The problem is very much like in the previous subsection, but we consider the percentage of co-occurrences with more frequent left part for both parts being NOT in the top-N-words.
Second line in table:
select 10, (select 1+count(*) from co_n where w1_id>110 and w2_id>110 and w2_id>w1_id) / (select 2+count(*) from co_n where w1_id>110 and w2_id>110);
5.1.9.1 Skewness in NN co-occurrences I