What settings must be made to reduce the number of false positives for Term Recognition?

With "largish" documents I get hundreds (sometimes thousands) of false positives for Term Recognition, how can this be reduced to a manageable level? Changing the "Fuzzy match level" does not help very much.