| 江长斌,陈子涵,黄英辉,王丹丹,何珂.基于ChatGPT的高校突发事件网络舆情情感分析研究[J].数字图书馆论坛,2025,21(6):53~62 |
| 基于ChatGPT的高校突发事件网络舆情情感分析研究 |
| Sentiment Analysis of Network Public Opinion in University Emergencies Based on ChatGPT |
| 投稿时间：2025-05-12 |
| DOI：10.3772/j.issn.1673-2286.2025.06.005 |
| 中文关键词: 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 |
| 英文关键词: Generative Artificial Intelligence; ChatGPT; Context Learning; Prompt Engineering; University Emergency; Network Public Opinion;Emotional Evolution |
| 基金项目:本研究得到国家社会科学基金项目“大数据视域下‘隐性’政治舆情演化规律及治理路径研究”（编号：19BSH013）资助。 |
| 作者 | 单位 | | 江长斌 | 武汉理工大学管理学院 | | 陈子涵 | 武汉理工大学管理学院 | | 黄英辉 | 武汉理工大学管理学院 | | 王丹丹 | 香港浸会大学传理学院 | | 何珂 | 武汉理工大学管理学院 |
|
| 摘要点击次数: 4 |
| 全文下载次数: 0 |
| 中文摘要: |
| 探索生成式人工智能在高校突发舆情事件管理中的应用，旨在推动高校网络舆情管理技术的创新，也为生成式人工智能在社会治理领域的应用探索提供新的视角。融合提示工程与上下文学习，构建基于生成式人工智能的情感分析框架，并以ChatGPT为核心模型展开研究。以“北京某大学女博士实名举报博导性骚扰”事件为研究案例，采用网络爬虫获取微博平台数据，结合信息生命周期理论划分舆情演化阶段。通过少样本学习策略筛选10条高质量标注示例，引导ChatGPT情感分析模型实现情感分类，并挖掘多阶段负向情感关键词以揭示演化规律。研究发现该事件舆情呈现质疑、愤怒、反思、理性四阶段特征，负向情感占比从发生期的48.3%攀升至爆发期峰值58.5%，随后逐步回落至消退期的40.4%，呈现先升后降的演化趋势。研究表明，ChatGPT情感分析整体性能优于TF-IDF-SVM与CNN-BiLSTM-Attention等传统基线方法，其模型整体准确率较传统基线模型分别提升5.87个百分点和1.56个百分点，且在隐喻与反讽等复杂语境下负向情感分类表现更优。 |
| 英文摘要: |
| Exploring the application of Generative Artificial Intelligence (GAI) in the management of public opinion emergencies in higher education institutions aims to advance innovation in network public opinion management technologies for universities while providing new perspectives for the application of GAI in social governance. This study integrates prompt engineering and context learning to construct a sentiment analysis framework based on GAI, with ChatGPT as the core model. Taking the case of a female doctoral student at a university in Beijing reporting sexual harassment by her supervisor in real name as the research subject, we collect data from the Weibo platform using web crawlers and divide the evolution stages of public opinion based on information lifecycle theory. By employing a few-shot learning strategy, 10 high-quality annotated examples are selected to guide the ChatGPT model in sentiment classification, and multi-stage negative sentiment keywords are extracted to reveal evolutionary patterns. We find that the public opinion evolution of this incident exhibits a four-stage pattern: skepticism, anger, reflection, and rationality, with the proportion of negative sentiment rising from 48.3% in the latency phase to a peak of 58.5% during the outbreak phase, followed by a gradual decline to 40.4% in the regression phase, illustrating a rise-then-fall evolutionary trajectory.The findings demonstrate that ChatGPT outperforms traditional baseline methods such as TF-IDF-SVM and CNN-BiLSTM-Attention in sentiment analysis. Specifically, the ChatGPT-based sentiment analysis model achieves overall accuracy improvements of 5.87% and 1.56%, respectively, over traditional models, while exhibiting superior performance in negative sentiment classification within complex contexts involving metaphor and irony. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
