Subspace similarity search using the ideas of ranking and top-k retrieval

Abstract

There are abundant scenarios for applications of similarity search in databases where the similarity of objects is defined for a subset of attributes, i.e., in a subspace, only. While much research has been done in efficient support of single column similarity queries or of similarity queries in the full space, scarcely any support of similarity search in subspaces has been provided so far. The three existing approaches are variations of the sequential scan. Here, we propose the first index-based solution to subspace similarity search in arbitrary subspaces which is based on the concepts of nearest neighbor ranking and top-k retrieval.