bert-base-cased-finetuned-ner-DFKI-SLT_few-NERd

This model is a fine-tuned version of bert-base-cased.

It achieves the following results on the evaluation set:

Loss: 0.1312
Person
- Precision: 0.8860048426150121
- Recall: 0.9401849948612538
- F1: 0.912291199202194
- Number: 29190
Location
- Precision: 0.8686381704207632
- Recall: 0.8152889539136796
- F1: 0.841118472477534
- Number: 95690
Organization
- Precision: 0.7919078915181266
- Recall': 0.7449641777764141
- F1: 0.7677190874452579
- Number': 65183
Product
- Precision: 0.7065968977761166
- Recall: 0.8295304958315051
- F1: 0.7631446160056513
- Number: 9116
Art
- Precision: 0.8407258064516129
- Recall: 0.8614333386302241
- F1: 0.8509536143159878
- Number: 6293
Other
- Precision: 0.7303024586555996
- Recall: 0.8314124132006586
- F1: 0.7775843599357258
- Nnumber: 13969
Building
- Precision: 0.5162234691388143
- Recall: 0.3648904983617865
- F1: 0.4275611234592847
- Number: 5799
Event
- Precision: 0.605920892987139
- Recall: 0.35144264602392683
- F1: 0.44486014608943525
- Number: 7105
Overall
- Precision: 0.8203
- Recall: 0.7886
- F1: 0.8041
- Accuracy: 0.9498

Model description

For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/tree/main/Token%20Classification/Monolingual/DFKI%20SLT%20few%20NERd

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology.

Training and evaluation data

Dataset Source: https://huggingface.co/datasets/DFKI-SLT/few-nerd

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Person Precision	Person Recall	Person F1	Person Number	Location Precision	Location Recall	Location F1	Location Number	Organization Precision	Organization Recall	Organization F1	Organization Number	Product Precision	Product Recall	Product F1	Product Number	Art Precision	Art Recall	Art F1	Art Number	Other Precision	Other Recall	Other F1	Other Number	Building Precision	Building Recall	Building F1	Building Number	Event Precision	Event Recall	Event F1	Event Number	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.1796	1.0	11293	0.1427	0.8741	0.9272	0.8999	29190	0.8576	0.8072	0.8316	95690	0.7699	0.7688	0.7694	65183	0.6711	0.75	0.7084	9116	0.8347	0.8154	0.8249	6293	0.6743	0.8195	0.7398	13969	0.4812	0.3951	0.4339	5799	0.5998	0.3253	0.4218	7105	0.8000	0.7852	0.7925	0.9483
0.1542	2.0	22586	0.1312	0.8860	0.9402	0.9123	29190	0.8686	0.8153	0.8411	95690	0.7919	0.7450	0.7677	65183	0.7066	0.8295	0.7631	9116	0.8407	0.8614	0.8510	6293	0.7303	0.8314	0.7776	13969	0.5162	0.3649	0.4276	5799	0.6059	0.3514	0.4449	7105	0.8203	0.7886	0.8041	0.9498

Framework versions

Transformers 4.30.2
Pytorch 2.0.1+cu118
Datasets 2.13.1
Tokenizers 0.13.3