VGU-M.Tech-AI at SemEval-2026: Multilingual Multi-Label Classification of Online Polarization Types via Weighted Transformer Fine-Tuning and Adaptive Per-Label Threshold Optimization

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

Summary

The VGU-M.Tech-AI team at SemEval-2026 introduced MMCOPT, a system for multilingual multi-label classification of online polarization types in social media posts. This research classifies posts into five categories: political, racial, religious, gender/sexual, or other. MMCOPT integrates a "distilbert-base-multilingualcased" model with a two-layer MLP head, employing a class-imbalance-weighted binary cross-entropy loss and adaptive per-label threshold optimization to enhance the validation micro-F1 score. Trained on the POLAR benchmark, a large multilingual dataset spanning seven languages, the model achieved an internal validation micro-F1 score of 0.7855 and a macro-F1 score of 0.7749. Ranked on the official Codabench leaderboard, MMCOPT demonstrated competitive performance across 22 language tracks, with its strongest results in Hindi (0.7429) and Urdu (0.7073).

Key takeaway

For NLP Engineers developing systems to detect online polarization, consider adopting the MMCOPT approach. Your models can achieve competitive multilingual multi-label classification by fine-tuning "distilbert-base-multilingualcased" with a weighted binary cross-entropy loss and optimizing per-label thresholds. This strategy is particularly effective for handling class imbalance and improving F1 scores across diverse languages, as demonstrated by its strong performance in Hindi (0.7429) and Urdu (0.7073).

Key insights

Multilingual transformer fine-tuning with adaptive thresholding effectively classifies online polarization types across diverse languages.

Principles

• Class imbalance weighting improves F1 scores.
• Per-label threshold optimization enhances multi-label classification.
• Multilingual models generalize across diverse languages.

Method

MMCOPT fine-tunes a "distilbert-base-multilingualcased" model with a two-layer MLP head, using class-imbalance-weighted binary cross-entropy loss and optimizing per-label thresholds on the POLAR benchmark.

In practice

• Apply "distilbert-base-multilingualcased" for multilingual text tasks.
• Implement adaptive thresholds for multi-label F1 optimization.
• Use weighted loss for imbalanced classification datasets.

Topics

• Multilingual NLP
• Online Polarization Detection
• Transformer Fine-tuning
• Multi-label Classification
• distilbert-base-multilingualcased
• 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.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• From Logistic Regression to GPT-2: Building a Complete Spam Detection & Sentiment Analysis Pipeline — For imbalanced text classification, prioritize F1-score and Precision-Recall AUC over accuracy and ROC-AUC.
• 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.
• 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.
• On GATE, Text and Social Media Analysis, and Detecting Misinformation Online — A multi-layer analytical pipeline effectively extracts and categorizes digital identity aspects of agrifood products from online discourse.

Open in AIssential →

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