Research map / Label-efficient and robust learning / Recognition and long-tail
Heavy-tailed Representations, Text Polarity Classification & Data Augmentation
Hamid Jalalzai, Pierre Colombo, Chloé Clavel, Eric Gaussier, Giovanna Varni, Emmanuel Vignon, Anne Sabourin
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 |
|---|---|---|---|
| Learn a heavy-tailed representation rather than classify directly on BERT embeddings (ablation) | Sentiment classification with LHTR · Amazon and Yelp extreme regions | classification loss | Improved |
| Add a separate classifier dedicated to extremes (ablation) | LHTR · Amazon extreme | classification loss: 0.091 -> 0.08 | Improved |
| Add a separate classifier dedicated to extremes (ablation) | LHTR · Yelp extreme | classification loss: 0.139 -> 0.1205 | Improved |
| Use NN for bulk examples and LHTR for extreme examples | Sentiment classification · Amazon overall | classification loss: 0.098 -> 0.084 | Improved |
| Use NN for bulk examples and LHTR for extreme examples | Sentiment classification · Yelp overall | classification loss: 0.111 -> 0.103 | Improved |
| Augment training data with GENELIEX | FastText classifier · Amazon medium | F1: 84.0 -> 86.3 | Improved |
| Augment training data with GENELIEX | FastText classifier · Amazon large | F1: 93.3 -> 94.0 | Improved |
| Augment training data with GENELIEX | FastText classifier · Yelp medium | F1: 86.7 -> 88.4 | Improved |
| Augment training data with GENELIEX | FastText classifier · Yelp large | F1: 94.1 -> 94.2 | Improved |
| Use GENELIEX instead of synonym-based augmentation | Text data augmentation · Amazon and Yelp | diversity (dist1/dist2) | Improved |
| Scale an extreme latent code to generate sequences | GENELIEX · Amazon and Yelp sentiment generation | sentiment label preservation | Mixed |
| Use the unconstrained extreme classifier rather than impose angular structure explicitly (ablation) | LHTR extreme classifier · Preliminary classification experiments | classification performance | Improved |
Closest papers by meaning
- DEALT: LLM-driven Diversity-Enhanced Data Augmentation for Long-Tail Text Classification (AAAI 2026 · similarity 0.86)
- On the Robustness of Text Vectorizers (ICML 2023 · similarity 0.84)
- Label-Specific Feature Augmentation for Long-Tailed Multi-Label Text Classification (AAAI 2023 · similarity 0.84)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · similarity 0.83)
- Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification (ICLR 2025 · similarity 0.83)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.83)
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.
