From Passive Generation to Investigation: A Proactive Scientific Peer Review Agent

The paper introduces ProReviewer, a proactive scientific peer review agent that leverages an 8 billion parameter large language model (LLM) to enhance the peer review process by enabling in-depth investigations of suspicious content in research papers. By formulating the review process as a Markov Decision Process (MDP) and utilizing a structured review log to track evidence, ProReviewer outperforms larger prompt-based LLMs by up to 39% across five quality dimensions and achieves superior human evaluation win rates. This advancement is significant for AI practitioners as it provides a framework for developing more effective automated review systems that mimic human-like scrutiny and evidence-based assessments.