Optimal-Nearest-Neighbor Queries

Given two sets D_A and D_B of multidimensional objects, a spatial region R, and a critical distance dc, an optimal-nearest- neighbor (ONN) query retrieves outside R, the object in D_B with maximum optimality. Let CAR (S_p,p) be the cardinality of the subset Sp of objects in D_A which locate within R and are enclosed by the vicinity circle centered at p with radius d_c. Then, an object o is said to be better than another one o´ if (i) CAR (S_o,o) = CAR (S_o,o´), or (ii) when CAR (S_o,o) = CAR (S_o´,o´) the sum of the weighted distance from each object in S_o to o is smaller than the sum of the weighted distance between every object in So´ and o´. This type of queries is quite useful in many decision making applications. In this paper, we formalize the ONN query, develop the optimality metric, and propose several algorithms for finding optimal nearest neighbors efficiently. Our techniques assume that both D_A and D_B are indexed by R-trees. Extensive experiments demonstrate the efficiency and scalability of our proposed algorithms using both real and synthetic datasets.