Extracting term relationships from Wikipedia

Abstract

When looking at the relationship between two terms, we are not only interested on how much they are related, but how we may explain this relationship to the user. This is an open problem in ontology matching, but also in other tasks, from information retrieval to lexicography. In this paper, we propose a solution based on snippets taken from Wikipedia. These snippets are found by looking for connectors between the two terms, e.g. the terms themselves, but also terms that occur often in both articles or terms that link to both articles. With a user study, we establish that this is particularly useful when dealing with not well known relationships, but well-known concepts. The users were learning more about the relationship and were able to grade it accordingly. On real life data, there are some issues with near synonyms, which are not detected well and terms from different communities, but aside from that we get usable and useful explanations of the term relationships. © 2013 Springer-Verlag.