Constrution Strategies on Metri Strutures for Similarity Searh

Abstract

The List of luster LC is an eetive tehnique to index high dimension metri spaes. LC is an array-type struture for similarity searh based on lustering. Sparse Spatial Seletion SSS is a new struture based on pivots for similarity searh in metri spaes. This array-type struture has shown good performane during the searh as ompared to other methods. This work shows dierent onstrution strategies on LC ; for instane, the use of SSS as a general seletion method of pivots or enters, among others. The artile also shows the advantages of the use of other tehniques like keeping the distane between the ob jets and the luster enter, apart from revising the eets of reursive appliation of suh methods. Finally, the inuene of the use of Voronoi partitions for the distribution of the ob jets within the struture will be shown. Preliminary experimental results show that the new versions of LC have a better performane in terms of distane evaluation than the original data struture and other renowned strutures. Keywords: Databases, Data strutures, Algorithms, Metri Spaes, Similarity Searh.