This is the Personalized Product Recommendation System. Three complementary models turn shopping signals into a considered top-10 collection. No hard-coded recommendation scores. Just real calculations, made understandable.
Every interaction helps us understand what you love.
Views, saves & purchases
Weighted signals
A map of shared taste
Three models, one score
Already purchased? Excluded.
New favorites, just for you
We build a user-item matrix from weighted interactions, find similar users using cosine similarity, and aggregate their product preferences. The neighbor scores are normalized before combining.
Product name, category, brand, description and tags become normalized TF-IDF vectors. Your activity forms a weighted content profile, which we compare with each product using cosine similarity.
Views, clicks, saves, bag additions and purchases each contribute weighted counts. These totals are normalized across the catalog. For new users, popularity provides the entire ranking until an activity profile exists.
A purchase says more than a glance. Interaction weights reflect how strongly a user expresses interest.
1
view
2
click
4
wishlist
5
cart
8
purchase
After score normalization, the hybrid combines 0.50 × collaborative + 0.30 × content + 0.20 × popularity. Already-purchased products are filtered out, and the 10 highest-scoring eligible products are returned with an explanation.
Try different sample profilesWe sort events by timestamp, train on the first 80%, and hold out the last 20%. Held-out wishlist, cart and purchase events define relevance. Previously purchased training items are excluded from evaluation. Precision@10 measures the share of relevant recommendations, Recall@10 the share of relevant items found, and NDCG@10 rewards placing relevant products higher in the ranking.
This is an educational evaluation over synthetic, category-correlated seed data. It is not a production accuracy claim. The app currently runs equivalent TF-IDF and matrix math in TypeScript, with saved records in the built-in key-value store—not a Python service.
Explore real evaluation resultsGood finds. Delivered.
Free standard shipping on every order
Room to change your mind
A thoughtful 30-day return policy
Safe from start to finish
Free demo checkout — no real charges