KG4ASTRA: question answering over Indian Missiles Knowledge Graph

Natural language, being unstructured, makes the tedious task for building a model to parse it into a query language successfully. An incorrect query for a particular question would lead to an absurd answer. As seen in many semantic parsing approaches, the inaccurate answering of complex questions increases significantly. This leads to many novel effective strategies that can apply to semantic parsing approaches to accurately parse complex natural language questions and generate an appropriate answer over a large knowledge graph. The defense system in a particular country is one of the most crucial components, and the lack of a proper defensive domain interface makes it a significant motivation. Hence, a knowledge graph has been constructed to collect and enquire about all information in one place. In this paper, KG4ASTRA has been designed with a Missile Knowledge Graph consisting of 177 entities linked using 400 relationships. A query-answering model then utilizes this manually created Missile Knowledge Graph, which generates tabular or graph representation for the natural language question. Neo4j platform has been used to prepare the knowledge graph, and Cypher queries are used to execute queries. The modeled knowledge has been evaluated by executing natural language queries on KG4ASTRA query-answering model and compared the search results with other existing knowledge graphs. As of our best knowledge, neither the knowledge graph nor the question-answering model has been designed for Indian Missiles. As a future scope, the proposed knowledge graph will be connected with the existing knowledge graph and extended to automatically extract domain-specific entities.