Pseudo validation to extract Schematic information

I would like to tag an instance to  identifying occurences of schema aggregates (e.g groups or complexTypes) within the instance.

I haven't got a schema to start with so plan A is to generate one with Examplotron and then see if I can tag the instance with whatever aggregates I get from it.

The big idea is pattern detection and for starters I take as  patterns things that equate to some sort of schema aggregate.

Thought best to ask to see if I can get a head start on this. Happy to follow pointers to research papers if there are any.

thanks.