Bruno - I had the same reaction. On a bit further thought, I wonder if he may be referring to some of the NSA's datasets, however you may have trouble getting access to them.
I also have been feeling that the reactions to the statements about facebook popularizing the term social graph is overblown. In fact, I had hoped - my eyes having prematurely saccaded to phrases such as "16th century" and "Francis Bacon" that the phrase "Of course, you all
know that" was going to be followed by some data on the actual early uses of the term "social graph". Not seeing such a thing, I looked it up in google ngrams, which finds the first use in 1922, with spikes in frequency every 10-15 years from 1950 to 2000. In fact, I was initially surprised that the degree of the spikes seem to be going down over time. However, after a little thought, I suspect this is an artifact of G Ngrams displaying only a percentage of the corpus with hits. So I think the reason there are spikes is because there are 2 books per year instead of 1 that use the word combined with the continually increasing number of books published each year. With only 1-3 hits per year up to 2000, it is not at all hard to regard that a statement that "popularization" occurred in 2007 in spite of fairly extensive noncommercial use prior to that. People just weren't using it in books or putting those books in ngram. Identifying a point of "popularization" is a bit challenging, and I am not sure that measuring periodiical's frequency of usage is the best way to move forward. But it's hard to deny that the term sees much greater use than once or twice a year and enough use to remain active.