Research on Food Safety Online Public Opinion Based on Text Mining—Taking the “3.15 Incident of a Brand of Cassoulet” as an Example

近年来，随着社交媒体和移动互联网的普及，食品安全事件的网络舆情呈现爆发式传播态势。公众通过抖音、小红书平台等渠道实时发布、讨论食品安全问题，形成具有社会影响力的舆论场。本研究以“3.15某品牌黄焖鸡事件”为切入点，依据2025年3月13日18时02分抖音央视新闻发布国务院食安办向山东省食药安办、河南省食安办发出挂牌督办通知书的视频为事件时间区分点，运用文本挖掘技术对抖音平台官方媒体与娱乐个人媒体发布的视频评论文本进行系统性分析。通过词云可视化、词频统计和LDA主题模型揭示两类媒体的话语建构差异，结合情感分析模型量化受众情感极化程度。结果发现，(1) 央视发声前后舆情阶段分化特征显著。(2) 官方媒体与娱乐个人媒体在舆情传播中形成二元分化结构。(3) 央视发声前后情感倾向呈现显著变化。(4) LDA主题模型显示两类媒体用户关注点存在时空错位。In recent years, with the popularization of social media and mobile Internet, online public opinion on food safety incidents has shown explosive propagation. The public has been posting and discussing food safety issues in real time through channels such as TikTok and rednote platforms, forming a public opinion field with social influence. This study takes the “3.15 incident of a brand of cassoulet” as the entry point, and based on the video released by TikTok CCTV News at 18:02 on March 13, 2025, in which the State Council’s Food Safety Office issued a registered supervisory notification to the Food and Drug Safety Office of Shandong Province and the Food Safety Office of Henan Province, the time point of the incident was differentiated, and the video released by official media and entertainment personal media on the TikTok platform was analyzed by using the text mining technology. The text mining technology is used to systematically analyze the video comment texts released by the media and entertainment personal media. The differences in discourse construction between the two types of media are revealed through word cloud visualization, word frequency statistics and LDA topic model, and the degree of audience sentiment polarization is quantified by combining with sentiment analysis model. The results found that (1) the stage differentiation of public opinion before and after CCTV’s voice is characterized significantly. (2) The official media and entertainment personal media formed a binary polarization structure in the dissemination of public opinion. (3) Emotional tendencies showed significant changes before and after CCTV’s voice. (4) The LDA topic model shows that there is a temporal and spatial mismatch in the focus of users of the two types of media.