Named-Enitty-Recognition

This repository explores different techiques used for Named Entity Recognition (NER) on the CONLL 2003 Reuters Dataset. The data can be found here: https://www.clips.uantwerpen.be/conll2003/ner/ and is under licence by Reuters.

Models

Feature Engineered

• Conditional Random Fields [w] (F1=77.73)
• Conditional Random Fields [w-1] (F1=83.54)
• Conditional Random Fields [w-2] (F1=83.25)
• Conditional Random Fields [w-3] (F1=83.91)

BiLSTM-CRF + Word Embeddings

Glove

• BiLSTM-CRF (Glove 50D) (F1=82.55)

SENNA

• BiLSTM-CRF (SENNA 50D) (F1=82.79)

Other Models

• BiLSTM-CRF (Glove + Casing Features) (F1=86.23)
• BiLSTM-CRF (SENNA + Casing Features) (F1=85.35)

Word Embeddings:

• GloVe
• SENNA

Requirements

• Python 3.6
• NumPy
• Pandas
• scikit-learn
• Tensorflow
• Keras