YNU-HPCC at SemEval-2026 Task 9: Hybrid Augmentation and Regularization Strategies for Multilingual Polarization Type Classification

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, medium

Summary

YNU-HPCC developed a system for SemEval 2026 Task 9: Multilingual Polarization Type Classification, designed to perform multi-label classification on texts in 22 languages. This system identifies five polarization types: political, racial/ethnic, religious, gender/sexual, and others. The primary challenges addressed were uneven data distribution, extreme class imbalance, and complex cross-lingual semantic understanding. Their proposed training framework, built upon XLM-RoBERTa-large, integrates hybrid augmentation and multi-strategy regularization. Specifically, it combines feature-space Mixup augmentation, an asymmetric loss function, adversarial training, and exponential moving average. Multi-label decisions are optimized using dynamic thresholds. The method achieved a macro-F1 score of 0.48 on the validation set, demonstrating improved classification performance and generalization in multilingual and imbalanced contexts.

Key takeaway

For NLP Engineers developing multilingual classification systems, especially with imbalanced data, you should consider integrating hybrid augmentation and regularization strategies. Implementing feature-space Mixup augmentation, an asymmetric loss function, and adversarial training, as demonstrated with XLM-RoBERTa-large, can significantly improve classification performance and generalization. Optimize your multi-label decisions using dynamic thresholding to achieve better macro-F1 scores in complex scenarios.

Key insights

A hybrid augmentation and regularization framework improves multilingual polarization classification on imbalanced data using XLM-RoBERTa-large.

Principles

• Handling data imbalance is crucial for multilingual NLP.
• Combining augmentation and regularization boosts generalization.
• Dynamic thresholds optimize multi-label classification.

Method

Fine-tune XLM-RoBERTa-large with feature-space Mixup augmentation, an asymmetric loss function, adversarial training, and exponential moving average. Optimize multi-label decisions via dynamic thresholds.

In practice

• Apply Mixup augmentation for imbalanced text data.
• Use asymmetric loss for multi-label tasks.
• Implement dynamic thresholding for decision optimization.

Topics

• Multilingual NLP
• Polarization Classification
• Data Augmentation
• Regularization Strategies
• XLM-RoBERTa
• SemEval 2026

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Exploring Data Augmentation and Resampling Strategies for Transformer-Based Models to Address Class Imbalance in AI Scoring of Scientific Explanations in NGSS Classroom — Targeted data augmentation significantly improves transformer-based model performance in automated scoring of imbalanced educational text data.
• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Data Augmentations for Data-Constrained Language Model Pretraining — Data augmentations effectively mitigate overfitting in data-constrained language model pretraining, enabling productive multi-epoch training.
• From Weights to Features: SAE-Guided Activation Regularization for LLM Continual Learning — SAE-guided activation regularization combats LLM catastrophic forgetting by using monosemantic features instead of polysemantic weights.
• SURGELLM: Rethinking Multi-Task Evaluation through Task-Aware Feature Gating with Class-Balanced Normalization — SURGELLM enhances multi-task NLP performance by integrating task-aware feature gating and class-balanced normalization within a unified transformer framework.
• Adversarial Label Invariant Graph Data Augmentations for Out-of-Distribution Generalization — RIA improves OoD generalization under covariate shift using adversarial label-invariant data augmentations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.