FEELFLOWAI
Full-stack web application that uses advanced artificial intelligence to analyze sentiment and detect 12 specific types of toxicity in YouTube comments
The Problem
Content creators and brands on YouTube face the challenge of manually moderating thousands of comments, without effective tools to automatically detect toxic content, hate speech, threats, and other types of problematic behavior. This manual moderation consumes significant resources and is prone to human error.
The Solution
Development of an AI system that combines sentiment analysis with multi-class toxicity detection using a custom hybrid BiLSTM model. The system processes YouTube comments in real-time, identifies 12 specific types of toxicity, and provides calibrated risk metrics through an interactive dashboard.
Impact
Identifies problematic content with 95.8% accuracy, facilitating automatic moderation and data-driven community management strategies. Detects potential sarcasm and provides risk metrics per video, optimizing resources through calibrated predictions.
Category
Artificial Intelligence & Full Stack
Completion date
July 2025
Tech Stack
Highlights
- Hybrid BiLSTM model with F1-macro of 95.8%
- Detection of 12 specific types of toxicity
- Interactive dashboard with advanced visualizations
- Automatic integration with YouTube API v3
- Cleaning pipeline with VADER sentiment analysis
- Enterprise architecture with Azure deployment
- Custom model with 2.7M parameters
Desktop View
Tablet View
Mobile View
Key Features
- Real-time simultaneous sentiment and toxicity analysis
- Advanced visualizations (radar charts, scatter plots, heatmaps)
- Sarcasm detector and sentiment-toxicity correlation
- ETL pipeline for massive comment extraction
- Feature engineering of 107 characteristics
- Risk metrics system per video
Next project
ACADEMICPREDICTOR