AgentReview

Peer review is central to scientific publishing, but its dynamics are hard to study because the data is sensitive and shaped by many latent factors. AgentReview is the first LLM-based peer-review simulation framework, disentangling the influence of reviewers, authors, and area chairs while sidestepping privacy concerns. The simulations reveal sizable effects — including a 37.1% variation in paper decisions attributable to reviewer biases — interpreted through sociological theories such as social influence and authority bias.

My role. I contributed to this collaboration as a co-author. AgentReview was published at EMNLP 2024 (Main Track, Oral).