EconoSentiment

This model is a fine-tuned version of samchain/econo-sentence-v2 on the Financial Phrase Bank dataset from FinanceMTEB. The full model is trained using a small learning rate isntead of freezing the encoder. Hence, you should not use the encoder of this model for a task other than sentiment analysis.

It achieves the following results on the evaluation set:

• Loss: 0.1293
• Accuracy: 0.962
• F1: 0.9619
• Precision: 0.9619
• Recall: 0.962

Model description

The base model is a sentence-transformers model built from EconoBert.

Intended uses & limitations

This model is trained to provide a useful tool for sentiment analysis in finance.

Training and evaluation data

The dataset is directly downloaded from the huggingface repo of the FinanceMTEB. The preprocessing consisted of tokenizing to a fixed sequence length of 512 tokens.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.5992	1.0	158	0.4854	0.805	0.7692	0.8108	0.805
0.0985	2.0	316	0.1293	0.962	0.9619	0.9619	0.962

Framework versions

• Transformers 4.50.0
• Pytorch 2.1.0+cu118
• Datasets 3.4.1
• Tokenizers 0.21.1