In my experience, a graph often represents mental models (and "conceptualizations") much more well than the tabular based onces, which are a good practical approach in many cases, but not all. I like to use RDF/SPARQL based models where possible (in order to introduce a certain level of standardization between different domains) or a property.graph approach (designin a proper physical model for domain, using cypher/gremlin and in general tools like neo4j/tinkerpop)