Publication at Faculty of Mathematics and Physics |
2019
Abstract
Nowadays, the metric space properties limit the methods of indexing for content-based similarity search. The target of this paper is a data-driven transformation of a semimetric model to a metric one while keeping the data indexability high.
We have proposed a genetic algorithm for evolutionary design of semimetric-to-metric modifiers. The precision of our algorithm is near the specified error threshold and indexability is still good.
The paper contribution is a proof of concept showing that genetic algorithms can effectively design semimetric modifiers applicable in similarity search engines.