title	Sentiment Analysis Bert
emoji	🧠
colorFrom	blue
colorTo	purple
sdk	gradio
sdk_version	4.0.0
python_version	3.10
app_file	app.py
pinned	false

🧠 Full Emotion & Sentiment Analyzer using BERT

Author: Aarica Raj GitHub: @Aaricacoding Tech Stack: Python · PyTorch · HuggingFace Transformers · DistilRoBERTa · DistilBERT · Gradio

---

📸 Demo Screenshot

📌 What is this project?

This project builds an AI-powered Full Emotion & Sentiment Analysis System that detects:

• 7 Emotions : Joy, Sadness, Anger, Fear, Surprise, Disgust, Neutral
• Overall Sentiment : Positive or Negative

It uses two models together:

• DistilRoBERTa : fine-tuned on emotion dataset for 7-emotion detection
• DistilBERT : fine-tuned on SST-2 for Positive/Negative sentiment

Real-world use cases:

• 📦 Product review analysis
• 🐦 Social media emotion monitoring
• 🎬 Movie/book review classification
• 💬 Customer feedback analysis
• 🏥 Mental health text monitoring

---

🧠 How Does It Work? (Architecture)

Input Text
    │
    ├──────────────────────┬─────────────────────────┐
    ▼                      ▼                         │
┌──────────────┐   ┌───────────────────┐             │
│  Tokenizer   │   │   Tokenizer       │             │
│  (RoBERTa)   │   │   (BERT)          │             │
└──────┬───────┘   └────────┬──────────┘             │
       │                    │                         │
       ▼                    ▼                         │
┌──────────────┐   ┌───────────────────┐             │
│ DistilRoBERTa│   │ DistilBERT        │             │
│ Emotion Model│   │ Sentiment Model   │             │
└──────┬───────┘   └────────┬──────────┘             │
       │                    │                         │
       ▼                    ▼                         │
┌──────────────┐   ┌───────────────────┐             │
│ 7 Emotions   │   │ Positive/Negative │             │
│ with scores  │   │ with scores       │             │
└──────┬───────┘   └────────┬──────────┘             │
       └────────────────────┘                         │
                    │                                 │
                    ▼                                 │
           Full Breakdown Output ────────────────────┘

Key Concepts:

Concept	Explanation
BERT	Reads text in both directions for better understanding
DistilBERT	40% smaller, 60% faster than BERT, keeps 97% accuracy
RoBERTa	Improved BERT with better training - more accurate
DistilRoBERTa	Lighter version of RoBERTa for faster inference
Fine-tuning	Pre-trained model trained further on specific data
SST-2	Stanford Sentiment Treebank - 67K labeled movie reviews
Softmax	Converts raw scores to probabilities that sum to 100%

---

📁 Project Structure

sentiment-analysis-bert/
│
├── app.py               ← Main Gradio application (run this!)
├── requirements.txt     ← All Python dependencies
└── README.md            ← This documentation file

---

⚙️ Setup & Installation

Step 1 : Clone the Repository

git clone https://github.com/Aaricacoding/sentiment-analysis-bert.git
cd sentiment-analysis-bert

Step 2 : Create Virtual Environment

python -m venv venv

# Windows:
venv\Scripts\activate

# Mac/Linux:
source venv/bin/activate

Step 3 : Install Dependencies

pip install -r requirements.txt

Step 4 : Run the App

python app.py

Step 5 : Open in Browser

https://aaricacoding-sentiment-analysis-bert.hf.space

💡 First run downloads both models (~550MB total). Only happens once!

---

🚀 How to Use

1. Open the app at https://aaricacoding-sentiment-analysis-bert.hf.space
2. Type any text in the input box
3. Click "🔍 Analyze" or press Enter
4. See the full breakdown:
   • ✅ Primary emotion detected
   • ✅ All 7 emotion scores with percentages
   • ✅ Positive/Negative sentiment scores

---

🧪 Example Outputs

Text	Primary Emotion	Sentiment
"I love this so much!"	😊 Joy (98%)	👍 Positive (99%)
"I am so angry right now!"	😠 Anger (94%)	👎 Negative (97%)
"The movie was okay, nothing special."	😐 Neutral (68%)	👍 Positive (54%)
"I am scared about tomorrow."	😨 Fear (91%)	👎 Negative (89%)
"Oh wow! What a surprise!"	😲 Surprise (87%)	👍 Positive (76%)
"This smells absolutely disgusting."	🤢 Disgust (93%)	👎 Negative (98%)
"I went to the store today."	😐 Neutral (82%)	👍 Positive (61%)

---

🔧 Technologies Used

Tool	Purpose
Python 3.10+	Core programming language
PyTorch	Deep learning framework
HuggingFace Transformers	Pre-trained models library
DistilRoBERTa	7-emotion detection model
DistilBERT (SST-2)	Positive/Negative sentiment model
Gradio	Web UI for ML demos

---

🌐 Models Used

Model	Task	Parameters	Accuracy
j-hartmann/emotion-english-distilroberta-base	7 Emotions	~82M	66% (6-class)
distilbert-base-uncased-finetuned-sst-2-english	Pos/Neg	~66M	91.3%

---

💡 What I Learned Building This

• How BERT and RoBERTa read text bidirectionally
• Difference between DistilBERT vs DistilRoBERTa
• How to run multiple AI models together in one app
• How fine-tuning adapts pre-trained models to specific tasks
• How tokenization works in NLP (WordPiece tokenizer)
• How to build and deploy NLP apps using HuggingFace + Gradio

---

🔮 Future Improvements

• Fine-tune model on custom emotions (jealousy, excitement, love)
• Add multilingual support (Hindi, Spanish, French)
• Add batch processing for multiple texts at once
• Deploy permanently on HuggingFace Spaces (free)
• Add REST API using FastAPI + Docker
• Add real-time Twitter/social media analysis

---

📄 License

MIT License — free to use, modify, and share!

---

🙏 Acknowledgements

• HuggingFace — for Transformers library and model hub
• j-hartmann — for the emotion detection model
• Stanford NLP — for SST-2 dataset
• Gradio Team — for the UI framework