It would be nice to divide the cooccurence count by counts of respective two headers, to get a sort of correlation. Now it's quite hard to see anything beyond "most popular things are paired with most popular things" which is obvious.