PeerCheck: Enhancing LLM-Generated Academic Reviews Towards Human-Level Quality

The PeerCheck framework has been introduced to enhance the quality of academic reviews generated by large language models (LLMs) by addressing the differences between LLM and human review focuses. Key techniques include prompt engineering with Chain-of-Thought (CoT) and retrieval-augmented generation (RAG), which significantly improve review quality, although RAG exhibits variability in effectiveness across different LLMs. This work is crucial for practitioners aiming to integrate LLMs into the peer review process, providing insights into optimizing model performance and aligning outputs with human standards.