“…The corpus was split into three different subsets, as in [6] , but in a different and more realistic way, since the three subsets are completely independent here. In [6] , the elements from the cohort set not selected to perform the score ratio (i.e., those different from the N with the lower scores) were used for score normalization. This was done so as to obtain a large and significant test set (183 elements) and a large ( H = 150 elements) cohort set.…”