📊 Product Recommendations POC

🔤 Textual Semantic Similarity vs 🕸️ Graph-Based Recommendations

What You'll Discover

This POC compares three AI-powered recommendation approaches on 20,000 products:

🔤 Textual Semantic Similarity
Finds products with seantically similar titles and descriptions
🕸️ Graph-Based Products
Discovers connections through shared metadata created by LLM (season, occasion and theme)
📂 Graph-Based Categories
Suggests relevant categories to explore from the same graph-based approach as Graph-Based Products recommendation

Key Finding: Only 4% overlap between textual semantic similarity and graph-based product recommendations (comparing top 20 recommendations). This means 96% of graph recommendations are unique - finding cross-category connections that textual semantic similarity cannot detect (e.g., brake pads → winter tire chains).

👀 See Examples First

🎯
All Three Recommendations Together
Start here! See side-by-side examples showing textual semantic similarity, graph-based products, AND graph-based categories for the same product.
👈 Start Here
🔍
Side-by-Side Case Studies
Real product examples comparing textual semantic similarity vs graph-based recommendations. See exactly how the recommendations differ.
Quick Examples

🎮 Try It Yourself

🎯
Interactive Demo
Search any product and explore recommendations. Switch between textual semantic similarity, graph-based, or compare both approaches side-by-side.
Interactive

📊 Understand the Differences

📊
Which Categories Benefit Most?
Heatmap showing recommendation overlap by category. Red zones show where graph-based adds the most unique value. Strategic for prioritization.
Strategic Insights
📈
Cross-Category Discovery
Metrics showing how graph-based recommendations discover products across more categories than textual semantic similarity. Quantifies "discovery potential."
Metrics

🔬 How It Works (Technical)

🌊
The Graph Structure
Interactive Sankey diagram showing how categories connect through seasons, occasions, and themes. Explains WHY graph-based finds cross-category connections.
🔵 Season 🟢 Occasion 🟠 Theme
🗺️
The Embedding Space
2D map of 20,000 products showing natural clustering by similarity. Zoom, pan, and explore where products land in semantic space.
Exploratory

🎯 Recommended 10-Minute Journey

  1. See examples (2 min): All Three Recommendations Together - understand what you're comparing
  2. Try it yourself (3 min): Interactive Demo - search products, explore differences
  3. Strategic insights (2 min): Which Categories Benefit Most? - see where graph adds value
  4. More examples (2 min): Side-by-Side Case Studies - real product comparisons
  5. How it works (1 min): The Graph Structure - understand the "why"

💡 Short on time? Just view All Three Recommendations Together and Which Categories Benefit Most? for the key insights.

20,000 products analyzed | 2 AI approaches tested | 18% overlap = Complementary value ✓