sentiment-final-model

Model Card for goatley/sentiment-final-model

This model is a fine-tuned DistilBERT model for binary sentiment classification (positive/negative) of English text reviews.
It was developed as part of an advanced NLP dashboard project demonstrating applied skills in deep learning, NLP engineering, and full-stack app deployment.

Model Details

Model Description

• Developed by: Keith Goatley
• License: MIT
• Model type: DistilBERT-based Sequence Classification (Binary)
• Language(s): English
• Fine-tuned from: distilbert-base-uncased
• Base model: Hugging Face Transformers v4
• Framework: PyTorch

Model Sources

• Repository: GitHub Repository
• Demo: Hugging Face Space (when deployed)

Uses

Direct Use

• Classifying short text reviews (e.g., Amazon product reviews) into positive or negative sentiment.

Downstream Use

• Embedding inside sentiment-driven recommendation engines
• As a component of multi-task NLP dashboards
• Fine-tuning for domain-specific sentiment (e.g., medical, finance, hospitality reviews)

Out-of-Scope Use

• Not designed for languages other than English.
• Not suited for emotion detection beyond binary sentiment.

Bias, Risks, and Limitations

This model was fine-tuned on Amazon reviews, which may carry biases toward product-related expressions and cultural language patterns.
Users should be cautious when applying the model outside typical e-commerce datasets.

Recommendations

For more robust domain generalization:

• Further fine-tuning on task-specific datasets is advised.

How to Get Started with the Model

from transformers import pipeline

classifier = pipeline(
    "sentiment-analysis",
    model="goatley/sentiment-final-model",
    tokenizer="goatley/sentiment-final-model"
)

classifier(["I love this!", "This was awful."])


Training Details
Training Data
Subset of Amazon Reviews Dataset

Balanced 50/50 split of positive and negative reviews

Approximately 5,000 examples used for fine-tuning

Training Procedure
Fine-tuned for 3 epochs

Learning rate scheduling with warmup

Optimizer: AdamW

Batch size: 16

Device: CPU-based training (GitHub Codespaces)

Training Hyperparameters
Learning Rate: 5e-5

Optimizer: AdamW

Max Sequence Length: 512

Epochs: 3

Evaluation
Testing Data
Held-out test split from the Amazon Reviews dataset

Metrics

Metric	Score
Test Accuracy	85%
Evaluation was performed using basic classification metrics (accuracy, precision, recall, F1-score).

Environmental Impact
Hardware Type: CPU (GitHub Codespaces)

Hours Used: ~2 hours

Cloud Provider: GitHub (Microsoft Azure backend)

Compute Region: North America

Carbon Emitted: Negligible (very small dataset + CPU-only fine-tuning)

Technical Specifications
Model Architecture and Objective
Architecture: DistilBERT Transformer Encoder

Task Objective: Sequence classification with 2 labels (positive, negative)

Compute Infrastructure
Training performed on GitHub Codespaces virtual machines.

No GPUs were used.

Software Environment
Hugging Face transformers==4.51.3

Datasets datasets==3.5.0

PyTorch torch==2.6.0

Citation
If you use this model or find it helpful, please cite:

APA:

Goatley, K. (2025). Sentiment Analysis Fine-Tuned DistilBERT Model [Model]. Hugging Face. https://huggingface.co/goatley/sentiment-final-model

BibTeX:
@misc{goatley2025sentiment,
  author = {Keith Goatley},
  title = {Sentiment

Analysis Fine-Tuned DistilBERT Model},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/goatley/sentiment-final-model}}
}

Model Card Authors
Keith Goatley

Contact
For questions or inquiries, please contact via:

GitHub: https://github.com/Keithgoatley

Hugging Face: https://huggingface.co/goatley