Measurement and Sentiment Analysis of YouTube Video Comments

Abstract

According to the latest video consumption statistics in 2022, 92.7 percent of global Internet users worldwide visits online video-sharing platforms, such as YouTube and TikTok, every week. These users share their videos and exchange image/text comments to establish crucial social network interactions. Based on the existing research, users’ likes and comments are evidence commonly used to quantify the popularity of videos and social media creators. However, it remains largely unclear if the sentiment of comments, e.g., negative comments, will also affect the video or video creators’ popularity. In this thesis, we take initial steps to explore YouTube video comments via sentiment analysis. We present an in-depth measurement study of commenting and user’s comment behaviors on a sample of more than 7 million comments on 4 million YouTube videos. Our measurement indicates that Music and Gaming videos attract more feedback and are more likely to be affected by the sentiment of comments. To better understand this, we utilize three popular machine learning models and two deep learning models to analyze the sentiment of video comments. Unlike Twitter and Facebook-based research, our study proves that negative comments do not significantly impact the popularity of YouTube videos. This means the online video-sharing platforms are more robust against unhealthy comments or rumors.