How does spacy use word embeddings for Named Entity Recognition (NER)?

Question

I am trying to train the NER model with spaCylocation help , (person) names and organizations. I am trying to understand how spaCyobjects are recognized in the text, and I could not find the answer. From this question on Github and this example , it is that spaCy uses a number of functions that are present in the text, such as POS tags, prefixes, suffixes and other characteristic and text functions in the text to train the average perceptron.

However, it doesn't appear anywhere in the code that it spaCyuses GLoVe attachments (although every word in a sentence / document seems to have them if they are present in the GLoVe body).

My questions -

• Are they used in the NER system?
• If I were to switch word vectors to a different set, should I expect performance to change in a meaningful way?
• Where in the code can I find out how (if all) spaCyuses the words vectors?

I tried looking at the Cython code, but I cannot figure out if the marking system uses dictionary attachments.

Answer 1

spaCy NER, CNN. , , spaCy, , NER . GloVe, , "" , , link.

. spaCy docs, Github.