Duke AIPI 540 · Module 3 Project

DeepReads
Explainable
AI Recommendations

A Two-Tower neural recommender trained on Amazon Books — powered by BPR loss, metadata-grounded explanations, and benchmarked against classical and popularity baselines.

Two-Tower Neural Network
BPR Loss
Explainable Recommendations
GitHub →
100K+Books
80K+Users
5M+Ratings
Scroll
Step 1

Choose a Persona

Select a demo user profile to see personalized recommendations, or build your own by picking items manually.

Step 2

Your Recommendations

Powered by our best-performing Two-Tower neural network.

🎯

Select a persona above to get personalized book recommendations.

🎯

Select a persona and click Get Recommendations to see personalized results.

Step 3

Model Comparison

See how all three models rank books differently for the same user. Items appearing in all three lists are highlighted.

⚖️

Select a persona first, then compare all three models side-by-side.

📊

Naive (Popularity)

Recommends globally popular items

🌲

Classical (LightGBM)

Feature-based reranker with SHAP

🧠

Deep (Two-Tower)

Neural embeddings with BPR loss