splink-dataset
收藏资源简介:
Splink Business ER数据集是一个专为商业实体解析任务设计的合成数据集,旨在支持实体链接或记录匹配管道的开发与评估。数据内容主要包括两部分:一是来自多个业务来源的原始数据表,具体包括规范商业实体表、商业注册表、税务注册表、许可证记录表和检查记录表,这些表格模拟了真实世界中可能涉及企业信息的不同数据源;二是用于模型评估的辅助表,包括真实实体映射表和评估标签表。数据集以CSV表格形式组织,不包含预先生成的链接视图,鼓励使用者基于提供的四个源表构建自己的匹配流程。该数据集适用于实体解析、记录链接、数据清洗和商业智能等应用场景的研究与实验。
The Splink Business ER dataset is a synthetic dataset specifically designed for business entity resolution tasks, aimed at supporting the development and evaluation of entity linking or record matching pipelines. The data content mainly includes two parts: first, raw data tables from multiple business sources, specifically including standardized business entity tables, business registration tables, tax registration tables, license record tables, and inspection record tables, which simulate different data sources that may involve enterprise information in the real world; second, auxiliary tables for model evaluation, including true entity mapping tables and evaluation label tables. The dataset is organized in CSV table format, without pre-generated link views, encouraging users to build their own matching processes based on the four provided source tables. This dataset is suitable for research and experimentation in application scenarios such as entity resolution, record linking, data cleaning, and business intelligence.
数据集概述
数据集名称:Splink Business ER Dataset
数据集地址:https://huggingface.co/datasets/trainMd/splink-dataset
数据集类型:合成商业实体解析数据集。
数据集配置与文件结构
该数据集包含7个配置(config),每个配置对应一个CSV文件,全部使用train分割:
- canonical_business_entity:
canonical/canonical_business_entity.csv(标准商业实体) - business_registry:
observed/business_registry.csv(商业注册记录) - tax_registry:
observed/tax_registry.csv(税务注册记录) - license_record:
observed/license_record.csv(许可证记录) - inspection_record:
observed/inspection_record.csv(检查记录) - ground_truth_entity_map:
evaluation/ground_truth_entity_map.csv(真实实体映射) - evaluation_labels:
evaluation/evaluation_labels.csv(评估标签)
关键说明
- 该数据集仅包含原始源表(四个源表)和评估表(两个评估表)。
- 不包含
linkage_view(链接视图);匹配/流水线笔记本应从四个源表创建linkage_view。 - 所有文件路径为相对路径,对应数据集仓库内的实际目录结构。




