Solving the Top-K problem for sequence counting using differential privacy

Online databases currently store large quantities of sensitive information. This information includes sequences, for example trajectories or network packet flows. While interesting conclusions might be drawn from analyzing it, the process is not trivial, as the results risk compromising the privacy of participating users. Differential privacy proposes a solution for the safe analysis of such data, using statistical means to perform anonymization. We look at two differential privacy algorithms for sequence counting and see how they perform when adapted to solve the top-K problem, which selects the greatest K values from a set. We analyze the performance of both algorithms in a variety of scenarios, and compute the precision and errors of the results. We conclude with a series of recommendations on which algorithm is best suited to solving the top-K problem depending on user goals.