EasyDeepLearn

How is a recommender system a bandit / RL problem?

medium

Answer

  • Contextual bandit: user context = state, item shown = action, click/watch = reward.
  • Long-term reward → full RL for session-level or LTV optimization.
  • Challenges: (1) partial feedback (only chosen item observed).
  • (2) Non-stationarity (user preferences drift).
  • (3) Delayed rewards (retention).
  • (4) Off-policy evaluation critical (can't A/B every candidate).
  • Modern: Thompson sampling for exploration, contextual bandits for retrieval, RL for long-horizon (Netflix, YouTube, TikTok).
Check yourself — multiple choice
  • Random
  • Contextual bandit: user context = state, item = action, click = reward; full RL for long-horizon LTV; challenges: partial feedback + drift + delayed + OPE; Thompson / RL at Netflix / YouTube
  • Same as CF
  • Not real

Recsys as bandit/RL: user=state, item=action, click=reward; Thompson + off-policy eval.

#applications

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions