A Semantic Similarity-based Subgraph Matching Method for Improving Question Answering over RDF.

RDF question/answering (Q/A) system can explore RDF data by translating natural language questions into SPARQL queries. In this poster, we design a generation-and-ranking approach to translate natural language questions into SPARQL queries based on semantic similarity between questions and SPARQL queries. In the generation stage, we employ a directed super semantic query graph to extract the structural query intention of the question, based on which Q/A on RDF is reduced to the graph matching problem. After building the query graph, we generate a set of candidate queries of the question. In the ranking stage, we rank the query in the candidate query set according to the semantic similarity between the query and question. Finally, we pick the query with the highest semantic similarity to the original question.