Improvements in distance-based indexing

Abstract

This work offers some improvements in the current distance-based indexing techniques. An optimal similarity search algorithm that is adopted from vector-based indexing is shown to be also optimal for distance-based indices. Farther similarity between the two types of indexing is revealed, leading to a general description of search structures. A probabilistic analysis of distance-based tree indices is also shown to be possible, allowing direct comparisons of structures without the need for extensive experimentation. This analysis will lend itself to future improved index construction algorithms.