Random Walk with Pre-filtering for Social Link Prediction

The prosperity of content-oriented social media services has raised the new chances for understanding users' social behaviors. Different from traditional social networks, the links in social media are usually influenced by user preferences rather than the real world connections, thus the traditional methods based on social network evolvement may fail to reveal the adequate links. Meanwhile, the existing link prediction algorithms considering both social topology and nodes attributes might be too much computationally complex. To deal with these challenges, in this paper, we propose a twosteps link prediction framework, in which a filter is functioned to select the candidates firstly, and then the adapted Supervised Random Walk (SRW) is executed to rank the candidates for prediction. Experiments on the real world data set of social media indicate that our framework could effectively and efficiently predict the appropriate links, which outperforms the baselines including ordinary SRW with acceptable margin.