Dataset for Concise Math Expression Images from Web Documents (URL-based)

This dataset is constructed for studies on the extraction of Concise Math Expressions (CMEs) from web documents. A CME is defined as an expression that is sufficiently informative for human understanding without requiring additional explanation, while being shorter than the expressions from which the relevant formula is derived. The dataset is reconstructed from web-based mathematical expression images, and it is used for training and evaluating classifiers that identify CMEs. To comply with copyright restrictions, this dataset does not include the image files themselves. However, it does provide URLs of the webpages on which the images appear. Each record in the dataset includes - image path (identifier) - binary label (1: CME, 0: non-CME) - data split (training/test) - original page title in Japanese - page title in English - original URL of the corresponding web page - finally accessed URL - link accessibility result (OK / NOT_FOUND / CLIENT_ERROR / SERVER_ERROR / ERROR) The dataset includes both training and test sets used in the paper’s experiments. The original webpages were collected between April and August 2016. URL accessibility information reflects the status observed in May 2026, and some URLs are currently inaccessible due to changes in web availability over time. This dataset is released to ensure reproducibility and support further research on mathematical expression retrieval and analysis.