A Social Network Approach for Automated Generation of YouTube Playlists

Abstract

Video online sharing and social media platform YouTube has more than a billion monthly users, where videos continue to be uploaded at a high rate per minute. With so many videos being created every minute, the viewers have to choose from this pool of videos available on the platform. And not only the viewers but even the content creators also aim to maximize their presence on YouTube, and having playlists on their channels can turn out to be a real help in sorting such a huge number of videos. Creation of the playlists manually can be a very tedious task, and hence, the paper proposes to automate the process of the creation of playlists for a channel. The research is performed on multiple YouTube channels with more than 100 videos. The solution involves the creation of the Social Network Graph for the channel with the help of the NLP framework. Four different clustering algorithms, namely, K-Means++, EM-GMM, Mean-Shift, DBSCAN, and OPTICS algorithms, are also performed on the network graphs. A comparative analysis is performed on the results of the clustering algorithms to determine the best algorithm that provides the most useful playlist distribution of the videos. The approach concludes with the results showing that k-means++ outperformed the other algorithms on the basis of the Silhouette and Calinski Harabasz scores.