🔍 ReviewVision: Multi-Modal E-Commerce Sentiment Analysis

A production-ready sentiment analysis system combining Amazon reviews, YouTube comments, and AWS image quality analysis with Neo4j graph database.

---

🎯 Key Findings

• Amazon reviews are +27pp more positive than YouTube across all categories
• Fashion shows highest platform divergence (+39pp gap)
• Image quality shows weak but significant correlation (r=0.214, p<0.001)
• Analyzed 42K+ reviews, 1,703 images, 36 products

---

📊 Project Architecture

Data Collection → Balancing → VADER NLP → AWS Rekognition → Neo4j Graph
   (Kaggle +        (21K per    (42K texts    (1,703 images   (1,245 nodes
   YouTube API)     platform)    analyzed)      quality scored) 3,936 edges)

---

🛠️ Tech Stack

• Cloud: AWS S3, Rekognition
• NLP: VADER Sentiment Analysis
• Graph DB: Neo4j
• Processing: Python, Pandas, NumPy, SciPy, Boto3
• Visualization: Matplotlib, Seaborn
• APIs: YouTube Data API v3
• Web Scraping: Selenium (Amazon product images)

---

🔬 Research Questions

Q1: Platform Comparison

Do Amazon reviews match YouTube sentiment?

• ❌ No - Amazon consistently +27pp more positive
• Fashion shows biggest gap (+39pp)
• Electronics smallest gap (+1.6pp)

Q2: VADER Consistency

Is sentiment analysis consistent across platforms?

• ✅ Yes - VADER reliable (Amazon 80.4% pos, YouTube 55.4% pos)
• Significant platform bias detected

Q3: Image Quality Impact

Does image quality influence sentiment?

• ⚠️ Weak positive correlation (r=0.214, p<0.001)
• Fashion has highest quality (21.7) AND positive rate (90.9%)

---

📈 Visualizations

Platform Comparison

VADER Consistency

Image Quality Analysis

Neo4j Graph Database

---

💼 Business Applications

1. 📸 Invest in product photography - Higher quality → better sentiment
2. 🎯 Platform-specific strategy - Amazon buyers more positive
3. 👗 Fashion YouTube strategy - Biggest gap (+39pp)
4. 📱 Electronics: focus on specs - Smallest gap (+1.6pp)

---

📊 Dataset

• Amazon Reviews: 21,026 (Kaggle - balanced)
• YouTube Comments: 21,026 (API scraped)
• Product Images: 1,703 (AWS Rekognition)
• Categories: Beauty, Electronics, Fashion, Food
• Products: 36 tracked across platforms

---

🏆 Technical Achievements

✅ Multi-modal data (text + image + video)
✅ AWS cloud infrastructure
✅ Neo4j graph database (1,245 nodes, 3,936 edges)
✅ Balanced dataset (21K each platform)
✅ Statistical significance (p<0.001)
✅ Production visualizations

---

🚀 Quick Start

git clone https://github.com/Safae211/Review_Vision.git
cd Review_Vision

pip install pandas numpy scipy matplotlib seaborn
pip install vaderSentiment boto3 neo4j selenium
pip install google-api-python-client

aws configure
neo4j start

---

👤 Author

Safae Ouâllal
Data Scientist | Multi-Modal Analytics

💼 LinkedIn

---

🙏 Acknowledgments

• Kaggle for datasets
• YouTube Data API
• AWS Rekognition
• Neo4j

---

⭐ Star this repo if you found it helpful!