Evaluation of Automatic Linking Strategies for Wikipedia Pages

Wikipedia contains an enormous amount of human knowledge. The wide range of covered topics is hierarchically organized in categories and strongly inter-linked. Its structure, its size and the fact that it is generated by humans are the reasons for the attention Wikipedia receives from researchers in different fields. Especially the link structure of Wikipedia is of huge importance not only for humans browsing the collection, but also as a resource for bootstrapping machine intelligence and the semantic web. Motivated by the fact that manual maintenance and creation of hyperlinks is labor intensive, this paper explores properties for automatic link creation between Wikipedia pages in this paper. Focusing on ad-hoc linking approaches we evaluate linking strategies on the word as well as on the document level using a standard test data set. As it is shown, rather simple approaches yield to reliable results and may be applicable in different application scenarios. Disambiguation strategies based on standard IR techniques help to boost accuracy delivering reasonable results.