The explosion of multimedia content in social media networks raises a great demand of developing tools to facilitate producing, sharing and viewing media content. Particularly, Flickr groups, self-organized communities with declared, common interests, are able to help users to conveniently participate in social media network. In this paper, we address the problem of automatically recommending groups to users. We propose to simultaneously exploit media contents and link structures between users and groups. To this end, we present a probabilistic latent topic model to model them in an integrated framework, expecting to jointly discover the latent interests for users and groups and simultaneously learn the recommendation function. We demonstrate the proposed approach on the dataset crawled from Flickr.com.