Finding RkNN Straightforwardly with Large Secondary Storage

Abstract

In this paper, we proposes an efficient algorithm for finding reverse k nearest neighbor (RkNN) search. Given a set V of objects and a query object q, a RkNN query returns a subset of V such that each element of the subset has q as its kNN member according to a certain similarity metric. Early methods pre-compute NN of each data objects and find RNN. Recent methods introduce index based on the mutual distance between two objects. Our method can find RkNN for any k straightforwardly with constant running cost. It can be applied to any RkNN searches whenever the mutual distance between objects can be figured out. It does not require the triangle inequality even. It is also based on pre-compute information, under the assumptions that secondary storage (hard disk drive) is cheap and the current computers are powerful enough so their spare power can be used to update data offline. We evaluate the efficiency and effectiveness of the proposed method.