This dataset contains over 22,000 peer reviews from ICLR and NeurIPS spanning three authorship classes: human-written, fully AI-generated, and LLM-refined (human reviews polished by an LLM).
It is the primary benchmark for training and evaluating Sem-Detect, an AI-Text Detection approach that combines textual features with claim-level semantic analysis, tailored for the peer-review domain.
Paper 📄… See the full description on the dataset page:
https://huggingface.co/datasets/Sem-Detect/ML_Conferences-Peer-Reviews.
