Anti-patterns are among my favorite topics 馃檪 There are good indications for using KGs: e.g., having linked data, having suitable ontologies/controlled vocabularies which are well-defined, etc.,, and of course one could also create or extend vocabularies as needed. KGs are great for representing relationships; in contrast RDBMS have relatively static definitions of relationships. The intended use case is also importance to consider: we understand about uses for embedding, link prediction, entity linking, neighborhood search, disambiguation, and we could go on with articulating others ... but if the intended use case doesn't fit, then some research/invention may be required. That can be difficult, expensive, unlikely, etc.
