Question answering over knowledge graphs: A graph-driven approach

With the growth of knowledge graphs (KGs), question answering systems make the KGs easily accessible for end-users. Question answering over KGs aims to provide crisp answers to natural language questions across facts stored in the KGs. This paper proposes a graph-driven approach to answer questions over a KG through four steps, including (1) knowledge subgraph construction, (2) question graph construction, (3) graph matching, and (4) query execution. Given an input question, a knowledge subgraph, which is likely to include the answer is extracted to reduce the KG's search space. A graph, named question graph, is built to represent the question's intention. Then, the question graph is matched over the knowledge subgraph to find a query graph corresponding to a SPARQL query. Finally, the corresponding SPARQL is executed to return the answers to the question. The performance of the proposed approach is empirically evaluated using the 6th Question Answering over Linked Data Challenge (QALD-6). Experimental results show that the proposed approach improves the performance compared to the-state-of-art in terms of recall, precision, and F1-score.