The Argonauts at SemEval-2026 Task 9: Multilingual Polarization Detection and Classification Using LLM Prompting and Transformer Fine-Tuning

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

Summary

The Argonauts' participation in SemEval-2026 Task 9 focused on multilingual polarization detection and classification. For Subtask 1, binary classification, they used Qwen3-14B with structured few-shot prompting in 4-bit mode, achieving macro-F1 scores of 0.847 for Bengali (4th place) and 0.808 for English (9th place). Subtask 2, multi-label classification in English and Bengali, involved fine-tuning XLM-RoBERTa-large and RoBERTa-base with an uneven loss (γ+ = 1, γ− =4) and label-specific thresholds, which boosted development macro F1 by up to 24.6 points. The final English test macro F1 was 0.454 (21st place). Analysis showed LLM prompting improves binary detection, while threshold adjustment is crucial for multi-label class imbalance.

Key takeaway

For NLP engineers developing multilingual polarization detection systems, consider integrating LLM prompting for binary classification tasks to enhance performance. When tackling multi-label classification with class imbalance, fine-tuning Transformer models with uneven loss and label-specific thresholds can significantly improve results, as demonstrated by a 24.6 point F1 increase.

Key insights

LLM prompting and fine-tuning with specific loss/thresholds effectively detect multilingual online polarization.

Principles

• LLM prompting enhances binary polarization detection.
• Threshold adjustment is critical for multi-label class imbalance.

Method

Qwen3-14B with 4-bit few-shot prompting for binary tasks; XLM-RoBERTa-large/RoBERTa-base fine-tuning with uneven loss (γ+ = 1, γ− =4) and label-specific thresholds for multi-label.

In practice

• Employ Qwen3-14B for binary polarization tasks.
• Apply uneven loss and label-specific thresholds for imbalanced multi-label classification.

Topics

• SemEval-2026
• Polarization Detection
• Multilingual NLP
• LLM Prompting
• Transformer Fine-tuning
• Class Imbalance

Best for: Research Scientist, AI Scientist, 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.
• Multi-View Attention Multiple-Instance Learning Enhanced by LLM Reasoning for Cognitive Distortion Detection — Combining LLMs with MIL and ELB components improves cognitive distortion detection by enhancing interpretability and reasoning.
• Revisiting the Systematicity in Negation in the Era of In-Context Learning — LLMs partially grasp negation via in-context learning, but struggle with robust scope recognition and function vector composition for it.
• A Multitask Transformer for Offensive Language Detection and Target Identification in HateBR — Multitask learning improves offensive language detection consistency but may reduce fine-grained target attribution.
• Detecting HIV-Related Stigma in Clinical Narratives Using Large Language Models — LLMs can effectively detect HIV-related stigma in clinical notes, with encoder models outperforming generative models in zero-shot settings.
• 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.

Open in AIssential →

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