Similarity Searching using Hybrid Technique Permutation Graph and Clustering

• Autores: Rocha, Gerardo; Figueroa, Karina; Reyes, Nora Susana
• Año de publicación: 2025
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; Similarity Search; Metric Space; Distance Calculations
• Institución: Universidad Nacional de La Plata

Similarity searches aim to find the elements most similar to a query within a database. One way to model this is by only considering the distance between the searched element and the rest of the database, which allows it to be defined as a metric space. The advantage is that large amounts of training data are not required, which is the main challenge today. The obvious strategy is to compare the entire database; however, with computationally expensive distances, this can consume time and resources. This work proposes using a data structure to perform searches on it, to reduce the number of distance calculations used. In particular, we propose combining strategies that are proven to be efficient: algorithms based on permutations and those based on clustering and graphs. Experiments show that we can achieve reductions of up to 20% on the number of distance calculations needed for the permutation based algorithm.
Red de Universidades con Carreras en Informática