Reproduce “Supply-Side Equilibria in Recommender Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Meena Jagadeesan, Nikhil Garg, Jacob Steinhardt

The authors model producers choosing direction and magnitude, while hardmax recommendations match users to highest-scoring items. With norm-power production costs, changing exponent β can shift symmetric equilibria from one genre to multiple directions. For two orthogonal users, two producers, and β=10, an equilibrium yields 0.8 profit per producer; this model result does not establish real creators’ behavior or welfare effects.

Key insights

  • Specialization is governed by the geometry of user preferences and the curvature of production costs, rather than by the number of producers alone. Increasing β makes serving several dimensions simultaneously more expensive and can force differentiation.
  • For two user populations, the threshold β* depends on their angular separation: similar users are easier to serve with one product, while orthogonal users favor specialization at lower cost.
  • Above the specialization threshold, finite-genre equilibria may disappear and equilibrium support can become continuous or countably infinite, representing a spectrum of tradeoffs between audience groups.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.