Semantic Analysis using Wikipedia, PhD Proposal
Dr. Evangelos Milios – Faculty of Computer Science
Dr. Vlado Kešelj – Faculty of Computer Science
Dr. Jeannette C.M. Janssen – Mathematics & Statistics
“Concepts are the constituents of thoughts. Consequently, they are crucial to such psychological processes as categorization, inference, memory, learning, and decision-making. This much is relatively uncontroversial. But the nature of concepts—the kind of things concepts are—and the constraints that govern a theory of concepts have been the subject of much debate.”, Stanford Encyclopedia of Philosophy
The main challenges in concept representation are (1) what dimensions (a.k.a features, properties) should be used to represent a concept(2) how to project a concept in this space, i.e. how to derive the relevance of each feature.
The answer to the first question can come from domain expert knowledge in form of ontologies, or distributional analysis of domain specific corpora. Ontologies are very labour intensive and do not exist most of the times. The existing ones are also focusing on a few relations and are far from complete. The distributional methods can be quite successful but cannot prevent the computational noise completely and get to the level of of success of ontology based methods in equivalent tasks (for example in semantic relatedness problem).