An Accurate Negative Survey Using Answer Confidence Level

Negative survey is an effective method to protect the privacy of survey participants and has many applications. Different from normal survey, it hides personal privacy by requiring an individual participant to provide a negative option as the answer to the survey question. Answers of participants can further be reconstructed into the distribution of positive options. However, the current negative survey that assumes all participants are 100 percent confident in their answers introduces some precision loss to the reconstruction process. In this paper, we provide an accurate negative survey by allowing participants to annotate their confidence levels to the answers. In particular, we develop a novel Negative Survey to Positive Survey with Answer Confidence Level (NStoPS-CL) algorithm to reconstruct the negative survey with answer confidence level and further increase the accuracy of reconstruction. Experiments demonstrate that NStoPS-CL reliably improves the reconstruction accuracy by testing answer confidence level under different conditions (i.e., dataset size, number of question options and dataset distributions), while balancing reconstruction accuracy and privacy well.