Ontology Learning - Suggesting Associations from Text

Abstract

In many applications, large-scale ontologies have to be constructed and maintained. A manual construction of an ontology is a time consuming and resource demanding process, often involving some domain experts. It would therefore be beneficial to support this process with tools that automates the construction of an ontology. This master thesis has examined the use of association rules for suggesting associations between words in text. In ontology learning, concepts are often extracted from domain specific text. Applying the association rules algorithm on the same text, the associations found can be used to discover candidate relations between concepts in an ontology. This algorithm has been implemented and integrated in GATE, a framework for natural language processing. Alongside the association rules algorithm, several information extraction and natural language processing techniques have been implemented, in which this algorithm is built upon. This has resulted in a framework for ontology learning. A qualitative evaluation of the associations found by the system has shown that the associations found by the association rules algorithm has promising results for detecting relations between concepts in an ontology. It has also been found that this algorithm is dependent on an accurate extraction of keywords. Further, a subjective evaluation of GATE has shown that it is suited as a framework for ontology learning.