Better NER BERT Named-Entity-Recognition

Named-Entity-Recognition-with-Bidirectional-LSTM-CNNs

A keras implementation of Bidirectional-LSTM_CNNs for Named-Entity-Recoganition. The original paper can be found at https://arxiv.org/abs/1511.08308

The implementation differs from the original paper in the following ways :

1. lexicons are not considered
2. Bucketing is used to speed up the training
3. nadam optimizer used instead of SGD

Result

The model produces a test F1_score of 90.9 % with ~70 epochs. The results produced in the paper for the given architecture is 91.14 Architecture(BILSTM-CNN with emb + caps)

Dataset

conll-2003

Network Model in paper

Network Model Constructed Using Keras

To run the script

   python3 nn.py

Requirements

0) nltk
1) numpy 
2) Keras==2.1.2
3) Tensorflow==1.4.1

Inference on trained model

from ner import Parser

p = Parser()

p.load_models("models/")

p.predict("Steve Went to Paris")
##Output [('Steve', 'B-PER'), ('went', 'O'), ('to', 'O'), ('Paris', 'B-LOC')]