文摘
With the rapid development of location-based social networks (LBSNs), increasing media data is ceaselessly uploaded by users. The multimedia data is often scattered and not informative and consequently they can not directly represent the semantics of each venue. Most of prior works leverage the user’ travelling histories to recommend new venues to users. However, these works often focus on the users’ travelling histories, while ignore the concepts or the popular levels of venues. In this paper, we proposed a quality model for venue recommendation by utilizing multimedia data to predict the interested level of each venue. First, we apply the graph cut method to generate the latent textual topics. Second, we leverage visual data from Flickr to train concept detectors to automatically label visual information. Third, the weighted bipartite matching algorithm is implemented to generate the venue multimedia topics by bridging the textual information and the visual information. Finally, we utilize the matching cost to predict the popular level of venue for recommendation. The experiments have been conducted on the cross-platform datasets. The results demonstrate the superiority of the proposed model.