Abstract

Social network study has become an important topic in many research fields. Early works on social network analysis focus on real world social interactions in either human society or animal world.With the explosion of Internet data, social network researchers start to pay more attention to the tremendous amount of online social network data. There is ample space for exploring social network research on large-scale online visual content. In this paper, we focus on studying a multi-label collective classification problem and develop a model that can harness mutually beneficial information among visual appearance, related semantic content and social network structure simultaneously. Our algorithm is then tested on CelebrityNet, a social network constructed by inferring the implicit relationship of people based on online multimedia content. We apply our model to a few important multimedia applications such as image annotation and community classification. We demonstrate that our algorithm significantly outperforms traditional methods on community classification and image annotation.