Ranking Multi-Source Push Notifications: Eligibility, De-Duplication, Fatigue and Send Decisions
Problem statement
How would you coordinate push notifications across a large social app? Several teams generate candidates independently, causing spam and duplicate messages while more users disable push. A shared system must quickly choose whether to send, hold or drop each candidate and order queued messages to support long-term engagement. Explain the objective, training data, models, serving and monitoring.
Show hint
What does the system ultimately want, and what can go wrong for a user who receives a push they open but regret?
Relevant roles: ML Engineer and Applied Scientist.