Research map / Label-efficient and robust learning / Recognition and long-tail
Extrapolated Markov Chain Oversampling Method for Imbalanced Text Classification
Aleksi Avela, Pauliina Ilmonen
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 |
|---|---|---|---|
| add oversampling with EMCO gamma=1 to SVM | Linear SVM · Reuters full text, very low frequency, 20% ratio | balanced accuracy: .656 -> .815 | Improved |
| add majority-class transitions (gamma=1 instead of plain MCO gamma=0) (ablation) | EMCO · Reuters full text, very low frequency, 20% ratio | balanced accuracy: .730 -> .815 | Improved |
| add majority-class transitions (gamma=1 instead of MCO) (ablation) | EMCO · Reuters titles, very low frequency, 10% ratio | balanced accuracy: .679 -> .736 | Improved |
| use EMCO instead of EDA | Linear SVM · HuffPost full text, very low frequency, 20% ratio | balanced accuracy: .613 -> .722 | Improved |
| use EMCO instead of EDA | Linear SVM · Reuters titles, low frequency, 10% ratio, HuffPost titles* | balanced accuracy: .721 -> .695 | Mixed |
| use EMCO gamma=1 instead of SVM without oversampling | Linear SVM · Reuters titles, very low frequency, 10% ratio | F1-score: .300 -> .277 | Hurt |
| use EMCO gamma=1 instead of SVM without oversampling | Linear SVM · Reuters full text, very low frequency, 10% ratio | F2-score: .333 -> .547 | Improved |
| lower majority weight gamma from 1 to 0.1 (ablation) | EMCO · Reuters titles, 20% ratio synthetic vocabulary | synthetic words added: 465.68 -> 435.85 words; recall 0.57 -> 0.55 | Mixed |
| add EMCO oversampling to LSTM | BiLSTM with fastText · Reuters titles, very low frequency, 20% ratio | balanced accuracy: .603 -> .705 | Improved |
| use EMCO instead of EDA for LSTM oversampling | BiLSTM with fastText · Reuters titles, very low frequency, 20% ratio | F1-score: .294 -> .161 | Hurt |
| use EMCO synthetic vocabulary instead of EDA synonym replacement | minority vocabulary expansion · Reuters titles, 20% ratio | balanced accuracy: 0.53 -> 0.69 | Improved |
| use EMCO vocabulary instead of DECOM uniform prior | minority vocabulary expansion · Reuters titles, 20% ratio | balanced accuracy: 0.51 -> 0.69 | Improved |
Closest papers by meaning
- Synthetic Tabular Data Generation for Imbalanced Classification: The Surprising Effectiveness of an Overlap Class (AAAI 2025 · similarity 0.85)
- Concentration and excess risk bounds for imbalanced classification with synthetic oversampling (NeurIPS 2025 · similarity 0.84)
- Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification (ICLR 2025 · similarity 0.84)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.84)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.83)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (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.
