This document discusses Netflix's global deep learning recommender system model. It describes how Netflix recommends content to over 150 million members across 190 countries using personalized recommendations. The system utilizes collaborative filtering techniques like soft clustering models to group users with similar tastes and generate weighted popularity votes. It also leverages topic models to model users' tastes as distributions over topics and content. The challenges of scaling these models globally to account for factors like country-specific catalogs and trends over time are discussed. The solution presented is to incrementally train the models by first censoring unavailable content and adding contextual variables, then periodically training warm start models with new embeddings and parameters to efficiently update the models at scale.
Boost Fertility New Invention Ups Success Rates.pdf
Building Incrementally Trained Global Deep Learning Recommender System
1. Building an Incrementally
Trained Global Deep Learning
Recommender System Model
Anoop Deoras, Ko-Jen (Mark) Hsiao
adeoras@netflix.com
MLConf, San Francisco
11/08/2019
@adeoras
4. ● Recommendation Systems are means to an end.
● Our primary goal:
○ Maximize Netflix member’s enjoyment of the selected show
■ Enjoyment integrated over time
○ Minimize the time it takes to find them
■ Interaction cost integrated over time
Personalization
● Personalization
13. Basic Intuition behind Collaborative Filtering
● Imagine you walked into a room full of movie enthusiasts, from all over
the world, from all walks of life, and your goal was to come out with a
great movie recommendation.
● Would you obtain popular vote ? Would that satisfy you ?
14. Basic Intuition behind Soft Clustering Models
● Now consider forming groups of people with similar taste based on the
videos that they previously enjoyed.
15.
16.
17.
18. Basic Intuition behind Soft Clustering Models
● Describe yourself using what you have watched.
● Try to associate yourself with these groups and obtain a weighted
“personalized popularity vote”.
20. Topic Models (Latent Dirichlet Alloc)
K
U
P
α θ φt v
β
Total
Topics
Taste
Convex Combinations of
topics proportions and movie
proportions within topic
23. Country Context in LDA models
Users in Country A play both Friends and HIMYM Users in Country B cannot play both Friends and
HIMYM
Country A catalog Country B catalog
Model is forced to split HIMYM plays.
topic k
Outcome: Parameters are being consumed to explain catalog differences.
topic j
Topic with
high mass
on Friends
and HIMYM
Topic with
high mass
on HIMYM
Thanks to Ehtsham Elahi for contributing this slide.
24. Catalogue Censoring in Topic Models
K
U
P
α θ φt v
β
Total
Topics
Taste
c
Censoring
pattern
m
Global Recommendation System for Overlapping Media Catalogue, Todd et.al., US Patent App
26. Time context in Topic Models
K
U
P
α θ φt v
β
Total
Topics
Taste
m
Observed
time
µ
Topics over Time: A Non Markov Continuous-Time Model fo Topic Trends. , Wang et.al., KDD 2006
27. Fully contextualizing Topic Models
K
U
P
α θ φt v
β
Total
Topics
Taste
m
Observed
time
µ
c
Censoring
pattern
m
37. RECIPE
1. CENSOR
2. ADD CONTEXT VARIABLES TO THE MODEL
3. DO; EVERY FEW DAYS
a. TRAIN A WARM START MODEL WITH (1 & 2)
4. DO; EVERY FEW HOURS
a. TAKE THE MODEL FROM (3)
b. ADD NEW EMBEDDINGS
c. ADD NEW PARAMETERS
d. FINE TUNE
38. THANK YOU !
Questions ?
Anoop Deoras, Ko-Jen (Mark) Hsiao
adeoras@netflix.com
@adeoras
Sincere thanks to a lot of my Netflix Colleagues: Aish Fenton, Dawen Liang and
Ehstham Elahi for contributing to the ideas discussed here.