We introduce a comprehensive approach that improves OCSR evaluation, training data, and model reliability.
Key Contributions:
Better Metrics: We introduce a comprehensive evaluation scheme that directly quantifies manual correction costs, with Perfectly-Matched Annotation Accuracy (PMAA) as the most stringent metric, ensuring exact alignment between model outputs and input… See the full description on the dataset page:
https://huggingface.co/datasets/anonymousttt2025/MolA.
