Distributed k-Nearest Neighbor Search Based on Angular Similarity

Abstract

The k-nearest search algorithm (KNNS) is widely used in those applications based on angular similarity. However, the current KNNS uses Euclidean distance to index dataset and retrieve the search object, which is not suitable for those applications. And existing centralized KNNS does not scale up to large volume of data because the response time is linearly increasing with the size of the searched file. In this paper, a distributed KNNS based angular similarity (DASKNNS) is proposed, which affords the distributed indexing structure to the performance of finding k-nearest neighbor of the search object. DASKNNS firstly proposes the distributed indexing structure (DAS-INDEX) based on angular similarity, which refers to the axis and a reference-line to organize the dataset into some shell-hyper-cone, and linearly stores them at each peer. Then it determines the object peer where the search object locates, makes a search hyper-cone which takes the line connecting the origin point and the search object as the axis, and determines those peers which intersect the hyper-cone. Finally those peers parallelly search the k-nearest neighbors of the search object. The experiment shows that the performance of AS-KNNS is superior to those other KNNS.