I've been thinking about the RDF data model a lot lately. Including reading up on SPARQL. Initially I didn't like it. However after a while it struck me that the RDF model + SPARQL actually matches most of what I do with databases a lot more than the relational model + SQL.
The problem with the relational model is that I normally work with "resources" that consists of data items that are generally accessed at the same time and are tightly related, yet if I want to get a properly normalised relational database, I end up with insanely complex queries if I want to gather all the information back together again.
With SPARQL queries on RDF data this suddenly becomes simplicity itself because I'm not required to try to figure out a way to map the data I want to query about back into a single row per entity - instead I'm figuring out a way to find the triples I need, and optionally provide a pattern for extracting just the data I want, or alternatively return all the found RDF triples.
The question is whether performance will be good enough - I haven't yet had a chance to experiment with a large scale RDF model.
The ease of querying for a graph of data, as opposed to being constrained into a very simplistic row/column model is a compelling incentive to spend more time on it.