In modern social platforms, which factors are comprehensively evaluated by the content display ranking algorithms, and how do they impact users' visibility and engagement rates?

02 Ağu 10:39

On social platforms' ranking models, I often compare them to e-commerce recommendation systems. The core metrics for e-commerce are "conversion rate" and "purchase probability," so they assign high weights to user behaviors like clicks, adding to cart, and purchase history. Social platforms, however, focus more on "dwell time" and "interaction depth" (likes, comments, shares), so their models incorporate content freshness, trending topics, and strong connections in the social graph. Both use deep learning embeddings to capture user interests, but social platforms add an extra "diversity regularization"—by including an exposure balance term in the ranking objective function, they force a portion of low-exposure content into the top N%, curbing the formation of information cocoons. If I had to prioritize metrics, I’d first ensure steady growth in "basic interaction rate" (CTR + interaction rate), then use a "content diversity index" (based on topic distribution entropy) as a constraint. At the exposure level, I’d introduce "tiered exposure" (e.g., fixed exposure quotas for new users, potential interest users, and niche creators). This balances immediate user satisfaction with high-quality content while giving niche creators visibility, preventing algorithmic bias from creating echo chambers. For implementation, we could take inspiration from YouTube’s "recommendation rounds" mechanism—before each ranking, a batch of "exploration streams" is sampled and merged with the main stream. This dual-track strategy maintains personalization while effectively boosting overall engagement and platform content ecosystem health.