Indexação de dados métricos em bancos de dados complexos

Abstract

We live in a moment where information and knowledge of this information play a significant role in the development of humanity. With all this information present and in the most varied types, from texts, numbers, videos, images, and etc, the need arises to store them and retrieve them when necessary in an efficient and fast way. Current SGBDRs have great support for indexing traditional data, such as text and numbers, but it does not achieve the same efficiency when dealing with complex or non-traditional data. With the technological evolution, the need arises to store and retrieve such data, for this are developed metric indexing structures, which are access methods of a database management system. When performing a search operation with this data, it is not possible to use total order comparators such as =, <, >, but rather to perform similarity queries, which generally depend on a distance function to execute. Our goal is to optimize a metric indexing structure known as FMM-tree, improving its efficiency and performance in the insertion and recovery of complex data. To achieve this goal, we will use tools such as Arboretum and Artemis, which are specific frameworks for indexing and feature extraction structures, and similarity query models such as queries by scope and number of neighbors. In the experiments it was possible to observe that after an application of the triangular version in the query similarity methods by span and a substitution variable for the data systems, a more structured structure with the best performance and more efficient in the return of the data.