GikiCLEF focused on the evaluation of the reasoning capabilities of systems to provide right answers for geographically-challenging topics. As we did not have previous experience in question answering, we participated in GikiCLEF with the goal of understanding best practices in extracting answers from documents though a hands-on experience. We developed a prototype that used DBpedia and the Portuguese Wikipedia as raw knowledge resources, and created a new world geographic ontology, also derived from the Wikipedia, for supporting geographic reasoning. The prototype was not ready to produce automatic runs at the submission deadline, but we observed that the best results we could aspire with the devised approach would be under our initial expectations. Wikipedia and DBpedia information coverage and location revealed to be much different from what we were initially expecting. We learned that when planning on improving a GIR system with modules aimed to reason over the query before steppi...
Nuno Cardoso, David Batista, Francisco J. Ló