Improve your marketing reach using ...
summary
Case study using a Hadoop/Spark cluster on 6,000,000 clients builds a daily test-and-learn pipeline: Hive/Pig for ETL, Apache Spark / Spark ML for model training, and DMP tagging to push campaigns. Uplift models (independent models using random forest and logistic regression) pick which of two messages to send; initial randomization sizes are ~100k–300k. Reported outcomes show measurable positive uplift for specific segments but warn of organizational friction, slow POC cycles, and fragile control-group joins that can invalidate results.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:40Can we really do better than regular scori
- 4:29Observed scoring approach limitation
- 6:03When one message is not enough
- 6:39Test & learn on the client database
- 7:41Test & learn multiple messa
- 13:43Number of models: N messages X MC
- 17:36Uplift model : improve an effect treatment Use case observed uplift and marketing
related talks
32mWhy start using uplift models for more efficient marketing campaigns
01h 04mOnline Advertising: Hybrid Approach to Large-Scale Data Analysis
41mFrom 5 PB of Daily Logs to Thousands of ML Models #bbuzz
31mML practices in the industry
01h 35mDeep Dive into the Big Data Landscape
37mBig Data Analytics, Machine Learning, Real Time Processing
42mHow mobile.de brings Data Science to Production...
21mCase Law and ML on Spark