jagoldz commited on
Commit
7ec4ddb
1 Parent(s): 879ec28

Update README

Browse files
Files changed (1) hide show
  1. README.md +53 -0
README.md CHANGED
@@ -1,3 +1,56 @@
1
  ---
2
  license: cc-by-4.0
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-4.0
3
+ task_categories:
4
+ - text-classification
5
+ language:
6
+ - de
7
+ tags:
8
+ - hate-speech-detection
9
+ - hate-speech
10
+ pretty_name: GAHD
11
+ size_categories:
12
+ - 10K<n<100K
13
  ---
14
+ # Dataset Card for GAHD
15
+
16
+ ## Dataset Description
17
+
18
+ GAHD is a **G**erman **A**dversarial **H**ate speech **D**ataset containing 10,996 examples. We collected the dataset via four rounds of Dynamic Adversarial Data Collection and explored various methods of supporting annotators in finding adversarial examples.
19
+
20
+ - **Paper:** https://arxiv.org/abs/2403.19559
21
+ - **Repository:** https://github.com/jagol/gahd
22
+
23
+ ## Dataset Structure
24
+
25
+ `gahd.csv` contains the following columns:
26
+ - `gahd_id`: unique identifier of the entry
27
+ - `text`: text of the entry
28
+ - `label`: `0` = "not-hate speech", `1` = "hate speech"
29
+ - `round`: round in which the entry was created
30
+ - `split`: "train", "dev", or "test"
31
+ - `contrastive_gahd_id`: `gahd_id` of its contrastive example
32
+
33
+ `gahd_disaggregated.csv` contains the following additional columns:
34
+ - `source`:
35
+ - if annotators entered the entry via the Dynabench interface: `dynabench`
36
+ - if the entry was translated from the Vidgen et al. 2021 dataset: `translation`
37
+ - if the entry stems from the Leipzit news corpus: `news`
38
+ - `model_prediction`: label predicted by the target model, `0` or `1`
39
+ - `annotator_id`: unique identifier of the annotator that created the entry
40
+ - `annotator_labels`: a string containing a forward slash-separated list of all labels by annotators
41
+ - `expert_labels`: `0` or `1` if an expert annotator annotated the entry, otherwise empty
42
+
43
+ ## Citation
44
+
45
+ When using GAHD, please cite our preprint on Arxiv:
46
+
47
+ ```
48
+ @misc{goldzycher2024improving,
49
+ title={Improving Adversarial Data Collection by Supporting Annotators: Lessons from GAHD, a German Hate Speech Dataset},
50
+ author={Janis Goldzycher and Paul Röttger and Gerold Schneider},
51
+ year={2024},
52
+ eprint={2403.19559},
53
+ archivePrefix={arXiv},
54
+ primaryClass={cs.CL}
55
+ }
56
+ ```