Background: Social media platforms have become a viable source of medical information, with patients and healthcare professionals using them to share health-related information and track diseases. Similarly, YouTube, the largest video-sharing platform in the world contains vlogs where individuals talk about their illnesses. The aim of our study was to investigate the use of Natural Language Processing (NLP) to identify the spoken content of YouTube vlogs related to the diagnosis of Coronavirus disease of 2019 (COVID-19) for public health screening. Methods: COVID-19 videos on YouTube were searched using relevant keywords. A total of 1000 videos being spoken in English were downloaded out of which 791 were classified as vlogs, 192 were non-vlogs, and 17 were deleted by the channel. The videos were converted into a textual format using Microsoft Streams. The textual data was preprocessed using basic and advanced preprocessing methods. A lexicon of 200 words was created which contained words related to COVID-19. The data was analyzed using topic modeling, word clouds, and lexicon matching. Results: The word cloud results revealed discussions about COVID-19 symptoms like "fever", along with generic terms such as "mask" and "isolation". Lexical analysis demonstrated that in 96.46% of videos, patients discussed generic terms, and in 95.45% of videos, people talked about COVID-19 symptoms. LDA Topic Modeling results also generated topics that successfully captured key themes and content related to our investigation of COVID-19 diagnoses in YouTube vlogs. Conclusion: By leveraging NLP techniques on YouTube vlogs public health practitioners can enhance their ability to mitigate the effects of pandemics and effectively respond to public health challenges.

翻译：背景：社交媒体平台已成为医疗信息的重要来源，患者和医疗专业人员利用这些平台分享健康相关信息并追踪疾病动态。同样，全球最大的视频共享平台YouTube上存在大量个人分享疾病经历的博客视频。本研究旨在探索利用自然语言处理技术识别YouTube博客中与2019冠状病毒病诊断相关的口语内容，以支持公共卫生筛查。方法：通过相关关键词检索YouTube上的COVID-19视频。共下载1000个英语口语视频，其中791个被归类为博客视频，192个为非博客视频，17个因频道删除而失效。视频内容通过Microsoft Streams转换为文本格式，并采用基础与进阶预处理方法处理文本数据。建立包含200个COVID-19相关词汇的词典。采用主题建模、词云和词典匹配对数据进行分析。结果：词云结果显示视频中讨论了COVID-19症状（如"发热"）以及"口罩"、"隔离"等通用术语。词汇分析表明，96.46%的视频中患者讨论了通用术语，95.45%的视频中提及了COVID-19症状。LDA主题建模结果成功提取出与YouTube博客中COVID-19诊断相关的主要主题和内容。结论：通过利用YouTube博客的自然语言处理技术，公共卫生从业人员能够增强减轻流行病影响的能力，并有效应对公共卫生挑战。