Freeqa-Hybrit question answering system on linked data

Question answering deals with natural language questions that can be answered using knowledge resources. With increasing in linked data resources such as DBpedia, Freebase and Yago, question answering systems start to use these types of resources more intensively. IBM Watson, Google Knowledge Graph, Wolfram Alpha, Siri can be example for these systems. Question answering systems over linked data basically try to answer using structured data. However, not all questions can be answered using only structured data. In such cases, textual or unstructured data, which are present in linked data as abstracts, labels, etc., can be utilized to answer such queries. This is called hybrid question answering. In this thesis, there is a different approach to hybrid question answering called FreeQA. This system utilizes natural language tools to map natural language questions to query templates and then to SPARQL queries, which can then be executed on linked data to answer the questions. FreeQA first builds a query tree with possible query templates and uses greedy search strategy to form the final SPARQL query. FreeQA can answer 37 of the 45 hybrid questions in the dataset, of which 33 of them are answered correctly and 4 of them are partially correctly answered.