README.md

---
annotations_creators:
- crowdsourced
- found
language:
- zh
language_creators:
- found
- crowdsourced
license:
- cc-by-sa-4.0
multilinguality:
- monolingual
paperswithcode_id: hansel
pretty_name: Hansel
size_categories:
- 1M<n<10M
- 1K<n<10K
source_datasets:
- original
tags: []
task_categories:
- text-retrieval
task_ids:
- entity-linking-retrieval
dataset_info:
  - config_name: wiki
    features:
      - name: id
        dtype: string
      - name: text
        dtype: string
      - name: start
        dtype: int64
      - name: end
        dtype: int64
      - name: mention
        dtype: string
      - name: gold_id
        dtype: string
    splits:
      - name: train
      - name: validation
  - config_name: hansel
    features:
      - name: id
        dtype: string
      - name: text
        dtype: string
      - name: start
        dtype: int64
      - name: end
        dtype: int64
      - name: mention
        dtype: string
      - name: gold_id
        dtype: string
      - name: source
        dtype: string
      - name: domain
        dtype: string
    splits:
      - name: test-few-shot
      - name: test-zero-shot
---


# 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://arxiv.org/abs/2207.13005) to learn more about our dataset.

## 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
@misc{xu2022hansel,
  title = {Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark},
  author = {Xu, Zhenran and Shan, Zifei and Li, Yuxin and Hu, Baotian and Qin, Bing},
  publisher = {arXiv},
  year = {2022},
  url = {https://arxiv.org/abs/2207.13005}
}
```