A Social Approach To The Construction Of Science Terminological Ontologies

Science terminological ontologies play an important role in research management, e.g., research funding management and research assessment exercise. We proposed a social approach to construct science terminological ontologies for a given research domain, to facilitate information retrieval or information browsing in the research management process.Previous terminological ontology construction studies relied heavily on domain decision makers such as managers in research funding agencies, journal editors or professional ontology engineers. It is possible to reduce domain decision makers’ burden by the continuous involvement of domain scholars for the ontology development and maintenance process. However, previous studies did not encourage domain researchers to participate actively in the process of terminological ontology construction. In this research, we proposed a social voting approach to design and develop science terminological ontologies. The use of the social voting approach facilitates the effective development of terminological ontologies and easy updating for future use.To create a domain specific terminological ontology, the following two issues are critical, (1) constructing a domain vocabulary, i.e., identifying all important keywords in a given research domain; (2) generating topic hierarchies from those identified keywords. These issues were addressed in this dissertation.Following the design science research approach, this research included a design part and an evaluation part. In the design part, we first proposed a social voting approach to construct domain vocabularies, based on the theory of linguistic arbitrariness borrowed from linguistics. Then, according to the framework that generating topic hierarchies by using term similarity and term specificity, we designed a topic hierarchy generation method by integrating the LDA topic modelling technique and Sanderson and Croft’s Subsumption Hierarchy Model.In the evaluation part, this study first evaluated one of the constructed domain vocabularies by a two-part questionnaire following the same procedure proposed by Barki et al. The results showed good performance of the social voting approach to construct domain vocabularies. Then, the LDA was evaluated by conducting experiments on a gold standard data set, and experimental results showed the LDA representation performed better than the traditional TFIDF vector space model. Next, the Subsumption Hierarchy Model was evaluated according to the information entropy criterion; experimental results showed it performed better than the KL divergence measure. Finally, a user study was conducted to evaluate the whole topic hierarchy generation method with promising results.This research has both theoretical and practical contributions. From the theoretical aspect, it first presents a new social voting approach based ontology development framework, which provides a unified and extendable theoretical framework for scientific domain vocabulary construction. Second, it presents a new topic hierarchy generation method which combines the LDA and the Subsumption Hierarchy Model to determine the hierarchical relationship of keywords.From the practical aspect, our approach provides an effective way of organizing domain knowledge to achieve higher performance in research management tasks. It provides many relevant agencies an alternative method to construct science terminological ontologies.