Deep Learning for Recommender Systems Justin Basilico & Yves - PowerPoint PPT Presentation

Deep Learning for Recommender Systems Justin Basilico & Yves Raimond March 28, 2018 GPU Technology Conference @JustinBasilico @moustaki

The value of recommendations A few seconds to find something ● great to watch… Can only show a few titles ● Enjoyment directly impacts ● customer satisfaction Generates over $1B per year of ● Netflix revenue How? Personalize everything ●

Deep learning for recommendations: a first try

Traditional Recommendation Setup Users 1 1 0 0 0 1 1 0 0 0 Items 1 1 1 0 0 1 0 0 0 0 1 0 0 0 0

A Matrix Factorization view V ≈ R U

A Feed-Forward Network view U V

A (deeper) feed-forward view U ? Mean squared loss V

A quick & dirty experiment ● ○ ○ ● ○ ■ ■ ○ ■ ■ ■ ■ ■ ●

GPU vs. CPU ● ● ●

What’s going on? ● ● ● ●

Conclusion? ● ●

Breaking the ‘traditional’ recsys setup ● ● ●

Alternative data

Content-based side information ● ● ●

Metadata-based side information ● ○ ● ○ ● ●

YouTube Recommendations ● ●

Alternative models

Restricted Boltzmann Machines ● ● ●

Auto-encoders ● ● ○ ● ● ●

prod2vec (Skip-gram) (*)2Vec ● ● ● user2vec (Continuous Bag of Words)

Wide + Deep models ● ● [Cheng et. al., 2016]

Alternative framings

Sequence prediction ● ○ ○ ● ○ ○ ●

Contextual sequence prediction ● ● ● ●

Contextual sequence data Sequence Context Action per user 2017-12-10 15:40:22 2017-12-23 19:32:10 2017-12-24 12:05:53 Time 2017-12-27 22:40:22 2017-12-29 19:39:36 ? 2017-12-30 20:42:13

Time-sensitive sequence prediction ● ○ ● ○ ■ ● ● ■ ○

Other framings ● ○ ● ○ ●

Conclusion

Takeaways ● ● ● ●

More Resources ● ● ● ● ● ●

Thank you. Justin Basilico & Yves Raimond @JustinBasilico @moustaki Yes, we’re hiring...