th-nuernberg/bert-base-german-counseling-gecco

gecco-bert-base-german-uncased

Text classification on German E-Counseling Conversations.

This model is a fine-tuned version of dbmdz/bert-base-german-uncased trained with the German E-Counseling Conversation Dataset, created at the Technische Hochschule Nürnberg (see github.com/th-nuernberg/gecco-dataset).

It achieves the following results on the evaluation set:

• Loss: 1.2341
• Accuracy: 0.6968
• F1: 0.4493

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 16

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
3.4151	1.0	20	3.0885	0.2935	0.0760
2.9316	2.0	40	2.7003	0.3484	0.1035
2.5556	3.0	60	2.3463	0.5032	0.2350
2.19	4.0	80	2.0714	0.5613	0.2841
1.904	5.0	100	1.8381	0.6	0.3085
1.6285	6.0	120	1.6712	0.6323	0.3633
1.4482	7.0	140	1.5518	0.6581	0.3774
1.2807	8.0	160	1.4796	0.6677	0.3880
1.1126	9.0	180	1.4207	0.6613	0.3787
1.0747	10.0	200	1.3461	0.6774	0.3885
0.9068	11.0	220	1.3097	0.6871	0.4132
0.8498	12.0	240	1.2893	0.6903	0.4235
0.8343	13.0	260	1.2549	0.7	0.4332
0.7375	14.0	280	1.2426	0.7	0.4497
0.7274	15.0	300	1.2385	0.7	0.4512
0.6916	16.0	320	1.2341	0.6968	0.4493

Framework versions

• Transformers 4.35.1
• Pytorch 1.10.1+cu111
• Datasets 2.14.7
• Tokenizers 0.14.1