What Is Novel? A Knowledge-Driven Framework for Bias-Aware Literature Originality Evaluation

Researchers have developed an AI system trained on 80,000 peer reviews from top AI conferences to assess the novelty of scientific papers. It extracts a paper's core ideas and compares them to prior work in a structured way, producing calibrated novelty scores and human-like explanations. The system aims to reduce bias and inconsistency, offering a more evidence-based alternative to the often subjective judgments made by human reviewers.