NLP Sentiment Analysis & Opinion Mining

Live Demo

Try the interactive dashboard: NLP Sentiment Analysis App

---

A production-ready sentiment analysis platform using real Amazon product reviews with multiple ML models, business insights, and interactive visualizations.

Key Features

Analysis Capabilities

• Multiple Sentiment Models: VADER, TextBlob, Logistic Regression (89.6% accuracy), Naive Bayes
• Deep Learning Models (NEW!): CNN and BiLSTM with TensorFlow/PyTorch, pre-trained embeddings (GloVe, Word2Vec)
• Business Insights: Automated alerts, top issues detection, actionable recommendations
• Comparison Mode: Side-by-side brand/category analysis with radar charts
• Opinion Mining: Aspect extraction, sentiment drivers, category analysis
• Temporal Analysis: Sentiment trends over time

Dashboard Features

• 8 Interactive Tabs: Overview, Business Insights, Compare, Categories, Aspects, Trends, Model Performance, Deep Dive
• Polished Custom Theme: Consistent, semantic light theme for all charts and components
• Export Functionality: Download filtered data as CSV or Excel
• Real-time Filtering: Category, brand, sentiment, rating, and date filters

Production Ready

• Docker Support: Multi-container deployment with docker-compose
• REST API: FastAPI endpoints for real-time predictions
• Structured Logging: JSON logging for monitoring and debugging
• Modular Architecture: Clean component-based code structure

Architecture

flowchart TB
    subgraph DataLayer[Data Layer]
        HF[HuggingFace Dataset]
        DL[Data Loader]
        PP[Preprocessor]
    end
    
    subgraph MLLayer[ML Layer]
        SA[Sentiment Analyzer]
        ML[ML Models]
        OM[Opinion Miner]
    end
    
    subgraph AppLayer[Application Layer]
        ST[Streamlit Dashboard]
        API[FastAPI REST API]
    end
    
    subgraph Components[Dashboard Components]
        OV[Overview]
        BI[Business Insights]
        CM[Compare Mode]
        EX[Export]
    end
    
    HF --> DL
    DL --> PP
    PP --> SA
    PP --> ML
    SA --> OM
    ML --> ST
    ML --> API
    OM --> ST
    ST --> Components

Loading

Project Structure

nlp-sentiment-analysis/
├── app.py                     # Streamlit dashboard (slim entry point)
├── main.py                    # CLI pipeline script
├── Dockerfile                 # Docker image for dashboard
├── docker-compose.yml         # Multi-service deployment
├── components/                # Modular UI components
│   ├── header.py              # Page header
│   ├── sidebar.py             # Filters and controls
│   ├── kpi_cards.py           # Metric cards
│   ├── charts/                # Chart components
│   │   ├── sentiment.py       # Sentiment charts
│   │   ├── category.py        # Category analysis
│   │   ├── temporal.py        # Time series
│   │   └── comparison.py      # Comparison charts
│   └── tabs/                  # Tab components
│       ├── overview.py        # Overview tab
│       ├── insights.py        # Business Insights tab
│       ├── compare.py         # Comparison Mode tab
│       ├── categories.py      # Categories tab
│       ├── aspects.py         # Aspects tab
│       ├── trends.py          # Trends tab
│       ├── performance.py     # Model Performance tab
│       └── deep_dive.py       # Deep Dive tab
├── utils/                     # Utility modules
│   ├── theme.py               # Theme and styling
│   ├── cache.py               # Data caching
│   ├── export.py              # Export functionality
│   ├── loading.py             # Loading states
│   └── logger.py              # Structured logging
├── src/                       # Core ML modules
│   ├── data_loader.py         # Data fetching
│   ├── preprocessor.py        # Text preprocessing
│   ├── sentiment_analyzer.py  # VADER + TextBlob
│   ├── ml_models.py           # ML training
│   ├── model_evaluator.py     # Evaluation
│   └── opinion_miner.py       # Aspect extraction
├── api/                       # REST API
│   ├── main.py                # FastAPI app
│   ├── schemas.py             # Request/response models
│   └── predictor.py           # Prediction service
└── tests/                     # Unit tests

Quick Start

Option 1: Docker (Recommended)

# Clone the repository
git clone https://github.com/AvishManiar21/nlp-sentiment-analysis.git
cd nlp-sentiment-analysis

# Start all services
docker-compose up -d

# Access the dashboard at http://localhost:8501
# Access the API at http://localhost:8000

Option 2: Local Development

# Clone and setup
git clone https://github.com/AvishManiar21/nlp-sentiment-analysis.git
cd nlp-sentiment-analysis
python -m venv venv
source venv/bin/activate  # Linux/Mac
# venv\Scripts\activate   # Windows

# Install dependencies
pip install -r requirements.txt

# Run the pipeline (downloads data, trains models)
python main.py

# Launch the dashboard
streamlit run app.py

# Or run the API
uvicorn api.main:app --reload --port 8000

Dashboard Tabs

Tab	Description
Overview	KPI metrics, sentiment distribution, confusion matrix
Business Insights	Automated alerts, top issues, recommendations
Compare	Side-by-side brand/category comparison with radar charts
Categories & Brands	Category sentiment analysis, brand positioning
Aspects & Drivers	Aspect-level opinion mining, word clouds
Trends	Temporal sentiment trends, VADER vs TextBlob comparison
Model Performance	Accuracy comparison, F1 scores, best models
Deep Dive	Sample reviews, search functionality

Results

Classical ML Models

Model	Accuracy	F1 Score	Precision	Recall
Logistic Regression	89.6%	0.896	0.896	0.896
Naive Bayes	87.1%	0.871	0.871	0.871
VADER	70.5%	0.699	0.770	0.705
Ensemble	70.3%	0.698	0.774	0.703
TextBlob	61.5%	0.622	0.794	0.615

Deep Learning Models (NEW!)

We now support state-of-the-art deep learning models with both TensorFlow and PyTorch:

Model	Framework	Embeddings	Expected Accuracy	Description
CNN	TensorFlow	Learned	~91-92%	1D CNN with multiple filter sizes (3,4,5-grams)
CNN + GloVe	TensorFlow	Pre-trained	~92-94%	CNN with frozen GloVe embeddings
CNN	PyTorch	Learned	~91-92%	Parallel PyTorch implementation
CNN + GloVe	PyTorch	Pre-trained	~92-94%	PyTorch CNN with GloVe embeddings
BiLSTM	PyTorch	Learned/Pre-trained	~90-93%	Bidirectional LSTM for sequence modeling

Training Deep Learning Models

# Train CNN models with both TensorFlow and PyTorch
python main.py --train-dl --dl-framework both --dl-model-type cnn

# Train with pre-trained GloVe embeddings for better accuracy
python main.py --train-dl --use-embeddings --embedding-name glove-wiki-gigaword-100

# Train LSTM model (PyTorch only)
python main.py --train-dl --dl-framework pytorch --dl-model-type lstm

# Customize training parameters
python main.py --train-dl --dl-epochs 20 --dl-batch-size 64

# Train all model types
python main.py --train-dl --dl-framework both --dl-model-type both --use-embeddings

Available Pre-trained Embeddings

• glove-wiki-gigaword-100 (100d) - Fast, good accuracy
• glove-wiki-gigaword-200 (200d) - Better accuracy
• glove-wiki-gigaword-300 (300d) - Best accuracy, slower
• word2vec-google-news-300 (300d) - Google News corpus
• glove-twitter-100 (100d) - Optimized for social media
• fasttext-wiki-news-subwords-300 (300d) - Handles rare words well

Deep Learning Features

• Hybrid Architectures: Combine pre-trained embeddings with CNNs for state-of-the-art results
• Multi-Framework Support: Compare TensorFlow and PyTorch implementations
• GPU Acceleration: Automatic GPU detection (CUDA, MPS, or CPU fallback)
• TensorBoard Integration: Visualize training metrics and model architecture
• Early Stopping: Prevent overfitting with automatic early stopping
• Model Checkpointing: Save best models during training
• Dashboard Integration: View trained models in the Model Performance tab

📖 Read the complete Deep Learning Guide for detailed training instructions, benchmarks, and best practices.

REST API

Endpoints

Method	Endpoint	Description
GET	/	API info
GET	/health	Health status
GET	/models	Available models
POST	/predict	Single prediction
POST	/predict/batch	Batch predictions

Example

curl -X POST "http://localhost:8000/predict" \
  -H "Content-Type: application/json" \
  -d '{"text": "This product is amazing!", "model": "logistic_regression"}'

{
  "text": "This product is amazing!",
  "model": "logistic_regression",
  "sentiment": "positive",
  "confidence": 0.92,
  "scores": {"positive": 0.92, "negative": 0.08}
}

Docker Deployment

# Build and run dashboard only
docker build -t nlp-sentiment .
docker run -p 8501:8501 nlp-sentiment

# Run with docker-compose (dashboard + API)
docker-compose up -d

# View logs
docker-compose logs -f

# Stop services
docker-compose down

Environment Configuration

Copy .env.example to .env and configure:

cp .env.example .env

Key settings:

• CLOUD_SAMPLE_SIZE: Number of reviews to process when generating the dataset (default: 30000)
• CLOUD_MODE: Set to true on Streamlit Cloud to enable lighter, Cloud-optimized defaults
• CLOUD_DISPLAY_SAMPLE_SIZE: Maximum number of reviews loaded into the dashboard when CLOUD_MODE=true (default: 20000)
• LOG_LEVEL: Logging level (DEBUG, INFO, WARNING, ERROR)
• API_AUTH_ENABLED: Enable API authentication

Technologies

• Python 3.10+ - Core language
• Streamlit - Interactive dashboard
• FastAPI - REST API
• scikit-learn - Classical ML models
• TensorFlow/Keras - Deep learning (CNN models)
• PyTorch - Deep learning (CNN, BiLSTM models)
• Gensim - Pre-trained word embeddings (Word2Vec, GloVe)
• NLTK/TextBlob - NLP processing
• Plotly - Visualizations
• TensorBoard - Training visualization
• Docker - Containerization
• HuggingFace - Dataset loading & transformers

License

MIT License - see LICENSE for details.

---

Built as a production-ready demonstration of NLP sentiment analysis, from data processing to deployment.