32
Music: Tuned to you Mohitdeep Singh Data Scientist Predictive Analytics Innovation Summit Feb 12-13, 2015 San Diego

Music: Tuned to you

Embed Size (px)

Citation preview

Page 1: Music: Tuned to you

Music: Tuned to you Mohitdeep Singh Data Scientist!

Predictive Analytics Innovation Summit Feb 12-13, 2015 San Diego

!

Page 2: Music: Tuned to you

!

Page 3: Music: Tuned to you

! http://www.rdio.com/about/!

Page 4: Music: Tuned to you

!

Page 5: Music: Tuned to you

!

Big Data @Rdio!

Tracks metadata!

Signal Processing!

Millions of hrs of music streamed every month!

Clicks!User Demography!

Social Info!

Every single interaction!

Page 6: Music: Tuned to you

!

Committed to opensource!

Page 7: Music: Tuned to you

!

Page 8: Music: Tuned to you

!

Scenario !

Page 9: Music: Tuned to you

!

Scenario !

Page 10: Music: Tuned to you

!

The answer lies in the matrix!

Page 11: Music: Tuned to you

!

2! 7! 44!22! 17!

9! 12!21! 18!77! 44!

Page 12: Music: Tuned to you

!

Baseline -Popularity!Recommend based on popularity of tracks!Pros:!

•  Again, a very simple model!

•  Easy to implement!

•  More efficient on Apache Giraph(by exploiting its property)!

•  Always a good baseline!

Cons:!•  Not really recommending anything!•  No element of discovery!

Page 13: Music: Tuned to you

!

Long Tail Problem!

Page 14: Music: Tuned to you

!

Nearest Neighbors!

2! 7! 44!22! 17!

9! 12!21! 18!77! 44!

Page 15: Music: Tuned to you

!

Distance matrix!

1! 0! 0! 0.0873! 0! 1! 0! 0!0! 1! 0! 0! 0! 0! 0! 0.3603!

0! 0! 1! 0! 1! 0! 0! 0!0.0873! 0! 0! 1! 0! 0.0873! 0.2621! 0.8967!

0! 0! 1! 0! 1! 0! 0! 0!..! ..! ..! ..! ..! ..! ..! ..!

Page 16: Music: Tuned to you

!

Top-N Recommendations!

*!

≈!

P = R* D!

Page 17: Music: Tuned to you

!

Top-N Recommendations!

*!

≈!

P = R* D!

Page 18: Music: Tuned to you

!

Pros!

•  Easy to reason models!

•  Easily scaled via Map Reduce.!

•  Gives decent performance on test set!

Cons!

•  If users and the items space are not stable, then things can and will go wrong.!

•  Lacks serendipity.!

•  No guarantee on the number of predictions/user. !

!

Page 19: Music: Tuned to you

!

Latent Factor Models!

Approach pioneered during Netflix Prize Competition.!

Key idea is to decompose rating matrix into multiple lower rank approximations.!

Page 20: Music: Tuned to you

!

≈! *! =!

Page 21: Music: Tuned to you

!

≈! *! =!

Page 22: Music: Tuned to you

!

≈! *! =!

Page 23: Music: Tuned to you

!

Pros!

•  Tries to learn the underlying concepts!

•  User/ item supplementary information can be baked in into learning algorithm (factorization machines).!

!

Cons:!

•  Doesn’t perform as well as simple nearest models!

•  Interpretation of latent space is hard.!

!

Page 24: Music: Tuned to you

!

Bayesian Personalized Ranking!•  Constructs a preference order for each user!

•  Directly optimizes the ranking function!

•  Takes into account the order preference.!

•  Implemented in scalable fashion on top of Apache Giraph!

Page 25: Music: Tuned to you

!

Results!

Popularity!

Nearest Neighbors!

Matrix Factorization!

Weighted Matrix Factorization!

Bayesian PR!

100%!50%!0! 150%!

Comparison of algorithms considering popularity as baseline!

Note: Offline metrics tracking MAP!

Page 26: Music: Tuned to you

!

Candidate Tracks !Catalogue of around 32M tracks!

Page 27: Music: Tuned to you

!

 P(Relevant | , Artist)!!

Track Id! Artist similarity! Track popularity! Artist popularity! Track duration! ..! ..! ..! ..! Relevant!0/1!

“My December”!

1! 0.992! 0.433! 482! ..! ..! ..! ..! 1!

‘’Shake it Off”! 0.03!!

0.04!!

0.88! 329!!

0!!

“Sugar”! 0.772!!

0.95!!

0.77!!

220!!

1!!

Page 28: Music: Tuned to you

!

Page 29: Music: Tuned to you

!

Many open problems!

Page 30: Music: Tuned to you

!

It’s a tough problem!!!

Page 31: Music: Tuned to you

!

Current/Future work!

•  Build an ensemble model to incorporate other models.!

•  Simplify A/B testing framework.!

•  Integrate content based recommendations.!

•  Experimenting with some deep-learning techniques.!

•  Incorporate information from the www.!

Page 32: Music: Tuned to you

!

Questions

Interested: Checkout https://www.rdio.com/careers/