People have presence across different information networks on the social web. The problem of user identity linking, is the task of establishing a connection between accounts of the same user across different networks. Solving this problem is useful for: personalized recommendations, cross platform data enrichment and verifying online information among others. In this paper, we propose a deep learning based approach that jointly models heterogeneous data: text content, network structure as well as profile names and images, in order to solve the user identity linking problem. We perform experiments on a real world problem of connecting the social profile (Twitter) and academic profile (DBLP) of researchers. Experimental results show that our joint model achieves a 97% F1 score outperforming state-of-the-art results that consider profile, content or network features only.