Recommender

Systems
Problem
formula4on
Machine Learning
 Example: Predic/ng movie ra/ngs
User rates movies using one to ve stars
Movie Alice (1) Bob (2) Carol (3) Dave (4)
Love at last
Romance forever = no. users
Cute puppies of love = no. movies
= 1 if user has
Nonstop car chases
rated movie
Swords vs. karate
= ra4ng given by
user to movie
(dened only if
)
Andrew Ng
 Recommender
Systems
Content-based
recommenda4ons

Machine Learning
Content-based recommender systems
Movie Alice (1) Bob (2) Carol (3) Dave (4)
(romance) (ac/on)
Love at last 5 5 0 0 0.9 0
Romance forever 5 ? ? 0 1.0 0.01
Cute puppies of love ? 4 0 ? 0.99 0
Nonstop car chases 0 0 5 4 0.1 1.0
Swords vs. karate 0 0 5 ? 0 0.9

For each user , learn a parameter . Predict user as

Andrew Ng
Problem formula/on
if user has rated movie (0 otherwise)
ra4ng by user on movie (if dened)
= parameter vector for user
= feature vector for movie
For user , movie , predicted ra4ng:
= no. of movies rated by user
To learn :

Andrew Ng
Op/miza/on objec/ve:
To learn (parameter for user ):

To learn :

Andrew Ng
Op/miza/on algorithm:

Gradient descent update:

Andrew Ng
 Recommender
Systems
Collabora4ve
ltering
Machine Learning
Problem mo/va/on
Movie Alice (1) Bob (2) Carol (3) Dave (4)
(romance) (ac/on)
Love at last 5 5 0 0 0.9 0
Romance forever 5 ? ? 0 1.0 0.01
Cute puppies of ? 4 0 ? 0.99 0
love
Nonstop car 0 0 5 4 0.1 1.0
chases
Swords vs. karate 0 0 5 ? 0 0.9

Andrew Ng
Problem mo/va/on
Movie Alice (1) Bob (2) Carol (3) Dave (4)
(romance) (ac/on)
Love at last 5 5 0 0 ? ?
Romance forever 5 ? ? 0 ? ?
Cute puppies of ? 4 0 ? ? ?
love
Nonstop car 0 0 5 4 ? ?
chases
Swords vs. karate 0 0 5 ? ? ?

Andrew Ng
Op/miza/on algorithm
Given , to learn :

Given , to learn :

Andrew Ng
Collabora/ve ltering
Given (and movie ra4ngs),
can es4mate

Given ,
can es4mate

Andrew Ng
 Recommender
Systems
Collabora4ve
ltering algorithm
Machine Learning
Collabora/ve ltering op/miza/on objec/ve
Given , es4mate :

Given , es4mate :

Minimizing and simultaneously:

Andrew Ng
Collabora/ve ltering algorithm
1. Ini4alize to small random values.
2. Minimize using gradient
descent (or an advanced op4miza4on algorithm). E.g. for
every :




3. For a user with parameters and a movie with (learned)
features , predict a star ra4ng of .

Andrew Ng
 Recommender
Systems
Vectoriza4on:
Low rank matrix
factoriza4on
Machine Learning
Collabora/ve ltering
Movie Alice (1) Bob (2) Carol (3) Dave (4)
Love at last 5 5 0 0
Romance forever 5 ? ? 0
Cute puppies of ? 4 0 ?
love
Nonstop car 0 0 5 4
chases
Swords vs. karate 0 0 5 ?

Andrew Ng
Collabora/ve ltering
Predicted ra4ngs:

Andrew Ng
Finding related movies
For each product , we learn a feature vector .

How to nd movies related to movie ?

5 most similar movies to movie :

Andrew Ng
 Recommender
Systems

Implementa4onal
detail: Mean
normaliza4on
Machine Learning
 Users who have not rated any movies
Movie Alice (1) Bob (2) Carol (3) Dave (4) Eve (5)
Love at last 5 5 0 0 ?
Romance forever 5 ? ? 0 ?
Cute puppies of love ? 4 0 ? ?
Nonstop car chases 0 0 5 4 ?
Swords vs. karate 0 0 5 ? ?

Andrew Ng
Mean Normaliza/on:

For user , on movie predict:

User 5 (Eve):

Andrew Ng