DHMSW: S-BERT

Use

This model corresponds to S_BERT in our paper Targeting Transparency: Early Evidence on Mandatory Adoption of European Sustainability Reporting Standards. We use the model to classify sentences from firms' sustainability reports.

This model is a fine-tuned version of distilbert/distilroberta-base on the None dataset. It achieves the following results on the evaluation set:

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 15
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.6011	1.0	550	0.2800	0.9095	0.8407	0.9094	0.7991
0.2239	2.0	1100	0.2299	0.9315	0.8894	0.8934	0.8923
0.1398	3.0	1650	0.2583	0.9284	0.8833	0.8712	0.8991
0.0904	4.0	2200	0.2564	0.9438	0.9098	0.8997	0.9224
0.0539	5.0	2750	0.2723	0.9489	0.9164	0.9267	0.9076
0.0368	6.0	3300	0.2961	0.9530	0.9177	0.9269	0.9100
0.0229	7.0	3850	0.3261	0.9509	0.9146	0.9361	0.8966
0.0188	8.0	4400	0.3585	0.9443	0.9079	0.9230	0.8955
0.0112	9.0	4950	0.3865	0.9484	0.9130	0.8973	0.9319