Yakread's Personalized Recommendation Algorithm Overhaul
2025-04-12
Yakread has rewritten its core recommendation algorithm that merges user subscriptions and bookmarked articles into a single personalized feed. The algorithm first sorts bookmarked articles by interaction (skips and bookmark time), applies a slight randomization to avoid monotony, and limits recommendations per website. For subscriptions, it calculates an "affinity score" based on the user's ten most recent interactions (views, skips, likes/dislikes) with each source. Pinned subscriptions are prioritized. Finally, it interleaves subscription and bookmark items using weighted random choice, balancing diversity and user preferences based on previous skips.