An interesting peek in Microsoft’s kitchen (the Beijing labs, by the looks of it): Probase and ReadWriteWeb writeup on it. It’s a very large web-fed knowledge-base, including concept hierarchies (2.7 million concepts, 4.5 million subclass relations, 16 million instances). Including all major knowledge sources (Freebase, WordNet, Cyc, DBPedia, Yago, a.o.), with pretty well researched quality measures. Unfortunately, none of the data is Linked in any way, none of this available, let alone in some standard format.This is interestingly different from IBM’s Watson knowledge base. That is mostly filled with knowledge extracted from linguistic sources (although structured data does play a limited role). Probase seems to rely much more on structured knowledge sources.