distilbert-base-uncased-Finanacial_Sentiment_Analysis
This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3079
- Accuracy: 0.8529
- F1 Score: 0.8564
Model description
This project classifies input samples as one of the following: negative, neutral, or positive.
For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Sentiment%20Analysis/Financial%20Sentiment%20Analysis/Financial%20Sentiment%20Analysis-Updated%20Version.ipynb
Intended uses & limitations
More information needed
Training and evaluation data
There were two datasets that I concatenated:
- https://www.kaggle.com/datasets/sbhatti/financial-sentiment-analysis
- https://www.kaggle.com/datasets/ankurzing/sentiment-analysis-for-financial-news
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: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 Score |
|---|---|---|---|---|---|
| 0.5569 | 1.0 | 134 | 0.3954 | 0.7591 | 0.7559 |
| 0.3177 | 2.0 | 268 | 0.3391 | 0.8135 | 0.8151 |
| 0.2479 | 3.0 | 402 | 0.3211 | 0.8322 | 0.8353 |
| 0.2049 | 4.0 | 536 | 0.3066 | 0.8463 | 0.8506 |
| 0.1802 | 5.0 | 670 | 0.3079 | 0.8529 | 0.8564 |
Framework versions
- Transformers 4.22.1
- Pytorch 1.12.1
- Datasets 2.4.0
- Tokenizers 0.12.1
Similar Models
You can find two models similar to this one that I completed at these links:
- https://huggingface.co/DunnBC22/fnet-large-Financial_Sentiment_Analysis_v3
- https://huggingface.co/DunnBC22/fnet-base-Financial_Sentiment_Analysis
