Fairest Neighbors: Tradeoffs Between Metric Queries

TL;DR

This paper introduces a method for metric search that balances multiple query objects fairly using inequality measures, improving efficiency over traditional linear scans and combination methods.

Contribution

It proposes a novel approach to multi-object metric queries using inequality measures, enabling faster and fairer tradeoffs in search results.

Findings

Significant speedup over linear scan methods

Effective balancing of multiple query objects

Empirical validation on index structures

Abstract

Metric search commonly involves finding objects similar to a given sample object. We explore a generalization, where the desired result is a fair tradeoff between multiple query objects. This builds on previous results on complex queries, such as linear combinations. We instead use measures of inequality, like ordered weighted averages, and query existing index structures to find objects that minimize these. We compare our method empirically to linear scan and a post hoc combination of individual queries, and demonstrate a considerable speedup.