A subset of DBpedia dataset

Hello everyone,
I am doing a study which requires a small and complete knowledge graph. The graph should have around 6 million triples and there is no cap on the number of relations per se.
Can anyone please guide me towards a subset of DBpedia dataset which is approximately the size I need and is as complete (or disjoint from the original DBpedia knowledge graph) as possible?

Thank you.

Dear @visaini,
there is a method described in http://jens-lehmann.org/files/2009/dllearner_sparql.pdf by iteratively crawling the graph with sparql.

small and complete is a contradicting request, i.e. what does complete mean?
– kurzum