YEZE at SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization via Heterogeneous Ensembling

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

Summary

The YEZE system, developed by Fengze Guo and Yue Chang for SemEval-2026 Task 9, addresses the detection and characterization of online polarization across multiple languages, cultures, and events. This multilingual approach independently models each of the three subtasks using a heterogeneous weighted ensemble. The ensemble combines XLM-RoBERTa-large and mDeBERTa-v3-base models. To manage severe label imbalance in multi-label settings, the system employs weighted binary cross-entropy. Trained exclusively on the provided task data, YEZE demonstrates robust performance across various languages, as detailed in the Proceedings of the 20th International Workshop on Semantic Evaluation (2026), pages 1860–1873.

Key takeaway

For NLP engineers developing systems for complex multilingual classification, consider adopting a heterogeneous ensembling strategy. Your models can achieve robust performance across diverse linguistic and cultural contexts by combining models like XLM-RoBERTa-large and mDeBERTa-v3-base. Furthermore, implementing weighted binary cross-entropy is crucial for effectively handling severe label imbalance in multi-label settings, ensuring more accurate and fair predictions.

Key insights

A heterogeneous ensemble of large language models effectively detects multilingual online polarization by independently modeling subtasks.

Principles

• Model subtasks independently for complex problems.
• Use weighted binary cross-entropy for label imbalance.
• Heterogeneous ensembling improves robustness.

Method

The system models each subtask independently using a heterogeneous weighted ensemble of XLM-RoBERTa-large and mDeBERTa-v3-base, applying weighted binary cross-entropy to mitigate multi-label imbalance.

In practice

• Apply XLM-RoBERTa-large for multilingual tasks.
• Integrate mDeBERTa-v3-base in ensemble architectures.
• Implement weighted binary cross-entropy for imbalanced data.

Topics

• Online Polarization Detection
• Multilingual NLP
• Heterogeneous Ensembling
• XLM-RoBERTa
• mDeBERTa-v3
• SemEval-2026

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.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources — Semantic deduplication and domain-balanced pretraining significantly enhance Persian PLM performance across diverse NLU tasks.
• Enhancing SDG-Text Classification with Combinatorial Fusion Analysis and Generative AI — Combinatorial Fusion Analysis (CFA) significantly enhances SDG text classification by combining diverse models and synthetic data.
• FLiP: Towards understanding and interpreting multimodal multilingual sentence embeddings — FLiP models effectively interpret sentence embeddings by linearly extracting lexical content, revealing underlying biases.
• 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.