Link Prediction in Online Social Networks Based on the Unsupervised Marginalized Denoising Model

As online social networking platforms change the ways and means of people communicating, accurate link prediction among a massive pool of users has become a difficult problem. The problem arises in many applications, such as friend recommendation, news feedback, and product recommendation. In this paper, we propose a novel algorithm to solve this problem. The existing online social network link prediction algorithms have some deficiencies in link prediction accuracy because they cannot make full use of information or capture all the features. From an unconventional perspective, this paper formulates the link prediction problem as a matrix denoising problem. We first propose an unsupervised marginalized denoising model (USMDM) and explain in detail its effectiveness. The core of the USMDM lies with a mapping function that is capable of identifying patterns in a vast amount of user information and also understands the topological structure of social networks. The mapping function projects the observed matrix onto a target matrix. To improve efficiency and prevent overfitting, a low-rank matrix is used to replace the original matrix in the learning process. Using the weak law of large number, the function can be learned on limited datasets. To illustrate the effectiveness of the proposed algorithm, experiments are conducted on four real social networks, and the results demonstrate the effectiveness of the model.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.