Violent Text Detection Model With Knowledge Distillation in Sensitive Domains
Li Tan, Dongfang Li, Jing Liu, Boya Zou, He Liu · International Journal of Intelligent Systems · 2025
With the increasing discussion on sensitive topics such as dietary choices, food safety (FS), and public health policies in social media and online comments, the importance of violent language detection has become increasingly prominent. Especially during public health crises, such as food poisoning outbreaks, violent language can undermine public trust and spread false information. We developed a Chinese food health dataset Food23 and proposed a rumor detection method based on knowledge distillation (KD) technology to reduce network parameters and facilitate model deployment. By constructing domain features, this method proposes a segmented domain feature vector update mechanism and a segmented domain attention mechanism. Experimental results show that our method performs well in using domain features compared with existing multidomain and cross‐domain violent language detection tasks. Experiments verify that the proposed method has a performance improvement of about 3% over the baseline method.