POS Tagger using Neural Net

A simple pos tagger using neural nets. The corpora used was brown corpus. The library for neural nets is keras.

Architecture

I used a simple feed forward neural network with a single hidden layer of 100 neurons, and a multiclass output for 472 tags.

Stats of the first few tags from report.txt on brown corpus

			 precision    recall  f1-score   support

	  b'AT'       0.92      0.93      0.93     17192
   b'NP-TL'       0.66      0.45      0.53       334
   b'NN-TL'       0.77      0.76      0.76      1056
   b'JJ-TL'       0.81      0.62      0.70       276
	 b'VBD'       0.89      0.97      0.93     10607
	  b'NR'       0.95      0.78      0.86       329
	  b'NN'       0.92      0.93      0.92     25750
	  b'IN'       0.80      0.89      0.85     19273
	 b'NP$'       0.43      0.24      0.31       597
	  b'JJ'       0.82      0.88      0.85      9751
	  b'``'       0.78      0.86      0.82      3353
	  b"''"       0.91      0.63      0.74      3344
	  b'CS'       0.92      0.84      0.88      4201
	 b'DTI'       1.00      0.83      0.91       482
	 b'NNS'       0.97      0.93      0.95      7076