Masking and labeling problems in social network data management

Abstract

Online social networks (OSN) contain data about persons or objects including private information and user generated labels. We study two problems on managing OSN data. Data is often shared with trusted parties. Still, data has to be adapted so that it does not trivially reveal identities of the users and their interactions. We formulate this problem of “masking” data by deliberately introducing uncertainty and trading it off with the utility of data for useful analyses. We present methods for masking static and dynamic OSN data and show high accuracy in experiments for answering a variety of queries over the masked data. User-generated labels have many uncertainties due to missing values, synonyms, and so on. The problem of Label Set Enhancing captures the task of reducing these uncertainties, by inferring missing values, replacing labels for larger concept labels and so on. We present first known, efficient, iterative solutions to this problem where the labels form a hierarchy. Our evaluations show significant benefits in using a hierarchy for reducing uncertainty in label sets in OSN data.