pascalrai/large-BERT-NER-email

Overview:

The Model is fine-tuned for 3 class + "0" class.
The Dataset is custom annotated and contains 400 texts and the model was trained on the split of 0.76, 0.12, and 0.12.

The validation classification report is as follows:

Class	Precision	Recall	f1
0	1.00	1.00	1.00
1	0.98	1.00	0.91
2	0.95	0.89	0.92
3	0.8	0.88	0.84
macro-avg	0.93	0.94	0.94

The test classification report is as follows:

Class	Precision	Recall	f1
0	1.00	1.00	1.00
1	0.98	1.00	0.99
2	0.66	0.97	0.79
3	0.84	0.78	0.81
macro-avg	0.87	0.94	0.90

Possible future direction:

1. Clean data to a good enough format as much as possible.
2. Increase the data as much as possible. (Make sure to have data that is seen in real use cases.)
3. Ponder: Is it possible to use sth like Grammarly to clean the sentences before tokenization such that proper nouns are Capital and the grammer is correct such that a pattern is formed?