National and Kapodistrian University of Athens
Θεμελιώδεις αρχές αναλυτικής επεξεργασίας πληροφοριών
Abstract
dc:descriptionA common operation involved with the majority of algorithms relevant to On-Line Analytical Processing is aggregation, which can be extremely time-consuming if applied over large datasets. To overcome this drawback, scientists have proposed the precomputation and materialization of a large volume of aggregated data into a structure called data cube. Nevertheless, the construction and usage of the data cube itself has been found very demanding in terms of computational and storage resources. In this dissertation, we study this problem in depth and propose comprehensive suites of scalable algorithms that perform efficient cube construction, storage, query answering, incremental updating, indexing, and caching. Our extensive experimental evaluation indicates that our solutions are viable even when applied over very large datasets with arbitrary hierarchies. Some key points in our work include the introduction of a novel storage scheme for cubes that is based on the use of row-id references, a new external partitioning algorithm, an efficient construction plan, and the use of on- demand approaches for processes that are too expensive to perform in a given window of time. A unique property of all our algorithms is their compatibility with the relational model, which makes them easy to incorporate into existing relational servers. Finally, as an application in data mining, we study the usage of aggregate queries for solving the problems of feature selection and classification and propose a disk-based, lazy, and accurate solution, which exhibits great potentials in a broad range of applications.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Morfonios, Konstantinos
- Μορφονιός, Κωνσταντίνος
Subjects
dc:subject × 14Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/22948