Design: News Feed

// Fan-out on write (push model)
// User posts → push to all followers' feed caches
//
// Alice (1000 followers) posts "Hello World"
// → Write to 1000 feed caches (one per follower)
// → Bob opens feed → read from pre-computed cache (fast!)
//
// Problem: Celebrity with 10M followers → 10M writes per post! 😱

// Fan-out on read (pull model)
// User opens feed → fetch posts from all followed users → merge + rank
//
// Bob follows 500 users → opens feed
// → Query 500 users' recent posts → merge → rank → return
// → Slow if following many users

// Hybrid (Twitter/X approach)
// Regular users (< 10K followers): fan-out on write
// Celebrities (> 10K followers): fan-out on read
//
// Bob's feed = pre-computed cache (regular follows)
//            + real-time merge (celebrity follows)

// Feed cache (Redis sorted set per user)
// Key: feed:bob
// Members: post_ids, scored by timestamp
ZADD feed:bob 1711700000 "post:123"
ZADD feed:bob 1711699000 "post:456"
ZREVRANGE feed:bob 0 19  // top 20 posts, newest first

// Feed ranking signals
// 1. Recency: newer posts score higher (decay function)
// 2. Engagement: posts with more likes/comments score higher
// 3. Relationship: posts from close friends score higher
// 4. Content type: video > image > text (per-user preference)
// 5. Diversity: avoid showing 5 posts from same user in a row

// Simple ranking formula
score = recency_weight * time_decay(post.created_at)
      + engagement_weight * log(post.likes + post.comments + 1)
      + affinity_weight * user_affinity(viewer, poster)
      + content_weight * content_type_boost(post.type)

// ML-based ranking (production systems)
// Training data: user interactions (like, comment, share, hide, time spent)
// Model predicts: P(user engages with post)
// Posts sorted by predicted engagement probability

// Pagination: cursor-based
// First page: top 20 by score
// Next page: cursor = last post's score → fetch next 20 below that score