Poster – Paper 555

Question Answering Benchmarks for Wikidata

Dennis Diefenbach, Thomas Pellissier Tanon, Kamal Singh and Pierre Maret

Poster

clock_event October 23, 2017, Poster and Demo Reception, 18:30-21:20
house Festsaal 1
download Download paper (preprint)

Abstract

Wikidata is becoming an increasingly important knowledge base whose usage is spreading in the research community. However, most question answering systems evaluation datasets rely on Freebase or DBpedia. We present two new datasets in order to train and benchmark QA systems over Wikidata. The first is a translation of the popular SimpleQuestions dataset to Wikidata, the second is a dataset created by collecting user feedbacks.