Datasets:

Modalities:
Tabular
Text
Languages:
Chinese
ArXiv:
Libraries:
Datasets
License:
imryanxu commited on
Commit
27b13e1
·
1 Parent(s): 68169d5

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +56 -0
README.md CHANGED
@@ -1,3 +1,59 @@
1
  ---
2
  license: cc-by-4.0
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-4.0
3
  ---
4
+
5
+
6
+ # Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark
7
+
8
+
9
+ ## Contents
10
+
11
+ - [Overview](#overview)
12
+ - [Dataset](#dataset)
13
+ - [Citation](#citation)
14
+
15
+ ## Overview
16
+
17
+ Hansel is a high-quality human-annotated Chinese entity linking (EL) dataset, focusing on tail entities and emerging entities:
18
+ - The test set contains Few-shot (FS) and zero-shot (ZS) slices, has 10K examples and uses Wikidata as the corresponding knowledge base.
19
+ - The training and validation sets are from Wikipedia hyperlinks, useful for large-scale pretraining of Chinese EL systems.
20
+
21
+ Please see our [WSDM 2023](https://www.wsdm-conference.org/2023/) paper [**Hansel**](https://arxiv.org/abs/2207.13005) to learn more about our dataset.
22
+
23
+ ## Dataset
24
+
25
+ ### Data Statistics
26
+
27
+ | | # Mentions | # Entities | Domain |
28
+ | ---- | ---- | ---- | ---- |
29
+ | Train | 9,879,813 | 541,058 | Wikipedia |
30
+ | Validation | 9,674 | 6,320 | Wikipedia |
31
+ | Hansel-FS | 5,260 | 2,720 | News, Social Media |
32
+ | Hansel-ZS | 4,715 | 4,046 | News, Social Media, E-books, etc.|
33
+
34
+
35
+ ### Data Format
36
+
37
+ {"id": "hansel-eval-zs-1463",
38
+ "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。",
39
+ "start": 29,
40
+ "end": 32,
41
+ "mention": "匹诺曹",
42
+ "gold_id": "Q73895818",
43
+ "source": "https://www.1905.com/news/20181107/1325389.shtml",
44
+ "domain": "news"
45
+ }
46
+
47
+ ## Citation
48
+
49
+ If you use our dataset in your work, please cite us.
50
+
51
+ ```bibtex
52
+ @misc{xu2022hansel,
53
+ title = {Hansel: A Chinese Few-Shot and Zero-Shot Entity Linking Benchmark},
54
+ author = {Xu, Zhenran and Shan, Zifei and Li, Yuxin and Hu, Baotian and Qin, Bing},
55
+ publisher = {arXiv},
56
+ year = {2022},
57
+ url = {https://arxiv.org/abs/2207.13005}
58
+ }
59
+ ```