Research & Papers

PeerBTS mechanism rewards effort in strategyproof peer selection

New algorithm incentivizes accurate evaluations in MOOCs and peer review.

Deep Dive

PeerBTS is a mechanism that combines a peer-prediction lottery, based on the Robust Bayesian Truth Serum, with existing peer-selection mechanisms to incentivize effort while remaining Bayes-Nash incentive compatible. The authors prove that incentivizing effort requires information beyond a single evaluation and note that achieving this requires adjustments to the problem context and limits other properties. The paper includes non-strategic simulations to evaluate performance and an initial study on the validity of peer-prediction from a small academic workshop.

Key Points
  • PeerBTS combines a peer-prediction lottery (Robust Bayesian Truth Serum) with existing peer-selection mechanisms to reward evaluation effort.
  • Theoretically proven that incentivizing effort requires information beyond a single evaluation—PeerBTS uses predictions to achieve this.
  • Non-strategic simulations and a small workshop study validate the mechanism's ability to incentivize high-quality evaluations without breaking strategyproofness.

Why It Matters

Improves fairness and accuracy in MOOCs and peer review by motivating quality evaluations.