Unmanned Aerial Vehicle Knowledge Graph Construction with SpERT

Abstract

Unmanned aerial vehicles are becoming more and more important in the military field. In recent years, global hot spot military events and local conflicts have fully proved its military value. Since knowledge graph is the information basis of intelligence, how to build a high-quality unmanned aerial vehicle knowledge graph is the focus of this paper. In this work, we propose an effective method to construct a knowledge graph from textual data. We first build the schema manually based on our domain knowledge. We then extract RDF triples with SpERT. Third, we disambiguate the instance by string comparison. Finally, We import the knowledge graph into neo4j for visualization. Our team takes part in the No. 10 evaluation task (i.e., military domain-specific knowledge graph construction for military unmanned aerial vehicles) in CCKS 2021. There are two stages in this evaluation, and our approach achieves the second place in the first stage, i.e., knowledge graph quality evaluation and the third place in the second stage, i.e., knowledge graph usage evaluation.