bert-for-patents-finetuned_ls-sys

This model is a fine-tuned version of anferico/bert-for-patents on a unique dataset consisting of 45392 patent applications, of which 10392 were defined as "LS-SYS-related" and 35000 as "Not LS-SYS-related". The fine-tuning is performed on patents' titles and abstracts. The base model was fine-tuned to perform a binary classification task, identifying patents related to the "Learning and Symbolic Systems" domain.

It achieves the following results on the evaluation set:

• Loss: 0.0529
• Accuracy: 0.981
• Auc: 0.998
• F1: 0.958
• Precision: 0.962
• Recall: 0.955

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Auc	F1	Precision	Recall
0.3254	1.0	993	0.1486	0.947	0.987	0.88	0.922	0.841
0.1237	2.0	1986	0.0887	0.969	0.995	0.931	0.945	0.918
0.0928	3.0	2979	0.0718	0.973	0.996	0.94	0.962	0.919
0.0811	4.0	3972	0.0635	0.977	0.997	0.949	0.96	0.939
0.0748	5.0	4965	0.0596	0.979	0.997	0.953	0.955	0.952
0.0695	6.0	5958	0.0563	0.98	0.997	0.955	0.962	0.949
0.0682	7.0	6951	0.0552	0.98	0.997	0.957	0.958	0.956
0.0664	8.0	7944	0.0537	0.981	0.997	0.958	0.961	0.954
0.0642	9.0	8937	0.0530	0.981	0.998	0.958	0.962	0.954
0.0647	10.0	9930	0.0529	0.981	0.998	0.958	0.962	0.955

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1