Research in Natural Language Processing (NLP) has made tremendous progress in the last two decades by employing data-driven techniques. However, further major advances can be achieved by integrating linguistic, domain and world knowledge into statistical approaches.
In this dissertation, a methodology is presented to extract this knowledge from Wikipedia, a resource which has attracted the attention of many researchers in the Artificial Intelligence (AI) community, mainly because it provides semi-structured information and a large amount of manual annotations.
The proposed approach uses the category system found in Wikipedia as a conceptual network. Semantic relations between categories are labeled to produce a large-scale taxonomy. This resource is evaluated by comparing it with Cyc and WordNet, as well as through computing semantic similarity between words and using semantic similarity measures as features for a state-of-the-art co-reference resolution system. The results show that this taxonomy can be successfully deployed for NLP tasks and represents a valuable semantic resource for AI applications.
