Constructing Social Networks Based on Near-Duplicate Detection in YouTube Videos

With the video-sharing websites springing up, more and more people would like to upload and share either their own videos or remix others'. Meanwhile, they could view and comment the videos that they are interested in. Therefore, social networks among videos and users exist implicitly. In this work, we construct two types of social networks, video networks (VN) and topic participant networks (TPN), by utilizing videos, related metadata and near-duplicate detection. In the networks, the nodes denote the videos or users while the weights of the directed edges represent the correlation between the nodes. Then, several indices are defined to quantitatively evaluate the importance of the nodes in the networks. Experiments are conducted by using YouTube videos and corresponding metadata related with a specific event. Experimental results show that the analysis of social networks and indices fits the evolution of the event and the roll topic participants plays in spreading Internet videos very well. Finally, we extensionally investigate to utilize the network for recognizing important videos and participants, summarizing video datasets, and tracking an event with few videos.