ISOFT at QALD-5: Hybrid Question Answering System over Linked Data and Text Data.
Seonyeong Park, Soonchoul Kwon, Byungsoo Kim, Gary Geunbae Lee · CLEF (Working Notes) · 2015
We develop a question answering system over linked data and text data. We combine knowledgebase-based question answering (KBQA) approach and information retrieval based question answering (IRQA) approach to solve complex questions. To solve this kind of complex question using only knowledgebase and SPARQL query, we use various methods to translate natural language (NL) phrases in question to entities and properties in knowledgebase (KB). However, converting NL phrases to entities and properties in KB many times usually has low accuracy in most KBQA. To reduce the number of converting NL phrases to words in KB, we extract clues of answers using IRQA and generate one SPARQL based on the extracted clues and analyses of the question and the semantic answer type.