Abstract

In this paper we present a question answering system using a neural networkto interpret questions learned from the DBpedia repository. We train asequence-to-sequence neural network model with n-triples extracted from theDBpedia Infobox Properties. Since these properties do not represent the naturallanguage, we further used question-answer dialogues from movie subtitles.Although the automatic evaluation shows a low overlap of the generated answerscompared to the gold standard set, a manual inspection of the showed promisingoutcomes from the experiment for further work.

 

Quick Read (beta)