A privacy preservation method for multiple-source unstructured data in online social networks

Abstract

Significant amounts of data are generated in different formats (e.g., text, audio, and video) in online social networks and from other online sources, which can be mined to facilitate decision-making in different applications. However, it is nontrivial to preserve user and/or data privacy when we are analyzing data from multiple sources, with different characteristics. In this paper, we propose a method to preserve privacy when dealing with data obtained from multiple/disparate sources. Taking images and corresponding text descriptions as examples, we set up different methods and protection scenarios to protect the private information in the data. Our proposed method is also designed to maintain the availability of existing information and links between the two forms to some extent. We then quantitatively analyze the performance of the proposed approach.