DISTANCE-BASED INDEX STRUCTURES FOR FAST SIMILARITY SEARCH

摘要

This review considers the class of index structures for fast similarity search. In constructing and applying such structures, only information on values or ranks of some distances/similarities between objects is used. The search by metric distances (satisfying the triangle inequality and other metric axioms) and by nonmetric distances is discussed. Structures that return objects of a base that represent the exact answer to a search query and also structures for approximate similarity search are presented (the latter structures do not guarantee precision, but usually return results close to exact and operate faster than structures for exact search). General principles of construction and application of some index structures are stated, and also ideas underlying concrete algorithms (both well-known and proposed lately) are considered.