Hansel

# Dataset Card for "Hansel"

## Table of Contents
- [Table of Contents](#table-of-contents)
- [Dataset Description](#dataset-description)
- [Dataset Structure](#dataset-structure)
- [Data Instances](#data-instances)
- [Data Splits](#data-splits)
- [Citation](#citation)

## Dataset Description

- **Homepage:** https://github.com/HITsz-TMG/Hansel
- **Paper:** https://arxiv.org/abs/2207.13005

Hansel is a high-quality human-annotated Chinese entity linking (EL) dataset, focusing on tail entities and emerging entities:
- The test set contains Few-shot (FS) and zero-shot (ZS) slices, has 10K examples and uses Wikidata as the corresponding knowledge base.
- The training and validation sets are from Wikipedia hyperlinks, useful for large-scale pretraining of Chinese EL systems.

Please see our [WSDM 2023](https://www.wsdm-conference.org/2023/) paper [**"Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark"**](https://dl.acm.org/doi/10.1145/3539597.3570418) to learn more about our dataset.

For models in the paper and our processed knowledge base, please see our [Github repository](https://github.com/HITsz-TMG/Hansel).

## Dataset Structure

### Data Instances

{"id": "hansel-eval-zs-1463",
"text": "1905电影网讯 已经筹备了十余年的吉尔莫·德尔·托罗的《匹诺曹》，在上个月顺利被网飞公司买下，成为了流媒体巨头旗下的新片。近日，这部备受关注的影片确定了自己的档期：2021年。虽然具体时间未定，但影片却已经实实在在地向前迈出了一步。",
"start": 29,
"end": 32,
"mention": "匹诺曹",
"gold_id": "Q73895818",
"source": "https://www.1905.com/news/20181107/1325389.shtml",
"domain": "news"
}

### Data Splits

| | # Mentions | # Entities | Domain |
| ---- | ---- | ---- | ---- |
| Train | 9,879,813 | 541,058 | Wikipedia |
| Validation | 9,674 | 6,320 | Wikipedia |
| Hansel-FS | 5,260 | 2,720 | News, Social Media |
| Hansel-ZS | 4,715 | 4,046 | News, Social Media, E-books, etc.|

## Citation

If you find our dataset useful, please cite us.

```bibtex
@inproceedings{xu2022hansel,
author = {Xu, Zhenran and Shan, Zifei and Li, Yuxin and Hu, Baotian and Qin, Bing},
title = {Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark},
year = {2023},
publisher = {Association for Computing Machinery},
url = {https://doi.org/10.1145/3539597.3570418},
booktitle = {Proceedings of the Sixteenth ACM International Conference on Web Search and Data Mining},
pages = {832–840}
}
```

# "Hansel"数据集卡片

## 目录
- [目录](#table-of-contents)
- [数据集概述](#dataset-description)
- [数据集结构](#dataset-structure)
- [数据实例](#data-instances)
- [数据划分](#data-splits)
- [引用格式](#citation)

## 数据集概述
- **项目主页**：https://github.com/HITsz-TMG/Hansel
- **相关论文**：https://arxiv.org/abs/2207.13005

Hansel是一个高质量人工标注的中文实体链接（Entity Linking，EL）数据集，聚焦于长尾实体与新兴实体：
- 测试集包含少样本（Few-shot，FS）与零样本（Zero-shot，ZS）子集，共计10000条样本，以维基数据（Wikidata）作为对应知识库。
- 训练集与验证集均取自维基百科（Wikipedia）超链接，可用于中文实体链接系统的大规模预训练。

如需了解该数据集的更多细节，请参阅我们发表于WSDM 2023的论文《Hansel：中文少样本与零样本实体链接基准数据集》，链接为：https://dl.acm.org/doi/10.1145/3539597.3570418。
如需获取论文中提及的模型与我们预处理后的知识库，请查阅我们的GitHub仓库：https://github.com/HITsz-TMG/Hansel。

## 数据集结构

### 数据实例
json
{"id": "hansel-eval-zs-1463",
"text": "1905电影网讯 已经筹备了十余年的吉尔莫·德尔·托罗的《匹诺曹》，在上个月顺利被网飞公司买下，成为了流媒体巨头旗下的新片。近日，这部备受关注的影片确定了自己的档期：2021年。虽然具体时间未定，但影片却已经实实在在地向前迈出了一步。",
"start": 29,
"end": 32,
"mention": "匹诺曹",
"gold_id": "Q73895818",
"source": "https://www.1905.com/news/20181107/1325389.shtml",
"domain": "news"
}

### 数据划分
| | 实体提及数 | 实体数 | 领域 |
| ---- | ---- | ---- | ---- |
| 训练集 | 9,879,813 | 541,058 | 维基百科 |
| 验证集 | 9,674 | 6,320 | 维基百科 |
| Hansel少样本子集 | 5,260 | 2,720 | 新闻、社交媒体 |
| Hansel零样本子集 | 4,715 | 4,046 | 新闻、社交媒体、电子书等 |

## 引用格式
如果您认为本数据集对您的研究有所帮助，请引用我们的工作。
bibtex
@inproceedings{xu2022hansel,
author = {Xu, Zhenran and Shan, Zifei and Li, Yuxin and Hu, Baotian and Qin, Bing},
title = {Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark},
year = {2023},
publisher = {Association for Computing Machinery},
url = {https://doi.org/10.1145/3539597.3570418},
booktitle = {Proceedings of the Sixteenth ACM International Conference on Web Search and Data Mining},
pages = {832–840}
}