mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection

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

Summary

"mdok-style" participated in SemEval-2026 Task 9, focusing on multilingual polarization detection across 22 languages. This task involves identifying polarization in multilingual, multicultural, and multievent contexts, specifically addressing its detection, type, and manifestation. The research highlights online polarization as a significant concern, often preceding hate speech, offensive discourse, and social fragmentation, making early detection vital for safer online spaces. The team addressed this challenge by finetuning mid-size Large Language Models (LLMs) for sequence-classification. They employed the QLoRA parameter-efficient finetuning technique. To enhance detection robustness, the training data was significantly augmented, incorporating anonymized, lower-cased, upper-cased, and homoglyphied versions of the original multilingual datasets.

Key takeaway

For NLP Engineers developing multilingual content moderation systems, this work demonstrates a viable approach to detecting online polarization. You should consider finetuning mid-size LLMs with QLoRA for sequence-classification tasks, especially when dealing with diverse language data. Augmenting your training datasets with anonymized, case-varied, and homoglyphied text can significantly improve model robustness across 22 languages, crucial for preventing the escalation of hate speech and social fragmentation.

Key insights

Finetuning LLMs with QLoRA and augmented multilingual data effectively detects online polarization across 22 languages.

Principles

• Early polarization detection prevents online harm.
• Data augmentation improves multilingual robustness.
• Parameter-efficient finetuning suits mid-size LLMs.

Method

Finetune mid-size LLMs for sequence-classification using QLoRA. Augment multilingual training data with anonymized, lower-cased, upper-cased, and homoglyphied versions to enhance robustness.

In practice

• Apply QLoRA to mid-size LLMs.
• Augment data with varied text transformations.
• Develop systems for multilingual hate speech prevention.

Topics

• Multilingual Polarization Detection
• LLM Finetuning
• QLoRA
• Data Augmentation
• Sequence Classification
• Content Moderation

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.
• Cross-lingual Relation Extraction with Large Language Models: Zero-Shot, Few-Shot, and Fine-Tuned Evaluation on Romanian — QLoRA fine-tuning significantly boosts cross-lingual relation extraction performance for low-resource languages, outperforming prompt-only LLMs.
• Multi-Stage Training for Abusive Comment Detection in Indic Languages — A multi-stage pipeline combining language preprocessing and model ensembles effectively detects abusive Indic language comments while minimizing false positives.
• 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.
• Linguistic Issues in Computational Linguistics for Non-English Languages — English-centric LLM design creates systematic linguistic weaknesses for non-English languages across multiple levels.
• Building an LLM fine-tuning Dataset — Building multi-speaker Reddit datasets for LLM fine-tuning is complex but achievable, with efficient QLoRA methods requiring minimal samples.

Open in AIssential →

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