Named entity Recognition Using Bi-LSTM-CRF

There was a long thread on this here.

It seems like, if you want to recognize words that aren’t in your training vocabulary, you might want to consider character-level embeddings like fasttext?