This study explores the application of generative artificial intelligence in university crisis public opinion management, aiming to innovate university online public opinion management technology and provide a new perspective for generative AI applications in social governance. By integrating prompt engineering and contextual learning, a sentiment analysis framework based on generative AI is constructed, with ChatGPT as the core model. Taking the case of a doctoral student's public accusation of sexual harassment at a Beijing university, we use web scraping to collect Weibo data and apply information lifecycle theory to identify public opinion evolution stages. Through few-shot learning strategies, 10 high-quality labeled examples were selected to guide ChatGPT's sentiment classification and extract negative sentiment keywords across stages to reveal evolution patterns. Findings show the incident exhibited four phases: doubt, anger, reflection, and rationality. Negative sentiment peaked at 58.5% during the outbreak phase, rising from 48.3% in the occurrence phase and declining to 40.4% in the fading phase. ChatGPT outperformed traditional models like TF-IDF-SVM and CNN-BiLSTM-Attention, improving accuracy by 5.87% and 1.56% respectively, with superior performance in metaphorical and sarcastic contexts.
关键词
生成式人工智能/ChatGPT/上下文学习/提示工程/高校突发事件/网络舆情/情感演化
Key words
Generative AI/ChatGPT/Contextual Learning/Prompt Engineering/University Crisis Events/Online Public Opinion/Sentiment Evolution
评论