Anti-hate Speech Framework: Leveraging Hedging Hyperbolic Learning

The proliferation of hate speech has intensified alongside advancements in internet technology, rendering the enhancement of hate speech detection accuracy. Traditional approaches have predominantly concentrated on augmenting pre-training and deepening network architectures to imbue neural networks with the capacity for profound semantic understanding. However, these methods often overlook the critical aspect of enhancing model capabilities to discern the subtleties and similarities inherent in hate speech. In this study, we introduce a novel framework for hate speech detection predicated on the concept of hedge hyperbolic learning. This framework employs contrastive learning within a hyperbolic space, focusing on the feature vectors to capitalize on the lexical and syntactic similarities that are pivotal in accurately identifying hate speech. Experiments on two public datasets demonstrate the effectiveness of our famework.