The experiments on NER field let me to build the following model. The model made use of two input, one the PoS embeddings (calculated starting from the SpaCy tool) and the Word embeddings (Word2Vec 300d from Google research) I also tested the model using the Conditional Random Field, as shown in the table result below, I got a more higher score without it when I started using the PoS embeddings.

Check the Report and the Slide Presentation for more information.

test.sh is a simple bash script. To run it:

conda create -n nlp2020-hw1 python=3.7
conda activate nlp2020-hw1
pip install -r requirements.txt
bash test.sh data/dev.tsv

Actually, you can replace data/dev.tsv to point to a different file, as far as the target file has the same format.