Recommendation & Personalization

Six case studies on deciding what to show a user. The recommendation system is the canonical case: one problem that forces candidate generation, ranking, multi task objectives, cold start, position bias, and A/B testing. The others reuse the same skeleton. Feed ranking adds the engagement trap, conversational recommendation adds dialogue and constraints, graph recommendation adds candidate generation over a social graph, autocomplete adds query completion under tight latency, and notifications add bandits and a fatigue budget.

The shared intellectual core across all six is the feedback loop: the user only acts on what you showed, so your next training set is biased by your current model.

• Recommendation System
• News Feed / Personalized Ranking
• Conversational Recommender
• People You May Know / Graph Recommendation
• Autocomplete / Typeahead Personalization
• Notification Optimization / Bandit System