Question Answering with DBpedia Based on the Dependency Parser and Entity-centric Index

The emerging Linked Open Data provides an opportunity to answer the natural language question based on knowledge bases (KB). This study proposes an approach to question answering (QA) on the DBpedia dataset. After parsing the question by a dependency parser, we locate the entity mention and property mention with predefined templates. We propose an entity-centric indexing model to help search referent entities in KB. After obtaining the referent entities, we expand the property mention with WordNet and ConceptNet to find the referent properties of the returned entities. The values of the referent property are then considered the answer to the question. Evaluations are performed on DBpedia version 2015. Results show that our approach reaches 46% precision when the top-10 entities are returned in the final QA stage. The evaluation tests show that our approach is promising in dealing with QA in Linked Data.