📺 YouTube Comment Sentiment & Emotion Analyzer 2.0

A high-performance, full-stack application that goes far beyond simple sentiment analysis. It performs deep psychological profiling, abuse detection, and time-series analysis of YouTube comments using an ensemble of custom AI models, providing actionable, data-driven insights for content creators.

( you run the dashboard locally to see the stunning dark-mode UI!)

---

✨ Key Features (2.0 Update)

🧠 Deep Psychological Profiling

• 28-Category Emotion Model: Upgraded from a basic 7-label model to a comprehensive, fine-tuned GoEmotions-based model (joy, anger, realization, disappointment, curiosity, etc.).
• Multi-Label Thresholding: Detects complex, nuanced emotional states. A single comment can now be correctly identified as both "gratitude" AND "joy".
• Mixed Sentiment Detection: Automatically flags comments expressing conflicting opinions (e.g., "Great video, but the audio is terrible").

🛡️ Automated Moderation & Toxicity Filtering

• Multilingual Toxicity Detection: Integrated citizenlab/distilbert-base-multilingual-cased-toxicity to detect abuse, hate speech, and spam.
• Strict Slang Tolerance: Precision-tuned thresholding (>0.985) ensures enthusiastic internet slang ("this is a banger!") isn't falsely flagged as toxic.
• Protective UI: Toxic comments are isolated into a priority queue and visually blurred by default to protect moderators.

📈 Advanced Analytics & Visualizations

• Time-Series Analysis: A dynamic Timeline graph tracks viewer sentiment over time, allowing creators to correlate audience mood swings with specific events or publishing times.
• Interactive Dashboard: A stunning, responsive React 19 dashboard featuring 28-color emotion mapping, sentiment pie charts, and real-time filtering.
• Data-Driven AI Suggestions: Automatically generates context-aware advice (e.g., "High negative sentiment detected regarding audio quality" or "Viewers are asking for a part 2").

---

🛠️ Tech Stack

Backend (Inference & API):

• Framework: FastAPI (Python 3.11+)
• Machine Learning: PyTorch, Hugging Face Transformers (XLM-RoBERTa, DistilBERT)
• MLOps: MLflow for experiment tracking and historical run logging.
• Data Processing: Pandas, langdetect

Frontend (Dashboard):

• Framework: React 19.2 (Vite)
• Styling: Vanilla CSS (Sleek Dark Mode, Glassmorphism, Micro-animations)
• Data Visualization: Recharts, Lucide Icons

---

📦 Setup & Installation

1. Environment Setup

# Clone the repository
git clone <your-repo-url>
cd sentiment-analyzer

# Create and activate environment
python -m venv venv
venv\Scripts\activate  # Windows
source venv/bin/activate  # Linux/macOS

# Install dependencies
pip install -r requirements.txt

2. Configure Credentials

Create a .env file in the root directory:

YOUTUBE_API_KEY=your_youtube_api_key_here

(You can get a free API key from the Google Cloud Console by enabling the YouTube Data API v3).

3. Run the Application

Terminal 1 (Backend API):

uvicorn api:app --reload

Terminal 2 (Frontend Dashboard):

cd dashboard
npm install
npm run dev

Visit http://localhost:5173 in your browser to access the dashboard!

---

🧠 Under the Hood: The AI Pipeline

The application passes scraped comments through a robust multi-stage NLP pipeline:

1. Language Detection: Identifies the language of the comment (handles mixed-language input).
2. Base Sentiment: A multilingual XLM-RoBERTa model calculates the foundational polarity (Positive, Neutral, Negative).
3. Emotion Extraction: Our custom fine-tuned DistilRoBERTa applies a sigmoid threshold (>0.3) across 28 distinct emotional axes.
4. Toxicity Screening: A specialized zero-shot classification model scans for abuse, shielding the UI with safety blurring.
5. Contextual Aggregation: The backend aggregates the dataset to discover topics, pin suggestions, and compute like-weighted sentiment graphs.

---

📂 Project Structure

sentiment-analyzer/
├── api.py                 # FastAPI routing, caching, and payload generation
├── models.py              # ML pipeline initialization and inference logic
├── analysis.py            # Pandas aggregations, timeline tracking, mixed sentiment
├── fetcher.py             # YouTube API integration and comment scraping
├── suggestions.py         # Heuristic engine for generating creator advice
├── dashboard/             # React frontend (Vite)
│   ├── src/
│   │   ├── components/    # Reusable UI elements (SentimentCards, CommentTable, etc.)
│   │   ├── index.css      # Core design system and CSS tokens
│   │   └── App.jsx        # Main dashboard view
├── fine-tuned-emotion-model/ # Exported weights for the 28-class GoEmotions model
└── requirements.txt       # Python dependencies

---

📈 MLflow Experiment Tracking

Every analysis run is automatically logged locally via MLflow to mlflow.db.

To view your historical performance data, run:

mlflow ui

Then open http://localhost:5000 to visualize how your content reception has evolved across multiple videos.

---

🤝 Contributing

Contributions are always welcome! Whether it's adding a word-cloud visualization, refining the topic modeling heuristic, or expanding the AI suggestions logic, feel free to open a Pull Request.

---

Originally created by Teshan G.H.P | University of Ruhuna