New auction model for social feeds lifts producer welfare 36%

The recommendation algorithms that power social media feeds are caught in a tug-of-war: producers optimize for engagement, platforms maximize time-on-site, and consumers are exposed to misinformation and polarization. Andy Lee and Hari Sundaram propose a radically different approach in a new arXiv paper: treat attention as a market, and let users bid for it. Their auction-based feed construction systematically balances the interests of content producers, consumers, platform operators, and social welfare. The mechanism is shown to be weakly incentive compatible under budget constraints, meaning no user can gain by misreporting their true valuation of another user's attention.

To counter negative externalities such as polarization and harmful content, the authors introduce a tax policy that raises the cost of bids on such content, discouraging its spread while preserving legitimate expression. Simulations on synthetic networks and an empirically observed network show that the auction mechanism yields 36.3% higher producer welfare on the empirical network and 31.4% higher on synthetic topologies, compared to baseline feed algorithms. It also produces more equitable distributions of attention across all network types tested. Rather than replacing recommendation systems outright, the paper proposes addressing attention allocation at a systematic level, a step toward feeds that align with societal goals rather than just engagement metrics.