I am training the new ner model manual from scripat, and I want to know the general acceptable score for stop training and correcting, for eg. 95?or some other scores level?
what is the general acceptable f and total scores for a mature ner model, as one can use in a science artical?
Hi! It is pretty impossible to answer in general, it really depends on the type of entities and domain of the texts you're using. To get some idea, have a look at this repository: NLP-progress/named_entity_recognition.md at master · sebastianruder/NLP-progress · GitHub
Thanks a lot!
You're welcome! FYI, there's a really good write-up about accuracy and utility in applied NLP in Ines' blog post here: Applied NLP Thinking: How to Translate Problems into Solutions · Explosion