Skip to content

Ranking algorithms

Ranking algorithms determine the order in which content appears in social media feeds, recommendation systems, and search results. They are the primary mechanism through which platforms decide which content each user sees and in what priority.

How ranking works

Ranking algorithms typically:

  1. Collect signals — User interaction history (likes, retweets, clicks, dwell time), user profile data, content features, social graph information
  2. Score content — Generate a relevance or importance score for each piece of content relative to each user
  3. Order — Rank content by score and display the top items in the user's feed or recommendations
  4. Optimize for a target metric — Different platforms optimize for different metrics (engagement time, click-through rate, user retention, satisfaction, diversity, etc.)

Design choices and tradeoffs

Ranking algorithms reflect explicit design choices about what constitutes "good" ranking:

  • Engagement optimization — Prioritize content that users interact with (likes, comments, shares). Tends to amplify emotionally charged, morally framed, and partisan content.
  • Diversity — Explicitly include diverse perspectives and sources. May reduce engagement but increase user satisfaction and reduce polarization.
  • User preferences — Rank based on explicit user preferences (stated preferences) rather than implicit user behavior (revealed preferences). Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media shows stated preference ranking reduces exposure to divisive content while maintaining engagement.
  • Chronological ranking — Simple reverse-chronological ordering; serves as a control baseline for measuring algorithmic effects.

Empirical evidence on effects

Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media directly compares engagement-based ranking to reverse-chronological ranking and finds that engagement-based ranking systematically amplifies partisan, emotionally charged, and hostile content beyond what users report wanting to see. An alternative stated-preference ranking reduces these negative effects.