QAnswer - Enhanced Entity Matching for Question Answering over Linked Data

QAnswer is a question answering system that uses DBpedia as a knowledge base and converts natural language questions into a SPARQL query. In order to improve the match between entities and relations and natural language text, we make use of Wikipedia to extract lexicalizations of the DBpedia entities and then match them with the question. These entities are validated on the ontology, while missing ones can be inferred. The proposed system was tested in the QALD-5 challenge and it obtained a F1 score of 0.30, which placed QAnswer in the second position in the challenge, despite the fact that the system used only a small subset of the properties in DBpedia, due to the long extraction process.