DeepSemantics at SemEval-2026 Task 9: Label-Wise Optimization with Adaptive Focal Loss for Polarization Manifestation Identification

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

Summary

The DeepSemantics system, presented at SemEval-2026 Task 9, addresses the fine-grained identification of polarization manifestations within multilingual social media content. Its methodology integrates transformer-based encoders, specifically RoBERTa-base for English and Afro-XLM-R-small for Hausa, within a One-vs-Rest (OvR) framework. The system further incorporates controlled oversampling, Adaptive Focal Loss, and label-wise threshold optimization. To effectively counter severe class imbalance and label sparsity, the approach employs language-specific optimization strategies, informed by pairwise χ2 independence analysis. On the official test sets, DeepSemantics achieved macro-F1 scores of 0.464 in English, ranking 14th, and 0.192 in Hausa, securing the 5th position on the leaderboard.

Key takeaway

For NLP Engineers developing systems to identify fine-grained polarization in multilingual social media, consider integrating a One-vs-Rest framework with transformer encoders like RoBERTa-base or Afro-XLM-R-small. Your approach should incorporate controlled oversampling and Adaptive Focal Loss to manage class imbalance effectively. Furthermore, implement label-wise threshold optimization and language-specific strategies, informed by pairwise χ2 independence analysis, to enhance performance, particularly for languages with limited data.

Key insights

DeepSemantics identifies social media polarization using transformer encoders, OvR, Adaptive Focal Loss, and language-specific optimization.

Principles

• Combine OvR with transformer encoders.
• Address class imbalance with Adaptive Focal Loss.
• Tailor optimization for language specifics.

Method

The system uses RoBERTa-base (English) and Afro-XLM-R-small (Hausa) in an OvR setup. It applies controlled oversampling, Adaptive Focal Loss, and label-wise threshold optimization, guided by pairwise χ2 analysis for language-specific strategies.

In practice

• Implement OvR for multi-label classification.
• Use Adaptive Focal Loss for imbalanced data.
• Apply χ2 analysis for language-specific tuning.

Topics

• Polarization Detection
• Multilingual NLP
• Transformer Models
• Class Imbalance
• Adaptive Focal Loss
• SemEval

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.
• Bölüm 1: Yapay Zekada Bir Devrin Kapanışı — Transformer Öncesi “Sequential” Kısıtlar — The Transformer architecture revolutionized NLP by enabling parallel processing and global dependency modeling through attention, overcoming sequential constraints.
• From TF-IDF to Transformers: A Comparative and Ensemble Approach to Sentiment Classification — A soft voting ensemble of diverse machine learning and transformer models significantly enhances sentiment classification accuracy, particularly for nuanced text.
• SURGELLM: Rethinking Multi-Task Evaluation through Task-Aware Feature Gating with Class-Balanced Normalization — SURGELLM enhances multi-task NLP performance by integrating task-aware feature gating and class-balanced normalization within a unified transformer framework.
• How to Fine-Tune an SLM for Emotion Recognition — Fine-tuning SLMs for fine-grained emotion recognition on imbalanced datasets requires a multi-technique approach to data balancing and loss weighting.
• Application of integrated gradients explainability to sociopsychological semantic markers — Integrated gradients reveal word-level contributions to sociopsychological text classification, even with limited data.

Open in AIssential →

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