KB-based Question Answering

Several knowledge bases like Dbpedia, Freebase have been published in recent years. They are compiled with public efforts, and linked data based. These knowledge bases are vast graphs consist of triples following RDF standards format. RDF triples are triplets of subject, predicate and object. It is atomic unit of knowledge which knowledge bases contain. Most recent knowledge bases can be used for integration or co-operation of other knowledge bases. It enables us to use several knowledge bases as single one.

However, to query these knowledge bases users have to comprehend how knowledge bases are organized and how to write structured queries for knowledge bases. We are developing a system which can ask and retrieve answers from linked data based KBs with natural language questions.

First, the system analyses the question, and identifies entities and relations between them. And extract pseudo triples from analysis result. Secondly, compose structured query from the pseudo triples and retrieve results from the knowledge base. Finally, the system ranks retrieved results with their relevance and provide the user the most relevant answer.