Gradient Descenders at SemEval-2026 Task 9: Data-Centric Counterfactual Augmentation for Multi-Label Hate Speech Detection

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

Summary

The Gradient Descenders team submitted a novel data-centric counterfactual augmentation pipeline to SemEval-2026 Task 9 Subtask 2 for multi-label hate speech detection. Existing Transformer models often struggle with this task due to severe class imbalance and complex class intersectionality, which fosters spurious correlations. To address this, the team utilized Large Language Models (LLMs) as semantic generators. They synthesized diverse, targeted training samples through three distinct prompting strategies: Additive Label-Flipping (Attribute Injection), Context Decoupling, and Cross-Domain Identity Substitution. Fine-tuning a RoBERTa classifier on this augmented corpus significantly improved sensitivity to minority classes. Their system ultimately achieved a Macro-F1 score of 44.15% on the official test set, demonstrating the effectiveness of targeted LLM-based augmentation in highly imbalanced, multi-label environments.

Key takeaway

For Machine Learning Engineers developing hate speech detection systems, if you are encountering degraded performance due to class imbalance, consider implementing LLM-based counterfactual data augmentation. This approach, using strategies like Additive Label-Flipping, can significantly improve your model's sensitivity to minority classes. A Macro-F1 score of 44.15% on SemEval-2026 Task 9 demonstrates its efficacy. You should explore targeted data synthesis to enhance robustness in complex, multi-label environments.

Key insights

Targeted LLM-based counterfactual augmentation improves multi-label hate speech detection in imbalanced datasets.

Principles

• Class imbalance degrades Transformer performance.
• Counterfactual augmentation mitigates spurious correlations.
• LLMs can synthesize diverse training data.

Method

Employ LLMs as semantic generators to synthesize training samples using Additive Label-Flipping, Context Decoupling, and Cross-Domain Identity Substitution, then fine-tune a RoBERTa classifier.

In practice

• Apply LLM prompting for data synthesis.
• Use counterfactuals to balance minority classes.
• Fine-tune RoBERTa on augmented datasets.

Topics

• Hate Speech Detection
• Data Augmentation
• Large Language Models
• Counterfactual Generation
• RoBERTa Classifier
• Class Imbalance
• SemEval-2026 Task 9

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Addressing Data Scarcity in Bangla Fake News Detection: An LLM-Based Dataset Augmentation Approach — LLM-based data augmentation significantly improves fake news detection in low-resource languages by generating diverse synthetic data.
• Aligning Implied Statements for Implicit Hate Speech Generalizability with Context-Bounded Semi-hard Negative Mining — The ImpSH framework improves implicit hate speech detection by aligning posts with implied statements and using context-bounded semi-hard negatives.
• Assisted Counterspeech Writing at the Crossroads of Hate Speech and Misinformation — LLMs can assist counterspeech against co-occurring hate and misinformation, but expert refinement and mixed knowledge strategies are crucial.
• From Weights to Features: SAE-Guided Activation Regularization for LLM Continual Learning — SAE-guided activation regularization combats LLM catastrophic forgetting by using monosemantic features instead of polysemantic weights.
• Listening Between the Lines: Joint Learning of ASR Embeddings and LLM-Augmented Linguistics for Dementia Detection — Jointly learning ASR embeddings and LLM-augmented linguistics significantly improves dementia detection from speech.
• 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.

Open in AIssential →

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