Skip to content

Recommendation algorithms and content discovery

Platform algorithms that rank, recommend, and surface content to users. Encompasses search ranking, feed algorithms, "suggested next" recommendations, and related discovery mechanisms. Central to platform dynamics because algorithm design (relevance, engagement, diversity) determines which creators reach which audiences.

Key papers

  • Horta Ribeiro et al. (2019) — audit of YouTube's recommendation graph across 349 channels; constructs recommendation network and performs random walks to measure reachability of Alt-right channels from milder communities; finds ~50% probability of reaching Alt-right from Alt-lite within 2 steps, suggesting algorithmic pathways to radicalization.
  • Hosseinmardi et al. (2022) — audits YouTube's recommendation engine via 4-year tracking of 309K users; finds 55% of far-right video views arrive from external URLs/search/homepage rather than recommendations; within-session viewing shows no bias toward extremism even in long sessions (where recommender has more influence); on/off-platform consumption patterns align, suggesting user preference over algorithmic steering.
  • Munger & Phillips (2022) — challenges "algorithm amplification" as sole explanation for far-right growth; shows viewership peaked before YouTube's 2019 algorithm changes and reflects supply-side affordances (low creation barriers, monetization) and demand-side audience appetite. Demonstrates algorithm is one of many platform design factors.
  • Cinelli et al. (2021) — compares platform architectures: algorithmic feeds (Facebook, Twitter) produce stronger echo chambers than community-based feeds (Reddit), suggesting algorithm design (engagement-optimized ranking) drives segregation more than user behavior alone.