Monochromatic Mutual Nearest Neighbor Queries Over Uncertain Data

摘要

As a variant of nearest neighbor queries, mutual nearest neighbor (MNN) search has important applications. In this paper, we formalize the MNN queries in uncertain scenarios, namely the UMNN queries, which return data objects with non-zero qualifying guarantees of being the query issuers' MNNs. We also present some properties of UMNN problems. To process UMNN queries efficiently, we develop two approaches which employ techniques including best-first based uncertain NN retrieval with minimal maximum distance bounding, existing uncertain reverse NN search with geometric pruning, and make use of the reusing technique. An empirical study, based on experiments performed using both real and synthetic datasets, has been conducted to demonstrate the efficiency and effectiveness of developed algorithms under various settings. The experiments also testify the correctness of properties that we proposed as for the monochromatic UMNN problem.