astronolan/galaxy-mentions

该数据集名为Galaxy Mentions，是一个从arXiv天体物理学论文中提取的星系提及数据集，涵盖了1995年9月至2024年5月期间几乎所有在论文中被讨论超过两句话的星系。数据集包含四个主要部分：galaxy_mentions（每个接受的星系提及对应一行，包括论文ID、原始星系名称、别名、摘要和证据引用计数）、evidence_quotes（每个支持性引用对应一行，通过mention_id与星系提及关联）、papers（每个处理的arXiv论文对应一行，包含接受/拒绝/错误状态）和batches（每个提取批次对应一行，用于运行审计）。数据来源基于Smith42/minty-astro-ph，这是一个专注于天文学的子集，源自mlfoundations/MINT-1T-ArXiv，并通过arXiv元数据过滤。数据集通过GPT-5.5子代理自动化提取生成，但未经过跨论文或跨名称的去重处理，因此同一物理星系可能出现在多行中，且别名（如M87、NGC 4486）可能未合并。数据集还包括历史Gemini标签（gemini_decision列），用于标识论文是否满足讨论非银河系星系超过两句话的标准。该数据集主要用于研究，包含提取错误、别名遗漏或不完整坐标信息等局限性。

The dataset is named Galaxy Mentions and contains a collection of nearly every galaxy discussed in more than two sentences on arXiv astro-ph papers. It includes four main components: galaxy_mentions (one row per accepted galaxy mention, with paper ID, raw galaxy name, aliases, summary, and evidence quote count), evidence_quotes (one row per supporting quote, linked to mentions via mention_id), papers (one row per processed arXiv paper, including acceptance/rejection/error status), and batches (one row per extraction batch for run auditability). The data is sourced from Smith42/minty-astro-ph, an astronomy-focused subset of mlfoundations/MINT-1T-ArXiv, filtered using arXiv metadata. It was produced via automated extraction using GPT-5.5 subagents, but is raw and pre-deduplication, meaning the same physical galaxy may appear in multiple rows, and aliases are not merged. The dataset also includes historical Gemini labels (gemini_decision column) to indicate whether papers met the criterion of discussing non-Milky-Way galaxies for more than two sentences. It is intended for research use, with limitations such as extraction errors and incomplete coordinate information.