Knowledge graphs (KGs) have become a central component of artificial intelligence and the semantic web, yet continues to be vaguely defined and distributed. In the literature, such views can be found in various definitions differing between ontological and logical formalization to a more practical, data-driven approach. This study collects and analyzes a corpus of definitions published between 2012 and 2025 to highlight their convergences and divergences. The results reveal three structuring dimensions: formal representation through triples, semantics supported by ontologies and their role as cognitive devices. An integrative definition is proposed, aiming to stabilize the concept and provide a shared theoretical foundation for future research.