Question Answering System Using Knowledge Graph Generation and Knowledge Base Enrichment with Relation Extraction

Question answering system is one of the predominant application that used to get the answer for the input natural language question. It reduces the search time to find the answer for particular question. In order to answer a natural language question, question understanding plays a vital role. Understanding a question semantically, leads to get the intention of the input question accurately. In order to understand the question, semantic query graph will be constructed to analyze the question phrases in a structured way by extracting entity and relation phrases. If there is no relevant relation phrases in knowledge base, then query graph cannot be constructed. In order to overcome this unavailability of relation phrases, using n-gram-based attention model and joint learning, relation phrases will be extracted. Using encoder-decoder model, extracted relation phrases along with that corresponding entity and predicate phrases will be converted as triples with unique IDs. Finally, using top-k sub–graph matching algorithm, the answer with the top score will be selected as the final answer. In this proposed question answering system with relation enrichment (QARE) using two datasets such as QALD5 and web questions benchmark, comparative analysis had been carried out with two existing systems. Experimental results show that proposed system achieves better performance than existing systems.