Amazon Electronics · 1,500+ users · 5,600+ products

Predicting the star
before it's given.

Three recommendation models — trained on real purchase ratings — compete to guess what a shopper hasn't rated yet. Below: the leaderboard, the models, and live predictions per user.

4.87predicted for a user who's never seen this product
View the full notebook on GitHub

Three models, one question

Lower RMSE means predicted ratings land closer to what users actually gave. Precision/Recall/F1 measured at k=10.

User-User KNN

0.992rmse
Precision0.836
Recall0.882
F10.858

Item-Item KNN

0.996rmse
Precision0.830
Recall0.876
F10.852

SVD (Matrix Factorization)

Best RMSE
0.933rmse
Precision0.842
Recall0.873
F10.857

Highest rated, minimum 50 reviews

No personalization here — just what the crowd actually loved.

01B001TH7GUU
4.92
02B003ES5ZUU
4.82
03B002WE6D44
4.76
04B000N99BBC
4.76
05B002V88HFE
4.75
06B004CLYEFK
4.73
07B0034CL2ZI
4.70
08B004CLYEDC
4.65
09B000JMJWV2
4.64
10B000LRMS66
4.51

Pick a shopper, see their picks

Predictions from the SVD model, for products this user hasn't rated yet.

#1B000N99BBC
5.00
#2B003NR57BY
5.00
#3B002WE0QN8
5.00
#4B00029U1DK
4.99
#5B0010Z28XG
4.99