Μοντελοποίηση Και Θέματα Λειτουργίας Βάσεων Προτύπων

Abstract

Databases both as a research and an application field have already several mature characteristics by computer science standards. Still, in the recent years the requirements are changing substantially as basic database technology is employed in several new fields, like biomedical data, data in peer-to-peer systems, data from sensor networks etc. The users that handle huge volumes of data want to explore them efficiently and view results in a concise way. The most popular methods for discovering information hidden in the data, or, in our terminology, for discovering patterns, come from the data mining field. Still, until now, data mining has not the practical success one would expect. An important problem that is identified and addressed in the thesis is the lack of tools for storing and handling patterns that come from the different data mining methods. In this thesis we contribute towards developing an environment for managing patterns, in the following ways: a)We propose a conceptual architecture and a logical model for an environment that handles both patterns and data, the pattern warehouse. The pattern warehouse facilitates the management of patterns, preserves the pattern semantics and allows the discovery of properties of the dataset by performing operations between patterns. b) We study the relationship between patterns and data and propose two different ways of modeling it. c) Having the relation between data and patterns as a starting point, we propose two indices for the efficient evaluation of containment queries; the Hybrid Trie Inverted file (HTI) and the Ordered Inverted File (OIF). By performing extensive experiments we establish the superior performance of both the HTI and the OIF, over the inverted files. Through the experimental evaluation we demonstrate the different gains and costs of each index.