Link prediction in online social networks based on supervised joint denoising model

Link prediction of social networks can capture important information about missing links for applications in many fields. Because of the failure to make full use of information as well as capture all properties, the link prediction precision of most of methods is low. For higher precision, we propose a novel algorithm, a supervised joint denoising model (SJDM) that formulates the link prediction problem as a supervised matrix “denoising problem. The central piece of our method is a function that is trained using the features of users and topological structures of social networks. The function can map the observed “corrupted matrix to an “uncorrupted matrix (target matrix). We performed community detection using the target matrix, which is better than using the original matrix. Five real networks are processed with this algorithm. The results show that SJDM algorithm is more efficient compared to the other four algorithms.