet: Interpretable Image Classification with Super-Prototypes and Argumentation (AAAI 2025 · similarity 0.83)
- Analyze Feature Flow to Enhance Interpretation and Steering in Language Models (ICML 2025 · similarity 0.83)
- Neurons to Words: A Novel Method for Automated Neural Network Interpretability and Alignment (AAAI 2025 · similarity 0.83)
- Automatically Interpreting Millions of Features in Large Language Models (ICML 2025 · 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.
