Modeling social tagging using latent interaction potential

Modeling social tagging plays a critical role in identifying statistical regularities and structural principles common to social tagging systems. Existing modeling approaches only consider imitations or background knowledge of users. However, common interests among users are ignored. In this paper, latent interactions are applied to present the common interests, and dynamic patterns in empirical data are investigated. Furthermore, the latent interaction driven model (LIDM) is proposed to model social tagging. Experimental results show that the tag frequency distribution generated by LIDM is consistent with that in real-world data. Moreover, the latent interaction graph generated by LIDM has a higher average clustering coefficient and lower average shortest path compared with that generated by preferential attachment methods. This demonstrates that LIDM outperforms traditional methods.