Re:What we learned from using GPT for 500k+ classifications
by dr_kiszonka on Friday July 14, 2023 @12:49AM
(#36719589)
If I understand your approach correctly, you could represent relations between words as graphs and use graph/network similarity measures (of which there are tons) to possibly get over the 92%. (Or not, I have never tried it.)
[ Parent
| on HN ]