PolDeck at SemEval-2026 Task 9: Multilingual Online Polarization Detection via Hybrid Model Ensembling and Data Augmentation

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

Summary

The PolDeck system, developed for SemEval 2026 Task 9, addresses Multilingual Online Polarization Detection using a hybrid ensemble framework. This framework integrates few-shot prompting with the Qwen3-30B model, a native multilingual XLM-R encoder, and a translation-augmented DeBERTa encoder. To counter label imbalance, PolDeck employs a multi-stage data augmentation pipeline that utilizes large language models for synthetic paraphrasing and cross-lingual translation. The system achieved a Top 10 ranking on both the English and German leaderboards, demonstrating the effectiveness of combining high-capacity monolingual models with flexible multilingual models within a unified system for detecting online polarization. Its code is publicly available on GitHub.

Key takeaway

For NLP Engineers developing multilingual text classification systems, particularly for tasks like online polarization detection, you should consider a hybrid ensemble approach. Integrating high-capacity monolingual models like Qwen3-30B with flexible multilingual encoders such as XLM-R and DeBERTa, combined with LLM-driven data augmentation, can significantly improve performance and address label imbalance. This strategy offers a robust path to achieving Top 10 benchmark results in complex cross-lingual tasks.

Key insights

Hybrid ensembles integrating monolingual and multilingual models, enhanced by LLM-driven data augmentation, effectively detect online polarization.

Principles

• Combine high-capacity monolingual and flexible multilingual models.
• Use LLMs for synthetic paraphrasing and cross-lingual translation.
• Multi-stage augmentation mitigates label imbalance.

Method

Build a hybrid ensemble using few-shot Qwen3-30B, XLM-R, and translation-augmented DeBERTa encoders. Augment data via LLM-driven synthetic paraphrasing and cross-lingual translation to address label imbalance.

In practice

• Explore Qwen3-30B for few-shot prompting.
• Implement LLM-based synthetic data augmentation.
• Utilize XLM-R and DeBERTa for multilingual encoding.

Topics

• Online Polarization Detection
• Hybrid Model Ensembling
• Multilingual NLP
• Data Augmentation
• Qwen3-30B
• SemEval 2026

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP 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.
• Graph-Augmented LLMs for Swiss MP Ideology Prediction — Graph-augmented LLMs can model complex political ideologies from text and relational data.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• Authorship Attribution in Multilingual Machine-Generated Texts — Multilingual authorship attribution for LLM-generated text is complex, with current monolingual methods showing limited cross-lingual transferability across diverse language families.
• Datasets for Verb Alternations across Languages: BLM Templates and Data Augmentation Strategies — This work addresses the underexplored ability of large language models (LLMs) to capture cross-sentence paradigmatic patterns, specifically verb alternations.

Open in AIssential →

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