I use dataset extensively. Once each DOM element is assigned a unique key id, many things get simplified: DOM manipulation, client server state sync, event handling, etc. Now if an additional data layer can resolve key id to semantic metadata, the only problem left would be to expose that data so third parties can read from it. So, the problem could be solved with three data points: url, dom key and the resolved metadata value.

What's really needed isn't a new protocol or set of rules and conditions. OpenGraph probably works just fine. It's a dedicated global database of metadata. A web white pages directory. As OpenGraph is perhaps dominant, it may make sense for someone like Facebook to provide this. Historically, third party non-profit services have not been popular.

All of this is of course contingent on the assumption that content hosters provide that metadata to begin with ;)