OZemi at SemEval-2026 Task 9: A Cross-Lingual Approach to Online Text Polarization Classification Using Multilingual Models and Adaptive Loss Formulation

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

Summary

The OZemi team submitted a unified multilingual approach to SemEval-2026 Task 9, focused on detecting multilingual, multicultural, and multievent online polarization. Their system integrates multilingual models with data-level techniques and a class-weighted cross-entropy loss. This combination aims to mitigate data imbalance across various languages, subtasks, and categories. The approach demonstrated consistent performance, achieving macro F1 scores above 70% in most languages for Subtask 1. Notably, the system secured the highest rank (1 out of 44) for Persian in Subtask 1. These results indicate the framework offers a flexible foundation for analyzing polarization across multiple languages and tasks.

Key takeaway

For NLP Engineers developing systems to detect online text polarization across multiple languages, the OZemi team's framework offers a robust solution. You should consider integrating multilingual models with data-level techniques and class-weighted cross-entropy loss to handle data imbalance effectively. This approach, which achieved macro F1 scores above 70% and top rankings in SemEval-2026 Task 9, provides a flexible foundation for consistent cross-lingual and multi-task analysis.

Key insights

The OZemi team's system uses multilingual models and adaptive loss to classify online text polarization across diverse languages and tasks.

Principles

• Unified multilingual approaches enhance efficiency.
• Adaptive loss mitigates data imbalance.
• Cross-lingual models offer consistent performance.

Method

The system combines multilingual models with data-level techniques and a class-weighted cross-entropy loss to address data imbalance in multilingual, multi-task online polarization classification.

In practice

• Apply class-weighted loss for imbalanced datasets.
• Use multilingual models for cross-lingual tasks.

Topics

• Online Polarization Detection
• Multilingual Models
• Cross-Lingual NLP
• Class-Weighted Loss
• Data Imbalance Mitigation
• SemEval-2026 Task 9

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.
• Lexical and Orthographic Variation in Portuguese Financial Tweets: Annotation, Analysis, and Implications for Embedding Models — Lexical and orthographic variations in financial tweets are systematic, influencing embedding model performance and requiring curated lexicons.
• IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources — Vector-based semantic deduplication and domain-balanced pretraining significantly enhance Persian language model performance across diverse NLU tasks.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• SHIFT: Semantic Harmonization via Index-side Feature Transformation for Multilingual Information Retrieval — SHIFT corrects language bias in Multilingual Information Retrieval by adjusting document embeddings at indexing using relative language vectors.

Open in AIssential →

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