Interactive Recommender Systems
11 likes3,089 views
The document discusses interactive recommender systems, specifically focusing on platforms like Netflix and Spotify, and their methods for enhancing user experience through personalized recommendations based on user feedback. It highlights challenges such as user interaction design, algorithmic efficiency, and the balance between exploration and exploitation of user preferences. Key insights include the importance of positioning interactive features within a product to ensure user engagement and satisfaction.
![Implicit Matrix Factorization [1] 78
1 0 0 0 1 0 0 1
0 0 1 0 0 1 0 0
1 0 1 0 0 0 1 1
0 1 0 0 0 1 0 0
0 0 1 0 0 1 0 0
1 0 0 0 1 0 0 1
•Aggregate all (user, track) streams into a large matrix
•Goal: Approximate binary preference matrix by inner product of 2 smaller matrices by minimizing the
weighted RMSE (root mean squared error) using a function of plays, context, and recency as weight
X YUsers
Songs
• = bias for user
• = bias for item
• = regularization parameter
• = 1 if user streamed track else 0
•
• = user latent factor vector
• = item latent factor vector
[1] Hu Y. & Koren Y. & Volinsky C. (2008) Collaborative Filtering for Implicit Feedback Datasets 8th IEEE International Conference on Data Mining](https://image.slidesharecdn.com/interactiverstutorial-recsys-2015-150916221009-lva1-app6891/85/Interactive-Recommender-Systems-78-320.jpg)
![Can also use Logistic Loss! [2] 79
1 0 0 0 1 0 0 1
0 0 1 0 0 1 0 0
1 0 1 0 0 0 1 1
0 1 0 0 0 1 0 0
0 0 1 0 0 1 0 0
1 0 0 0 1 0 0 1
•Aggregate all (user, track) streams into a large matrix
•Goal: Model probability of user playing a song as logistic, then maximize log likelihood of binary
preference matrix, weighting positive observations by a function of plays, context, and recency
X YUsers
Songs
• = bias for user
• = bias for item
• = regularization parameter
• = user latent factor vector
• = item latent factor vector
[2] Johnson C. (2014) Logistic Matrix Factorization for Implicit Feedback Data NIPS Workshop on Distributed Matrix Computations](https://image.slidesharecdn.com/interactiverstutorial-recsys-2015-150916221009-lva1-app6891/85/Interactive-Recommender-Systems-79-320.jpg)
![82
Deep Learning on Audio [3]
[3] http://benanne.github.io/2014/08/05/spotify-cnns.html](https://image.slidesharecdn.com/interactiverstutorial-recsys-2015-150916221009-lva1-app6891/85/Interactive-Recommender-Systems-82-320.jpg)
Interactive Recommender Systems
- 1. Interactive Recommender Systems RecSys 2015 Harald Steck Netflix Inc. hsteck@netflix.com Roelof van Zwol Netflix Inc. roelofvanzwol@netflix.com Chris Johnson Spotify Inc. cjohnson@spotify.com
- 3. Introduction
- 5. What are Interactive RSs? Define Interactive: Influencing or having an effect on each other. Allowing a two-way flow of information Responding to a user’s input Forms of Interaction Implicit vs explicit feedback User experiences SearchBrowse Interactive RS
- 6. Need for Interactive RSs? To better serve the user’s (entertainment) needs Fill the gap between browse and search Lean-back vs precise need To establish trust and credibility Instant gratification
- 7. Interactive RS in practice? Most online services offer different levels of personalization Video: Netflix, YouTube, Hulu, Amazon Video, HBO, … Browse (homepage/genre) and search Music: Spotify, Pandora, Apple Music, Google Music, … Browse, Radio, Search Books/Apps/Shopping: Amazon, Apple, Google (Video and Image) Search: Google, Bing, Yahoo Left rail suggestions (explore), search completion Travel: Booking.com, Kayak, Orbitz, Expedia, … Guided experience: Pick destination —> pick flight and hotel, arrange local transport, plan activities, go dining and sightseeing…
- 8. Research on Interactive RS “A successful recommender system must balance the needs to explore user preferences and to exploit this information for recommendation.” Exploration in Interactive Personalized Music Recommendation: A Reinforcement Learning Approach. X. Wang, Y. Wang, D. Hsu, Y. Wang (2013) Active learning / Multi-Armed Bandits (𝜀-greedy, UCB, Thompson sampling) Papers at RecSys this year and in the past :) When do explicit taste preferences trump play data? When we have insufficient knowledge about the user When context changes TasteWeights: A Visual Interactive Hybrid Recommender System. S. Bostandjiev, J. O’Donovan, T. Höllerer (RecSys 2012) When the user has a vague idea of what they want IntRS: Workshop on Interfaces and Human Decision Making for Recommender Systems
- 9. Use-case:
- 10. Section name 10
- 11. Section name 11
- 12. Section name 12
- 13. Section name 13
- 14. Interactive RSs @ Netflix Help user’s find a great video to watch as an alternative to browse and search. Address browse fatigue; failed search experience; or user has vague notion of what to watch; watching with friends Not for everyone, not all the time… Max - Rating Game Select a topic, rate some videos, get recommendations Hot-or-Cold Tell us what you’re into, one video at a time Discover Select one or more videos of interest to explore other similar options
- 15. Max - Rating Game
- 16. Hot or Cold Game
- 17. Discover App
- 18. Use-case:
- 19. Section name 19
- 20. Spotify in Numbers Started in 2006, now available in 58 markets 75+ million active users, 20 million paying subscribers 30+ million songs, 20,000 new songs added per day 1.5 billion playlists 1 TB user data logged per day 1k node Hadoop cluster 10,000+ Hadoop jobs run every day
- 21. Challenge: 20 Million songs... how do we recommend music to users?
- 22. Discover Weekly
- 23. Discover
- 24. Related Artists
- 25. Radio
- 26. Radio as an Interactive RS User has a vague idea of what they would like to listen to User starts a radio station seeded by an Artist, Album, Song, Playlist, or Genre Generally a lean back experience, but with the ability to tune recommendations with (+) and (-) feedback ( )
- 27. Section name 27 SmartBox: fully interactive radio experience
- 28. Challenges
- 29. Challenges Understanding when to present interactive experience - Binging, trying to find something new, experiencing browse fatigue X User Interaction Design - UI affordance - Messaging task - Efficient mechanism to collect user input - Instant gratification - Choice penalty - Navigation options (user control) X X
- 30. Challenges Workflow and Algorithmic Design - Computational efficient - Balancing relevancy, diversity and popularity of the recommendations - Support user navigation - Allow for both exploration and exploitation, e.g. avoid overfitting based on limited input from the user X X
- 32. Discover: Understanding the user
- 33. Meeting the users expectations Introducing interactive RS as a full page experience or integrate with other rows on the homepage?
- 34. Meeting the users expectations Introducing interactive RS as a full page experience or integrate with other rows on the homepage? Avoid conflict: browse vs discovery mode Browsing on the homepage is a lean-back experience Discover, is interactive… that means work
- 35. Understanding the user’s need Introducing Discover feature as a new destination Allow users to quickly navigate to feature
- 36. Understanding the user’s need Introducing Discover feature as a new destination Help users find new feature through messaging at top of homepage At session start After scrolling X rows down After 45 secs
- 37. Understanding the user’s need Introducing Discover feature as a new destination Help users find new feature through: messaging at top of homepage invitation in row on homepage
- 38. Understanding the user’s need Introducing Discover feature as a new destination Help users find new feature through: messaging at top of homepage invitation in row row-based invitation In position of Nth row or as last row on homepage
- 39. Understanding the user’s need Context is crucial for the success of the integration of interactive experiences into the product Never interrupt a user that knows what they want Message at start of user session Showing the invitation at the right moment (and place) is crucial Allow the user to explore options on the homepage first Challenge is to find the right trade-off between rate at which user tries the interactive experience, versus success Success rate can vary from 30-70% Not a goal to reach all users, just need to reach those who need help
- 40. Discover: Inside the interactive RS
- 41. Section name 41
- 42. Section name 42
- 43. Interaction Workflow for Discover Init - At start of new session, create an initial seed of videos to show to the user See Suggestions - After receiving input from the user (selected videos), compute recommendations
- 44. Interaction Workflow for Discover Need to find balance between relevancy, diversity and popularity of the videos shown to the user
- 45. Videos in a Latent Space Representation normalized item-vectors
- 46. Videos in a Latent Space Representation user-vector in same space
- 47. Relevance Score
- 49. Popularity Score
- 50. “See Suggestions” step Score of item i: , for adjusting the balance score(i) determined by “closest” item j in S “diminishing returns”
- 51. “See Suggestions” step Greedy computation of recommendation set S:
- 52. Interaction Workflow for Discover See Suggestions - After receiving input from the user (selected videos), compute recommendations What if the selected videos cover more than 1 genre?
- 53. User may select diverse set of videos Challenge: Recommendations should reflect all genres of the shows selected by user Solution: Multiple user-vectors, one per selected group of similar shows (clustering) Compute recommendations for each user vector Display clusters in separate rows, or combine via reciprocal rank aggregation
- 54. Interaction Workflow for Discover Init - At start of new session, create an initial seed of videos to show to the user Show diverse set of recognizable videos that are familiar to the user, and which is fresh across sessions
- 55. Producing the initial set of videos Objective: Show diverse set of recognizable videos that are familiar to the user, and which is fresh across sessions How? Sample videos from user’s play history score: no popularity & relevance component Sample shows from catalog with popularity bias score: no relevance component Randomize initial set for freshness
- 56. Learning the Latent Vectors
- 57. Learning the Latent Vectors Use implicit feedback data Compute item-item similarity matrix: covariance matrix transition matrix … Train (spectral) matrix factorization model
- 60. Graph Cut—Simplistic Example Cut here …not similar to Ballerina any more new similar of Ballerina:
- 61. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors 5. (kmeans) clustering of latent vectors ii j ij
- 62. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors 5. (kmeans) clustering of latent vectors ii j ij
- 63. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors 5. (kmeans) clustering of latent vectors ii j ij
- 64. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity (largely) removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors 5. (kmeans) clustering of latent vectors ii j ij
- 65. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity (largely) removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors 5. (kmeans) clustering of latent vectors ii j ij
- 66. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity (largely) removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors (remove remaining popularity) 5. (kmeans) clustering of latent vectors ii j ij
- 67. Graph Cut à Spectral Clustering 1. Graph à Affinity Matrix A 2. Laplacian: L = D … diagonal matrix, D = sum (A ) popularity (largely) removed prior to factorization 3. factorize the Laplacian L 4. normalize the latent vectors (remove remaining popularity) 5. (kmeans) clustering of latent vectors ii j ij
- 68. Example: Django Movies similar to: Viennese-born Christoph Waltz won Oscar for best supporting role only play-data used, no content information
- 69. Similar to Django via basic MF
- 70. Similar to Django via spectral MF
- 71. Example: Hercules in New York Movies similar to: Austrian-born Arnold Schwarzenegger’s debut movie only play-data used, no content information
- 72. Similar to Hercules in NY via basic MF
- 73. Similar to Hercules in NY via spectral MF
- 74. Deep Dive: Radio
- 75. Radio as an Interactive RS User has a vague idea of what they would like to listen to User starts a radio station seeded by an Artist, Album, Song, Playlist, or Genre Generally a lean back experience, but with the ability to tune recommendations with (+) and (-) feedback ( )
- 76. Overview of Radio Dataflow
- 77. Step 1: Train batch recommendation models
- 78. Implicit Matrix Factorization [1] 78 1 0 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 1 0 0 0 1 1 0 1 0 0 0 1 0 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 •Aggregate all (user, track) streams into a large matrix •Goal: Approximate binary preference matrix by inner product of 2 smaller matrices by minimizing the weighted RMSE (root mean squared error) using a function of plays, context, and recency as weight X YUsers Songs • = bias for user • = bias for item • = regularization parameter • = 1 if user streamed track else 0 • • = user latent factor vector • = item latent factor vector [1] Hu Y. & Koren Y. & Volinsky C. (2008) Collaborative Filtering for Implicit Feedback Datasets 8th IEEE International Conference on Data Mining
- 79. Can also use Logistic Loss! [2] 79 1 0 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 1 0 0 0 1 1 0 1 0 0 0 1 0 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 •Aggregate all (user, track) streams into a large matrix •Goal: Model probability of user playing a song as logistic, then maximize log likelihood of binary preference matrix, weighting positive observations by a function of plays, context, and recency X YUsers Songs • = bias for user • = bias for item • = regularization parameter • = user latent factor vector • = item latent factor vector [2] Johnson C. (2014) Logistic Matrix Factorization for Implicit Feedback Data NIPS Workshop on Distributed Matrix Computations
- 80. NLP models on News, Blogs, and Text 80
- 81. 81 NLP models also work on playlists! Playlist itself is a document Songs in playlist are words
- 82. 82 Deep Learning on Audio [3] [3] http://benanne.github.io/2014/08/05/spotify-cnns.html
- 83. Songs in a Latent Space Representation normalized item-vectors
- 84. Songs in a Latent Space Representation user-vector in same space
- 85. Step 2: Ensemble of batch models using explicit thumbs data
- 86. Ensemble of Batch Models
- 87. Explicit feedback at the expense of presentation bias…
- 88. Augment explicit feedback with random negatives
- 89. Thumbs also carry a subjective bias… Thumbs used to personally tune a user’s station But with enough data, they’re a pretty good signal
- 90. Interesting fact… Ratio of to does not always correlate with user engagement!
- 91. Step 3: Static song pools adaptively personalized as user interacts
- 92. Should be Simple, fast, and instantly gratifying
- 93. Adaptive personalization Probability of choosing song given user , station , at time = global rank of track in song pool = relevancy to user based on implicit CF vector = relevancy to user based on explicit thumbs = diversity of artist and album within radio session
- 94. Step 4: Radio thumbs feed back into ensemble!
- 96. Key Take Aways Implicit feedback used to train batch recommendation models Explicit feedback is augmented with random negatives and used as training data to combine batch models (feedback loop) Adaptive shuffling is a function of song pool rank, user’s implicit CF relevancy, user’s thumbs relevancy, and diversity of artist and album Interaction needs to be simple, fast, and instantly gratifying to the user
- 97. Insights
- 98. Insights - User perspective Positioning of interactive RS in a product is crucial to success Natural transition from browse and search to interaction Interactive RS needs to be lightweight.. Most users prefer lean-back experience Interaction means work —> instant gratification Evidence helps build trust and transparency Choice penalty How much input is needed and how many results to return
- 99. Insights - Algorithmic perspective Need to balance popularity, relevancy, and diversity of the recommendations during different phases of the interactive workflow. Real-time scoring vs offline computation of recommendations Video and music domain share many algorithmic characteristics, yet the application requires vastly different implementation On using implicit vs explicit feedback: What users say they like is not always what they watch/ listen