Ads Personalization at Spotify - NYC Data Engineering 10/23
11 likes2,185 views
The document discusses how Spotify utilizes ad personalization to enhance user experience through iterative engineering and product development. It outlines various ad targeting architectures and technologies employed to create context-aware and real-time targeted ads based on users' music preferences and activities. Additionally, it highlights the importance of user feedback in refining ad strategies and improving overall engagement.
Ads Personalization at Spotify - NYC Data Engineering 10/23
- 1. Ad Personalization at Spotify Building personalized ad experiences through iterative engineering and product development Kinshuk Mishra Noel Cody
- 5. Music… ...is with you throughout the day. ...fits your mood. ...fits your activity.
- 9. If your day looks like this: Wake up Work out Commute Focus at Work Relax at Home Sleep
- 10. Ads should follow. Wake up Work out Commute Focus at Work Relax at Home Sleep Ads
- 11. Ads should follow. Wake up Work out Commute Focus at Work (Classical) Relax at Home Sleep Electronic Music ad Not bad. WTF?
- 13. Why Personalization? “...it works well the advertisements are annoying though I am not a fan of mainstream music so hearing about pop bands is also driving me crazy” “Great way to listen to whatever music you want. The ads can be really annoying though since they don't seem to be targeted. I HATE rap music, yet I seem to get a lot of ads for it.”
- 14. Data confirms anecdotal evidence
- 18. User Stories Hypotheses + Goals Product MVPs + Experiments
- 19. User Stories to Hypotheses & Goals: ● Context-aware ads ● Music ads like music recommendations ● Ads that learn
- 20. Hypotheses to Products: ● Real-time genre targeting ● Historic genre targeting ● Real-time moment targeting
- 21. (Product MVPs to Experiments) Control Variation 1 Variation 2
- 22. INFRASTRUCTURE
- 23. Ad Targeting Architecture Feedback Loop
- 24. Ad Targeting Architecture OSS Data Infrastructure Spotify Backend Infrastructure
- 25. Ad Targeting Architecture V1.0 COTS Data Infrastructure Real-time Targeting Spotify Backend Infrastructure
- 26. Ad Targeting Architecture V2.0 Real-time + Batch Targeting (a.k.a. Lambda Architecture)
- 27. Ad Targeting Architecture V2.5 Transition to Persistent User Profile
- 28. Ad Targeting Architecture V3.0 Richer Profile Schema with Persistence
- 29. Tech Choices
- 30. Kafka ● Kafka is a distributed, partitioned, replicated commit log service. ● Guarantees ● Kafka provides a total order over messages within a partition ● Fault tolerance : handles N-1 failures for replication factor N.
- 31. Ad Targeting Architecture V1.0 COTS Data Infrastructure Real-time Targeting Spotify Backend Infrastructure
- 32. SSttoorrmm ● Real time stream processing ● Like hadoop without HDFS ● Like Map/Reduce with many reducer steps ● Fault tolerant and guaranteed message processing
- 33. Storm: Testing (since 0.8.1) Storm
- 34. Storm: Visualization (since 0.9.2) Storm
- 35. Ad Targeting Architecture V2.0 Real-time + Batch Targeting
- 36. Apache Crunch ● Framework for writing, testing, and running MapReduce pipelines ● Pipelines are composed of user-defined functions and higher-level abstractions of common MR tasks (filter, join, etc.)
- 37. Apache Crunch Data structures: ● PCollection<T> ● PTable<K,V> ● PGroupedTable<K,V> Functions: ● MapFn<T1,T2>: T1 → T2 ● CombineFn<K,V>: (K, Iterable<V>) → (K, V)
- 38. Apache Crunch What’s wrong with plain Python Streaming MapReduce? ● Testability ● Optimization ● Performance ● IDE support ● Type Safety ● Lack of higher-level operations (filter/join/aggregate) From Spotify Presentation: Scalding the Crunchy Pig for Cascading into the Hive
- 39. Apache Crunch ● About a 5x performance improvement over Python streaming MapReduce ● Readable functional-style API in plain Java ● Great local testing support ● First-class support for Avro records. From Spotify Presentation: Scalding the Crunchy Pig for Cascading into the Hive
- 40. Apache Crunch
- 41. Apache Crunch
- 42. Ad Targeting Architecture V2.5 Transition to Persistent User Profile
- 43. CASSANDRA Rich wide-column schema support Solid persistence and replication Slower reads MEMCACHED K/V only TTL is default (in-memory ● Rich schema ● Persistence mgmt) vs.
- 44. Ad Targeting Architecture V3.0 Richer Profile Schema with Persistence
- 45. DATA INGESTION: CASSANDRA CRUNCH STORM HDFS KAFKA LOGS
- 46. TESTING
- 48. User Stories Hypotheses + Goals Product MVPs + Experiments
- 49. AAR Vital Signs Ad-Specific Metrics
- 50. AAR Vital Signs Higher-level metrics are hard to move Ad-Specific Metrics
- 51. USER EXPERIENCE TEST ITERATION IMPACTS AAR
- 52. AAR Vital Signs Our focus Ad-Specific Metrics
- 53. Test evaluation ● Positive Signals: CTR, Downstream Effects ● Avoidance Signals: Volume, Audio Output ● An “Ad Quality Score”
- 54. Thanks! (We’re hiring): spotify.com/us/jobs/