Constructing lightweight ontologies from a conversational agent

Abstract

An ontology is a knowledge representation structure which models the information about a topic by creating a network of concepts linked together by relations. Ontologies are machine-readable and they encapsulate the shared knowledge of a domain. Lately, ontologies have been the subject of a lot of study since they are a fundamental component in the Semantic Web vision. The major drawback of ontologies is that they are difficult to construct mainly due to the knowledge acquisition bottleneck. The knowledge acquisition bottleneck is hindering the proliferation of ontologies and thus the progress of the Semantic Web. This problem is the motivation behind our thesis, where an attempt is made to learn ontologies from text available on the web. Various researchers believe that a fully-automatic approach to ontology learning is currently unrealistic, therefore we opted for a semi-automatic ontology learning system. The presented system is called Dial-an-Ont because it involves a dialogue system which learns ontologies. Dial-an-Ont takes Wikipedia articles as input and outputs lightweight ontologies. Wikipedia articles are used because tl1ey are written by multiple authors, therefore, their knowledge can be considered as shared knowledge. A lightweight ontology is a type of domain ontology which is poor in ontological axioms. Dial-an-Ont is composed of two main modules. A fully automatic high recall module which utilises WordNet and VerbNet as background lexical knowledge bases while extracting semantic relations from typed dependencies. The Stanford parser generates typed dependencies from text onto which specially designed patterns are matched to extract semantic relations. The semantic relations produced by the first module are used as input to the second semi-automatic module of the system which is focused on precision rather than recall. The second module is a dialogue system which initiates a conversation with the user. Based on the replies provided by the user, the dialogue system prunes away irrelevant semantic relations, and transforms the relevant ones into an ontology. The system was evaluated by comparing ontologies generated from Wikipedia article with the linked data present in DBpedia about the same articles. Encouraging results were achieved.