De-identification-of-medical-information

A conversation between a doctor and a patient contains private information. However, the amount of data is huge. Therefore, in this project, use "Named Entity Recognition" to extract them.

Environment

1. Torch 1.8.0
2. Numpy 1.19.2
3. Pandas 0.24.2
4. Pickle 0.7.5

Experiments

Model number	Architecture	Embedding dimension	Hidden layer dimension	F1 score
1	CRF	512	X	0.434
2	BiLSTM + CRF	512	512	0.58
3	BiLSTM + CRF	128	128	0.67
4	BiLSTM + CRF	256	256	0.71

Acknowledgment

1. named entity recognition