PEACE: Cross-Platform Hate Speech Detection- A Causality-guided Framework
This work addresses the challenge of building generalizable hate speech detection models for social media platforms, where distribution shifts and lack of labeled data hinder performance, representing an incremental advance by applying causal inference to a known bottleneck.
The authors tackled the problem of cross-platform hate speech detection by proposing a causality-guided framework called PEACE, which leverages inherent causal cues like sentiment and aggression to improve generalization across platforms, achieving state-of-the-art performance with up to 15% improvement in F1-score over baselines.
Hate speech detection refers to the task of detecting hateful content that aims at denigrating an individual or a group based on their religion, gender, sexual orientation, or other characteristics. Due to the different policies of the platforms, different groups of people express hate in different ways. Furthermore, due to the lack of labeled data in some platforms it becomes challenging to build hate speech detection models. To this end, we revisit if we can learn a generalizable hate speech detection model for the cross platform setting, where we train the model on the data from one (source) platform and generalize the model across multiple (target) platforms. Existing generalization models rely on linguistic cues or auxiliary information, making them biased towards certain tags or certain kinds of words (e.g., abusive words) on the source platform and thus not applicable to the target platforms. Inspired by social and psychological theories, we endeavor to explore if there exist inherent causal cues that can be leveraged to learn generalizable representations for detecting hate speech across these distribution shifts. To this end, we propose a causality-guided framework, PEACE, that identifies and leverages two intrinsic causal cues omnipresent in hateful content: the overall sentiment and the aggression in the text. We conduct extensive experiments across multiple platforms (representing the distribution shift) showing if causal cues can help cross-platform generalization.