Athens University Economics and Business (AUEB)
Εννοιολογική ανάκτηση από FDB βάσεις δεδομένων
Abstract
dc:descriptionIn a traditional Database Management System, data retrieval is mainly carried out using keywords, while any conceptual information is ignored. Since any term (concept) can be expressed in more than one word, a significant number of records are ignored during the data retrieval process, as only the a posteriori correlations among the terms are used. Query languages retrieve data containing specific keywords as defined by the user, who does not usually know the content of the Database since it is not possible to identify all the words that are used when defining a concept. Moreover, taking into consideration the fact that the data in a Database can be multilingual, it is obvious that the data set that is finally ignored by the data retrieval process may be important. The present research documents the utilization of the a priori conceptual correlations among terms that exist independently of any Database and are available for processing through the thesaurus.The proposed research approach is based on the FDB model (Framework DataBase) which is an ongoing research project in our department since 1999; it supports dynamically evolving Database environments and allows the definition of any multilingual Database in the FDB universal schema. The present dissertation extends the FDB model so that it can accommodate and manage any multilingual thesaurus. The thesaurus is part of the FDB universal schema and is dynamically defined using appropriate metadata.In the present research, the algorithms for the management of any multilingual or monolingual thesaurus in the FDB model are also proposed. Data correlation with the appropriate thesaurus terms, the enrichment of any existing thesaurus and the creation of a new thesaurus are long and demanding processes that require human intervention. Through the proposed algorithms, data instances of an FDB Database can be automatically correlated with one or more thesaurus terms, while any existing thesaurus can be enriched with terms derived from the FDB Database. Besides, the “core” for a new thesaurus with terms derived from data can be created. Through the present approach, procedures that demand time and effort can be carried out automatically and human intervention by experts is significantly minimized.Moreover, the present dissertation documents the use of the a priori conceptual correlations, which are identified by the underlying thesaurus in data retrieval from any FDB Database. Data search in any FDB Database is carried out using the queries submitted by the user. The present research extends the user queries by using the information provided by monolingual or multilingual thesauri, while data retrieval from any FDB Database is also carried out using the conceptual relationships among terms. Increased flexibility is provided at all stages of data retrieval as the user can choose from a variety of parameters which define the search process: one or more thesauri can be used up to the level desired during data retrieval, including more or less relationship types (equivalent terms, narrower terms etc.). User choices could be differentiated form one search to another, a fact that provides even greater flexibility to the whole model.Additionally, the thesis presents the results from the analysis of a bibliographic Database of the University, proving that the use of a thesaurus during the data retrieval process significantly increases the relative recall while the use of multilingual correlations does not significantly reduce precision. Finally, the findings from the implementation of the proposed algorithms in a real database environment are presented.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Petraki, Evangelia
- Πετράκη, Ευαγγελία
Subjects
dc:subject × 12Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/36754