Estimating long-term outcome of algorithms - Spotify Research

The post discusses the challenges of estimating long-term outcomes of algorithms and introduces a statistical approach called Long-term Off-Policy Evaluation (LOPE) as a solution. LOPE combines short-term rewards and historical data to estimate the long-term outcome, providing more accurate results compared to other methods. It does not rely on the restrictive surrogacy assumption and can optimize long-term outcomes.