Explore the Similarity of YouTube Video Creators

Abstract

The rapid growth of social media platforms, most notably Youtube, is generating great driving forces to boost the Internet industry. The Internet business and advertising service providers have made great efforts to find the best platforms, the best video creators, and the best videos for their commercials. In this business model, the video creators who have the most subscribers are normally preferred. However, it remains largely unclear if these popular video creators are more cost effective and better than others. In this thesis, we carry out an extensive measurement to understand the most popular videos and their video creators on YouTube. Our dataset consists of 100,000 recently published videos from the most popular 500 Youtube channels. The measurement indicates that the total view count of a video is highly related to the number of dislikes, especially when these videos have similar like counts. This observation motivates us to further explore the similarity of video creators and check if special types of video creators can better attract users' attention. Our statistical model analysis shows that some YouTubers with relatively low subscriptions can create better and more popular videos than the most-subscribed YouTubers. Therefore, we further applied Hierarchical Clustering to help businesses and advertising service providers find these valuable and cost-effective video creators. The follow-up analysis result shows that the RMSLE of our model is less than 0.7.