Indexação de chaves compostas por índices aprendidos

Abstract

Through machine learning, the first learned data indexing structure, called Recursive Model Index (RMI), was developed as a possible replacement for traditional indexes (e.g. B+tree), achieving remarkably fast queries. After the introduction of RMI, many frameworks were proposed by various researchers, using frameworks with different design and construction techniques to improve data indexing and make it competitive with existing traditional indices. The learned indexes, according to the authors, can improve query processing performance by avoiding full table scans, thus reducing query latency and the size of the index occupied on disk. The learned indexes, for the most part, index data of the integer type because they are easier to manipulate and extract any patterns that may exist. In this work, we propose the indexing of compound keys, when there is more than one key to be indexed, by evolving an existing learned indexing structure, with the purpose of optimizing the query process. As a case study, a set of data from the educational area was used, containing data from the disciplines. This contribution is expected to improve the learned data indexing process, focusing on compound key indexing.