Selective data acquisition for probabilistic K-NN query

Recently, management of uncertain data draws lots of attention to consider the granularity of devices and noises in collection and delivery of data. Previous works directly model and handle uncertain data to find the required results. However, when data uncertainty is not small or limited, users are not able to obtain useful insights and thereby tend to provide more resources to improve the solution, by reducing the uncertainty of data. In light of this issue, this paper formulates a new problem of choosing a given number of uncertain data objects for acquiring their attribute values to improve the solutions of Probabilistic k-Nearest-Neighbor (k-PNN) query. We prove that solutions must be better after data acquisition, and we devise algorithms to maximize expected improvement. Our experiment results demonstrate that the probability can be significantly improved with only a small number of data acquisitions.