Analysis of variants of similarity retrieval operators in Metric Access Methods

• Ano de defesa: 2025
• Autor(a) principal: Silva, William Zaniboni
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Billing reduction; Consultas por similaridade; DBMS; Métodos de acesso métricos; Metric access methods; Query variations; Redução de custo financeiro; SGBDs; Similarity query; Variações de consulta
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-06082025-183402/

The Database Management Systems (DBMS) are the tools responsible for storing and retrieving data par excellence. However, their features were developed focusing on scalar data, such as numbers and small text. When considering complex data (non-scalar data) such as images and videos, the resources available in these tools become less useful in the context of data retrieval and data organization. Furthermore, for complex data, recovery based on the similarity between elements makes more sense. The definition of similarity depends on particular characteristics of the application domain, and the Metric Access Methods (MAM) are employed to structure these data, aiming at processing the similarity retrievals more efficiently. The literature presents a few works that integrate open-source Relational Database Management Systems (RDBMS) and similarity queries using MAMs. In the majority, the available similarity retrieval operations are the most basic, and they tend to work only on complex data. The objective of this masters thesis is to define and evaluate useful variations of similarity queries that should be accelerated and create the support for query execution using Metric Access Methods (MAM) within the execution engine of the Relational DBMS, and assess potential applications of these concepts in NoSQL systems. The focus will be on MAM and the development of its retrieval operators, so that operations performed in DBMS can be varied and optimized (it is not within the scope of this work to integrate MAM into the RDBMS). As a result, this masters thesis: defines and details the applicability of several similarity query variations; develops a new MAM, called Include-Slim, which allows accelerating these queries within an RDBMS; defines and develops a new framework called Similarity-Slim, which enables similarity queries to be accelerated and their financial costs reduced when executed on a NoSQL cloud database using document collections; and presents two new post-processing algorithms, Slim-Bloat and Slim-Replace, that can be applied to any MAM to generate a better structure. The evaluations performed show that Include-Slim can accelerate query variations by up to four orders of magnitude, and Similarity-Slim can reduce the financial cost of queries by up to three orders of magnitude. To the best of our knowledge, this is the first work that addresses accelerating the presented variations and the first one applied to reduce the financial costs in NoSQL environments.