Private Weighted Histogram Aggregation in Crowdsourcing

Histogram is one of the fundamental aggregates in crowdsourcing data aggregation. In a crowdsourcing aggregation task, the potential value or importance of each bucket in the histogram may differs, especially when the number of buckets is relatively large but only a few of buckets are of great interests. This is the case weighted histogram aggregation is needed. On the other hand, privacy is a critical issue in crowdsourcing, as data contributed by participants may reveal sensitive information about individuals. In this paper, we study the problem of privacy-preserving weighted histogram aggregation, and propose a new local differential-private mechanism, the bi-parties mechanism, which exploits the weight imbalances among buckets in histogram to minimize weighted error. We provide both theoretical and experimental analyses of the mechanism, specifically, the experimental results demonstrate that our mechanism can averagely reduce $$20\,\%$$ of weighted square error of estimated histograms compared to existing approaches (e.g. randomized response mechanism, exponential mechanism).