Link prediction via latent space logistic regression model*

Nowadays, link prediction is of vital importance in the operation of social network platforms. One typical application is to make accurate recommendation to enhance users’ activeness. In this article, we propose a latent space logistic regression model for link prediction. The model takes both the users’ attributes and the latent social space into consideration. Two pseudo maximum likelihood estimators are proposed for parameter estimation. They correspond to the concepts of reciprocity and transitivity, respectively, and are computationally efficient for large-scale social networks. Extensive simulation studies are provided to evaluate the finite sample performance of the newly proposed methodology. At last, a real data set of Sina Weibo is presented for illustration purposes.