Recommendation System Model Selection

Design and justify the model choices for TikTok’s For-You-Page short-video recommendation system. You must pick (and can combine) classical and/or deep-learning approaches for the two core stages: (1) candidate generation (≈10K videos) and (2) ranking (≈500→50 videos shown). Explain which algorithm you would use in each stage, what features and training data are required, how you handle the cold-start problem for both new users and new videos, and how your choices satisfy the production constraints of <100ms p95 latency for 1B+ daily-active users while optimizing for multiple objectives (watch-time, like, share, follow, “not-interested”). Provide a concise comparison of at least two alternative model families and give the trade-offs (accuracy, latency, data volume, compute cost) that led to your final decision.