Does training a random forest on short vs long samples improve the results on short vs long comparisons?

[stephaniereinders]

The current random forest is trained on pairwise comparisons between paragraph-long samples. Previous experiments indicate that using a random forest trained on comparisons between short and long samples might improve the accuracy when testing on short versus long comparisons.