Empirical Analysis of Attribute Inference Techniques in Online Social Network

Social network is a popular platform that users share their videos, photos, and other media. To protect the information from being abused for malicious purposes, many social network platforms provide privacy protection mechanisms, such as anonymizing personal identification information, hiding users' profiles that only could be disclosed by social friends, etc. However, recent researches demonstrate that the hidden private attributes still can be inferred by using auxiliary information obtained from social data, e.g., social structures, online behaviors, and correlations among social attributes, etc. Intuitively, most of these methods are sensitive to specific datasets structures and their performance is influenced significantly by the parameter configurations. Thoroughly understanding of the existing inference attacks is very important to develop efficient social data protection solutions. In this paper, we conduct a systematic analysis on typical attribute inference approaches and develop several experiments to evaluate the efficiency of these methods with different social datasets, under different pre-configured environments as well. Our experiment results disclose the impacts on the approach performance caused by different factors, e.g., dataset properties, critical parameters.