DistilBERT Fine-Tuned for Financial Sentiment Analysis

Model Description

This model is a fine-tuned version of distilbert-base-uncased specifically tailored for sentiment analysis in the financial domain. It has been trained on the Financial PhraseBank dataset to classify financial texts into three sentiment categories:

  • Negative (label 0)
  • Neutral (label 1)
  • Positive (label 2)

Model Performance

The model was trained for 5 epochs and evaluated on a held-out test set constituting 20 of the dataset.

Evaluation Metrics

Epoch Eval Loss Eval Accuracy
1 0.2210 94.26%
2 0.1997 95.81%
3 0.1719 96.69%
4 0.2073 96.03%
5 0.1941 96.69%

Training Metrics

  • Final Training Loss: 0.0797
  • Total Training Time: Approximately 3869 seconds (~1.07 hours)
  • Training Samples per Second: 2.34
  • Training Steps per Second: 0.147

Training Procedure

Data

  • Dataset: Financial PhraseBank
  • Configuration: sentences_allagree (sentences where all annotators agreed on the sentiment)
  • Dataset Size: 2264 sentences
  • Data Split: 80% training (1811 samples), 20% testing (453 samples)

Model Configuration

  • Base Model: distilbert-base-uncased
  • Number of Labels: 3 (negative, neutral, positive)
  • Tokenizer: Same as the base model's tokenizer

Hyperparameters

  • Number of Epochs: 5
  • Batch Size: 16 (training), 64 (evaluation)
  • Learning Rate: 5e-5
  • Optimizer: AdamW
  • Evaluation Metric: Accuracy
  • Seed: 42 (for reproducibility)

Usage

You can load and use the model with the Hugging Face transformers library as follows:

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("AnkitAI/distilbert-base-uncased-financial-news-sentiment-analysis")
model = AutoModelForSequenceClassification.from_pretrained("AnkitAI/distilbert-base-uncased-financial-news-sentiment-analysis")

text = "The company's revenue declined significantly due to market competition."
inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)

logits = outputs.logits
predicted_class_id = logits.argmax().item()

label_mapping = {0: "Negative", 1: "Neutral", 2: "Positive"}
predicted_label = label_mapping[predicted_class_id]

print(f"Text: {text}")
print(f"Predicted Sentiment: {predicted_label}")

License

This model is licensed under the Apache 2.0 License. You are free to use, modify, and distribute this model in your applications.

Citation

If you use this model in your research or applications, please cite it as:

@misc{AnkitAI_2024_financial_sentiment_model,
  title={DistilBERT Fine-Tuned for Financial Sentiment Analysis},
  author={Ankit Aglawe},
  year={2024},
  howpublished={\url{https://huggingface.co./AnkitAI/distilbert-base-uncased-financial-news-sentiment-analysis}},
}

Acknowledgments

  • Hugging Face: For providing the Transformers library and model hosting.
  • Data Providers: Thanks to the creators of the Financial PhraseBank dataset.
  • Community: Appreciation to the open-source community for continual support and contributions.

Contact Information

For questions, feedback, or collaboration opportunities, please contact:

Downloads last month
298
Safetensors
Model size
67M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for AnkitAI/distilbert-base-uncased-financial-news-sentiment-analysis

Dataset used to train AnkitAI/distilbert-base-uncased-financial-news-sentiment-analysis

Evaluation results