Diplomarbeit DIP-3015

Bibliograph.
Daten
Keller, Christine: Theoretical and Practical Perspectives on Ontology Learning from Folksonomies.
Universität Stuttgart, Fakultät Informatik, Diplomarbeit Nr. 3015 (2010).
113 Seiten, englisch.
CR-Klassif.H.3.3 (Information Search and Retrieval)
I.2.6 (Artificial Intelligence Learning)
Kurzfassung

In this work, a wide range of aspects of ontology learning from folksonomies is presented and discussed. Applying ontology learning methods on folksonomies from Social Web applications aims at exploiting the "wisdom of the crowds" and incorporating it into the formal and explicit representation of an ontology. Folksonomies are up-to-date and can quickly adopt new topics that the community is interested in. This dynamics can be used to support the maturing of ontologies and to keep them up-to-date as well. This work reviews the methods for ontology learning and their adaptation to learning from folksonomies. The different results of already implemented approaches are discussed. Apparently, the possible results of ontology learning from folksonomies do not directly fit into a formal representation of an ontology. Therefore, additional ontology engineering steps are necessary. This raises the question of how much information is possibly contained in a folksonomy and how much information can therefore be extracted from it at all. Another part of this work therefore focuses on a theoretical perspective on folksonomies. In order to describe the expressiveness of a folksonomy, its relational characteristics are explored, using the relational algebra that comes from the field of relational databases. With the relational algebra, different possible views on the information in a folksonomy can be described. It becomes apparent that in fact the informational content of a folksonomy is limited. Nevertheless, it still can be valuable to extract the information that is present in the folksonomy. Therefore several ontology learning methods are implemented on a test data set. After applying cluster analysis and association rule mining on the test data, the drawbacks of both approaches can be determined. On the one hand, a clustering analysis results in very fuzzy groups of related tags. On the other hand, the association rule mining approach has problems finding many meaningful rules of related tags, because of the structure of the data and the different kinds of noise in the tag set. This leads to a combined ontology learning approach (COLA), which utilises the strength of both approaches to overcome their drawbacks. The results of the COLA approach are discussed in detail, outlining further development possibilities.

Volltext und
andere Links
PDF (2138891 Bytes)
Zugriff auf studentische Arbeiten aufgrund vorherrschender Datenschutzbestimmungen nur innerhalb der Fakultät möglich
Abteilung(en)Universität Stuttgart, Institut für Informatik, Intelligente Systeme
BetreuerKlenk, Sebastian
Eingabedatum18. November 2010
   Publ. Informatik