Attribute Inference by Link Strength Modeling in Online Social Networks with User Tags

Abstract

The link strength between two users in online social networks is generally latent and can not be observed directly. The strength is usually related to the interests, behaviors, posted texts, common friends, and common followings of two users. Most previous works have ignored the distinctions of the link strengths among different pairs of users, and some works simply classify the relationships into strong and weak instead of a particular value. Given the importance of link strength in link prediction or item recommendation system, in this paper, we propose a novel method for modeling the strength of links in social networks by jointly taking the common friends, common followings, user behaviors, and user tags into consideration. A new method to construct the tags for each user based on the semantics of open information is also presented. The attribute inference and tag prediction approach based on link strength is put forward and evaluated by the experiments on a real-world dataset, the inferred results prove the feasibility of the proposed model and demonstrate that the model substantially outperforms the compared methods.