Deep learning - Study of the impact of platform-specific social norms on hate speech detection.

Hate speech detection faces a critical challenge: content moderation systems must adapt to platform-specific social norms while maintaining generalization across communities. This issue is important because current NLP models often underperform when applied across different platforms, leading to inconsistent moderation. We address this by proposing a domain-aware transformer model that explicitly encodes platform information through specialized tokens. Our solution enables the same model to learn both platform-specific nuances and cross-platform patterns. Evaluating on Twitter and Reddit data, we demonstrate that this approach outperforms conventional generalist models while nearly matching platform-specific specialists. The findings highlight an efficient pathway for improving content moderation systems that operate across diverse online communities.