Indexing Probabilistic Nearest-Neighbor Threshold Queries

Data uncertainty is inherent in many applications, including sensor networks, scientic data management, data integration, location- based applications, etc. One of common queries for uncertain data is the probabilistic nearest neighbor (PNN) query that returns all uncertain objects with non-zero probabilities to be NN. In this paper we study the PNN query with a probability threshold (PNNT), which returns all objects with the NN probability greater than the threshold. Our PNNT query removes the assumption in all previous papers that the probability of an uncertain object always adds up to 1, i.e., we consider missing prob- abilities. We propose an augmented R-tree index with additional proba- bilistic information to facilitate pruning as well as global data structures for maintaining the current pruning status. We present our algorithm for eciently answering PNNT queries and perform experiments to show that our algorithm signicantly reduces the number of objects that need to be further evaluated as NN candidates.