Reconstructing Positive Surveys from Negative Surveys with Background Knowledge

Negative Survey is a promising technique for collecting sensitive data. Using the negative survey, useful aggregate information could be estimated, while protecting personal privacy. Previous work mainly focuses on improving the general model of the negative survey without considering background knowledge. However, in real-world applications, data analysts usually have some background knowledge on the surveys. Therefore, in this paper, for the first time, we study the usage of background knowledge in negative surveys, and propose a method for accurately reconstructing positive surveys with background knowledge. Moreover, we propose a method for evaluating the dependable level of the positive survey reconstructed with background knowledge. Experimental results show that more reasonable and accurate positive surveys could be obtained using our methods.