An improved social attribute inference scheme based on multi-attribute correlation

In an online social network (OSN), users share their personal media such as hobbies, photos, and videos. These information can help users find potential friends, but the exposure of their characteristics may also cause problems like their features being mined and utilized. To solve these problems, many social network providers choose to hide users' sensitive information like their real names and exact positions. To ensure the effectiveness of the methods that can protect users' information, there are several attacks to infer users' hidden information. In this paper, based on the previous analysis of the relevance between two attributes, we analyze the relevance of multiple attributes and find that there is also a strong correlation between multiple attributes as well as the correlation between attributes and behaviors, which also has an impact on the attribute inference. Therefore, based on the correlation between multiple attributes and behaviors, we propose a new attribute inference scheme. We use the Apriori algorithm based on data cube to dig the correlation between multiple attributes and behaviors and analyze them with Kulczynski measure and Cosine measure. We model a social network as a Markov Random Field (MRF) and use Loopy Belief Propagation (LBP) for attribute inference attacks. We compare our method with the previous algorithm. The results show that our method has better performance than the traditional methods.