aquiro1994/naics-gh

NAICS-GH 是一个多区域、GPT-4.1标记的GitHub仓库语料库，每个仓库用北美行业分类系统（NAICS 2022）的两位数行业部门代码进行标记。该数据集包含6,588个来自美国、欧盟和澳大利亚的公共GitHub仓库，保留了19个两位数NAICS代码（代码55“企业管理”因样本不足被排除）。标签通过两阶段的“检索-验证”流程生成：首先从GitHub数据仓库检索候选仓库，然后使用BAAI/bge-large-en-v1.5嵌入和FAISS索引进行语义相似性检索，最后通过GPT-4.1评分（评分≥8的仓库被保留）。手动验证显示美国部分的标签精度为96.98%。数据集用于文本分类任务，特征包括仓库名称、描述、主题、README内容、整数标签和NAICS代码。数据集按行业分层随机分割为训练集、验证集和测试集。NAICS-GH is a multi-region, GPT-4.1-labeled corpus of GitHub repositories tagged with 2-digit NAICS industry sectors. It contains 6,588 public GitHub repositories drawn from the USA, the European Union, and Australia, each labeled with one of the 19 retained 2-digit codes from the North American Industry Classification System (NAICS 2022). Sector 55 ("Management of Companies and Enterprises") is absent due to insufficient samples. Labels were produced via a two-stage retrieve-and-verify pipeline: initial retrieval from GitHubs data warehouse, semantic similarity retrieval using BAAI/bge-large-en-v1.5 embeddings and FAISS indexing, and scoring by GPT-4.1 (only repositories scoring at least 8 are kept). Manual verification shows 96.98% label precision on the USA portion. The dataset is for text-classification tasks, with features including repository name, description, topics, README content, integer label, and NAICS code. It is split into train, validation, and test sets with stratification by NAICS sector.