Building Next-Generation Audience Targeting and Analytics at NBCUniversal
summary
NBCUniversal builds a cloud data lake that runs Apache Mesos/DC/OS with Spark, Kafka and Cassandra, storing Parquet on S3 to avoid HDFS sprawl. It exposes data via a Spark Thrift server and REST APIs so BI tools and notebooks query raw parquet instead of copying data. Audience Studio stitches batch and streaming signals into audience profiles, processes roughly 100 TB and supports a $1 billion ad-sales initiative out of a $10B business while improving targeting and pacing guarantees.
AI-written summary · about
chapters (10)
- 0:00Start
- 0:39Content & distribution are in a major state of flux
- 7:38NBCUniversal's Portfolio
- 10:48INTRODUCING AUDIENCE STUDIO
- 13:35Advanced advertising technology Goals
- 14:51Audience targeting value proposition
- 22:47Audience Studio - Data Lake Stores
- 26:59Audience Studio Data Lake access Points
- 28:48Audience Studio Analytic Tool Sets
- 31:16DCIOS Packages
related talks
49mData Platform Architecture, Evolution, and Philosophy
05mHadoop in the cloud: A Nielsen use case
30mAn architecture for big data: Supporting journalism through transparency in advertising
56mSpark-driven Audience counting
29mData foundation for analytics and AI
17mDemocratising data analytics
01h 03mDiscovering the value of a cloud data platform
39mAnalysing Public Kafka Data from NASA Satellites