SugarTextNet: A Transformer-Based Framework for Detecting Sugar Dating-Related Content on Social Media with Context-Aware Focal Loss

The emergence of sugar dating-related content on social media has raised significant societal and regulatory concerns, particularly regarding the commercialization of intimate relationships. In response, researchers have developed SugarTextNet, a transformer-based framework aimed at effectively identifying such content. This framework incorporates a pretrained transformer encoder, an attention-based cue extractor, and a contextual phrase encoder to capture both prominent and subtle features in user-generated text. To tackle the challenges of class imbalance, SugarTextNet employs Context-Aware Focal Loss, a specialized loss function that enhances the detection of minority classes. Evaluated on a newly curated dataset of 3,067 Chinese social media posts from Sina Weibo, SugarTextNet has demonstrated substantial improvements over traditional machine learning models and large language models. This innovation highlights the importance of context-aware modeling in sensitive content detection…