Seeing Hate Differently: Hate Subspace Modeling for Culture-Aware Hate Speech Detection

2025-10-10

AI Summary

Existing hate speech detection methods often overlook label bias in training data and the substantial influence of cultural context on the definition of hate, leading to high misclassification rates in cross-cultural settings. To address this, we propose a culture-aware personalized hate subspace model: first, we construct fine-grained subspaces based on combinatorial multidimensional cultural attributes; second, we employ label propagation to disentangle cultural confounding from annotation ambiguity, enabling personalized hate representation learning. Our approach jointly integrates cultural attribute modeling, label propagation, and subspace learning to enhance feature discriminability under sparse supervision. Experiments on multilingual and multicultural benchmark datasets demonstrate that our method achieves an average F1-score improvement of 1.05% over state-of-the-art approaches. It establishes a novel paradigm for interpretable, fair, and robust cross-cultural hate speech detection.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Hate speech detection has been extensively studied, yet existing methods often overlook a real-world complexity: training labels are biased, and interpretations of what is considered hate vary across individuals with different cultural backgrounds. We first analyze these challenges, including data sparsity, cultural entanglement, and ambiguous labeling. To address them, we propose a culture-aware framework that constructs individuals' hate subspaces. To alleviate data sparsity, we model combinations of cultural attributes. For cultural entanglement and ambiguous labels, we use label propagation to capture distinctive features of each combination. Finally, individual hate subspaces, which in turn can further enhance classification performance. Experiments show our method outperforms state-of-the-art by 1.05% on average across all metrics.

Problem

• Detecting hate speech with cultural bias in training labels
• Addressing data sparsity through cultural attribute combinations
• Resolving cultural entanglement and ambiguous labeling issues

Innovation

• Hate subspace modeling for culture-aware detection
• Modeling cultural attribute combinations to reduce sparsity
• Label propagation for disentangling cultural features