Consider a data set that is naturally hierarchical and path related relationships are the central purpose of the data. Let’s say a genealogical database like some services run.

I can see a way of doing it with tags but mostly what I’m picturing has to add additional metadata to the tags that essentially represents the graph and has to add extra logic for resolving all of it.

If stored as nodes and edges you also have the capacity to add additional features to the relationships easily and naturally. That allows you do induce various subnetworks by edge flavor pretty easily. Network metrics such as centrality and clustering also fall out naturally.

Again, you can do it in tags because you can represent the network data as a table, which would in turn be translatable into possibly some long and complex tags. Or maybe there’s a more natural way, but for me the graph is easier to think about and write interesting algorithms for.