基于人工智能(ChatGPT)对供精受孕者YouTube视频的情感分析
本研究采用横断面设计,由两名独立评审筛选50个描述供精受孕者观点、故事或情感的YouTube视频,并将自动生成的英文转录文本输入ChatGPT(GPT-4o模型)进行情感分类。ChatGPT判定44%视频为积极情感、28%为消极、24%为混合、4%为中性,情感分布与期望分布差异有统计学意义(Χ2[3, n=50]=16.240,P=0.001);与人类评审相比,ChatGPT结果与至少一名评审的一致率为90%。结论认为ChatGPT在分析YouTube视频转录文本情感方面表现良好,供精受孕者对自身供精受孕状态的情感倾向偏积极。
要点
- ChatGPT判定44%视频为积极情感、28%为消极、24%为混合、4%为中性。
- 情感观察分布与期望分布差异有统计学意义(Χ2[3, n=50]=16.240,P=0.001)。
- ChatGPT与至少一名人类评审的情感结果一致率为90%。
核心结论
- ChatGPT在与人类评审比较时,分析YouTube视频转录文本情感表现出较强性能。
- 经ChatGPT分析,供精受孕者在YouTube视频中对自身供精受孕状态的情感偏积极。
研究设计与样本
- 横断面分析,使用关键词「donor conceived」检索YouTube视频,由两名独立评审筛选。
- 纳入标准为视频描述供精受孕者对其生活经历的观点、故事或情感。
- 将每个视频自动生成的英文YouTube转录文本输入ChatGPT(GPT-4o模型),提示词为「Given this text, what is the sentiment conveyed? Is it positive or negative? Text: {transcript}」。
- 共纳入50个YouTube视频,视频长度中位数6.91分钟,观看次数中位数2217.5。
- 采用卡方拟合优度检验和独立性检验比较观察到的情感比例与期望分布。
主要发现
- ChatGPT判定44%的视频为积极情感,28%为消极,24%为混合,4%为中性。
- 观察到的情感分布与期望分布差异有统计学意义(Χ2[3, n=50]=16.240,P=0.001)。
- 与两名人类评审相比,ChatGPT情感结果与至少一名评审的一致率为90%。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
RESEARCH QUESTION: What is the tone of donor-conceived people (DCP) towards their donor-conceived status in YouTube videos assessed for sentiment analysis via ChatGPT, a natural language-processing tool? DESIGN: A cross-sectional analysis of YouTube videos using the keyword 'donor conceived' was performed by two independent reviewers. Videos were selected for inclusion if they described DCPs' opinions, stories or emotions about their life experiences. Reviewers categorized the sentiments of the videos as neutral, positive, negative or mixed. The auto-generated English-language YouTube transcript for each video was then inputted into ChatGPT (GPT-4o model) using the following prompt: 'Given this text, what is the sentiment conveyed? Is it positive or negative? Text: {transcript}'. Sentiment results generated via ChatGPT were categorized as neutral, positive, negative or mixed. Chi-squared goodness-of-fit and independence tests were performed to assess the observed proportions of sentiments compared with the expected distribution. RESULTS: Fifty YouTube videos were identified. The median video length was 6.91 min and the median number of views was 2217.5. Via ChatGPT, 44% of videos were determined to have a positive sentiment, 28% were negative, 24% were mixed and 4% were neutral, with a significant difference in the observed distribution of sentiments from expected (Χ2 [3, n = 50] = 16.240, P = 0.001). When compared with two human reviewers, ChatGPT sentiment results had 90% concordance with at least one reviewer. CONCLUSION: ChatGPT shows a strong performance in analysing YouTube video transcript sentiments when compared with human reviewers. The sentiments of DCP towards their donor-conceived status in YouTube videos have a positive leaning when analysed by ChatGPT.
引用
- Galperin S, Wiener L, Chafitz O, et al. A sentiment analysis of YouTube videos from donor-conceived people, utilizing artificial intelligence (ChatGPT). Reprod Biomed Online. 2025. PMID: 40623336. doi:10.1016/j.rbmo.2025.104907
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。