pfr821 at SemEval-2026 Task 9: Multilingual Polarization Detection via Hybrid XLM-RoBERTa with Targeted Data Augmentation and Imbalance-Aware Training

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

Summary

Team pfr821's HYPOLDET system, submitted to SemEval-2026 Task 9 (Polarization Detection, Subtask 1), addresses multilingual binary classification across 22 diverse languages. The system integrates three key components: a Hybrid Architecture extending XLM-RoBERTa-Large with a custom transformer encoder and attention-based pooling for enhanced token-level signal aggregation; targeted data augmentation using an LLM-based Grok API to generate culturally grounded synthetic examples for low-resource and imbalanced languages; and an imbalance-aware training regime incorporating a per-language balanced batch sampler, weighted focal loss, and label smoothing. HYPOLDET's best single model achieved an unweighted macro-averaged F1 of 0.796, while a lightweight ensemble reached 0.798, securing a top 10 rank in 7 languages and 2nd place for Hausa.

Key takeaway

For NLP Engineers developing multilingual classification systems, especially when facing data scarcity or class imbalance, you should consider HYPOLDET's integrated approach. Its hybrid XLM-RoBERTa architecture, targeted LLM-based data augmentation via Grok API, and imbalance-aware training regime provide a robust framework. Implementing these strategies can significantly improve performance on diverse, low-resource languages, as demonstrated by its F1 of 0.798 and 2nd place for Hausa.

Key insights

Hybrid XLM-RoBERTa, LLM-based data augmentation, and imbalance-aware training effectively detect multilingual polarization.

Principles

• Extend base models for richer signal aggregation.
• Augment data with culturally relevant synthetic examples.
• Tailor training to address class imbalance per language.

Method

HYPOLDET extends XLM-RoBERTa-Large with a custom transformer and attention pooling, augments data via Grok API for low-resource languages, and uses a balanced batch sampler, focal loss, and label smoothing for imbalance-aware training.

In practice

• Use LLMs for culturally specific data generation.
• Implement per-language batch balancing.
• Apply weighted focal loss for imbalanced datasets.

Topics

• Multilingual NLP
• Polarization Detection
• XLM-RoBERTa
• Data Augmentation
• Grok API
• Class Imbalance

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

Related on AIssential

• 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.
• An End-to-End Hybrid Framework for Rumour Detection in Low-Resources Algerian Dialect — A hybrid framework combining transformer embeddings with classical classifiers enables effective rumour detection in low-resource Algerian dialect.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation — Transfer learning and data augmentation effectively overcome low-resource challenges in Chinese dialect discrimination.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• AFRILANGTUTOR: Advancing Language Tutoring and Culture Education in Low-Resource Languages with Large Language Models — Dictionary-based synthetic data generation and combined SFT+DPO significantly enhance LLM tutoring for low-resource African languages.

Open in AIssential →

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