Team HausaNLP at SemEval-2026 Task 9: Tackling Class Imbalance in Low-Resource Hausa Polarization Detection

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

Summary

Team HausaNLP submitted a system to SemEval-2026 Task 9, Subtask 2, focusing on detecting specific polarization categories like Political, Religious, and Ethnic in Hausa social media comments. This task presented significant challenges due to extreme class imbalance and Hausa's low-resource status. Their system utilized a pre-trained multilingual transformer, Afro-XLMRLarge, fine-tuned with Weighted Binary CrossEntropy loss and dynamic undersampling at a 1:3 ratio to address the scarcity of polarized examples. On the official test set, the system achieved a Macro-F1 score of 0.2346 and a Micro-F1 score of 0.2581. The model demonstrated strong recall (Micro-Recall: 0.6166) in detecting polarization, though precision remained low at 0.1632. Its best per-class performance was an F1 score of 0.48 in the Political domain.

Key takeaway

For NLP Engineers developing systems for low-resource languages or tasks with severe class imbalance, consider combining pre-trained multilingual transformers with explicit imbalance handling. Your approach should integrate weighted loss functions and dynamic undersampling, as demonstrated by Team HausaNLP. Prioritize recall if detecting all instances of a rare class is critical, even if it impacts precision. This can provide a robust baseline for challenging text classification problems.

Key insights

Low-resource language polarization detection benefits from transformer fine-tuning with class imbalance mitigation techniques.

Principles

• Class imbalance requires explicit mitigation.
• Multilingual transformers aid low-resource NLP.
• Recall can be prioritized over precision.

Method

Fine-tune Afro-XLMRLarge with Weighted Binary CrossEntropy loss and dynamic undersampling (1:3 ratio) for low-resource text classification with severe class imbalance.

In practice

• Apply weighted loss functions for rare classes.
• Experiment with dynamic undersampling ratios.
• Evaluate recall for critical detection tasks.

Topics

• Hausa Language Processing
• Polarization Detection
• Class Imbalance
• Low-Resource NLP
• Multilingual Transformers
• 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.
• 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.
• 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.
• The Data Problem in Low-Resource Languages — Data availability, quality, and standardization are critical for NLP model performance across diverse languages.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.
• 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.

Open in AIssential →

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