You must log in to comment.

in reply to @amypercent's post:

i haven't used it since, like, 2017... but even back then it was pretty nice and ergonomic, imo the API was designed thoughtfully. like all NLP stuff, the quality of the output is massively dependent on your models and the data you're trying to process, but for my purposes of ripping apart corpuses and restitching them back together in weird ways, it worked fine. i think the dependency parser was my favorite part of it.

a bit, but not a ton - i don't like that the entity extraction model is neural-net based because the training cycle time are awful and it makes stability testing and 5-fold cross validation a huge pain in the ass.

that said, it seems to work pretty well