Psychological Health Knowledge-Enhanced LLM-based Social Network Crisis Intervention Text Transfer Recognition Method

As the prevalence of mental health crises increases on social media platforms, how to effectively identify and deter potential harms has emerged as an urgent problem. To improve the detection ability of crisis-related content in social networks, this study proposes a large language model (LLM) text transfer recognition method for social network crisis intervention based on the enhancement of mental health knowledge that integrates mental health professional knowledge and transfer learning technology. We introduce a multi-level framework that employs transfer learning on a large language model BERT and integrates domain mental health knowledge, sentiment analysis as well as behavior prediction modeling techniques. This approach proposes a mental health annotation tool trained on social media datasets from crisis events, helping a large language model find potential language cues and then determine the presence of a psychological crisis and crisis acts. Experimental results indicate that the proposed model is superior to the traditional method in crisis detection accuracy, and demonstrate a greater sensitivity to underlying differences in context and emotion.