laugustyniak asawczyn commited on
Commit
6835dd2
1 Parent(s): 762030a

Update README.md (#3)

Browse files

- Update README.md (6a5de46eedb8b9c374d23af9edb5811131df0477)


Co-authored-by: Albert Sawczyn <asawczyn@users.noreply.huggingface.co>

Files changed (1) hide show
  1. README.md +52 -3
README.md CHANGED
@@ -3,9 +3,9 @@ annotations_creators:
3
  - hired_annotators
4
  language_creators:
5
  - found
6
- languages:
7
  - pl
8
- licenses:
9
  - other
10
  multilinguality:
11
  - monolingual
@@ -19,7 +19,7 @@ task_ids:
19
  pretty_name: Polish-Political-Advertising
20
  ---
21
 
22
- # Dataset Card for "Polish-Political-Advertising"
23
 
24
  ## Info
25
 
@@ -27,10 +27,59 @@ Political campaigns are full of political ads posted by candidates on social med
27
 
28
  > We achieved a 0.65 inter-annotator agreement (Cohen's kappa score). An additional annotator resolved the mismatches between the first two annotators improving the consistency and complexity of the annotation process.
29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  ## License
31
 
32
  [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
33
 
 
 
 
 
 
 
34
  ## Citing
35
 
36
  > ACL WiNLP 2020 Paper
 
3
  - hired_annotators
4
  language_creators:
5
  - found
6
+ language:
7
  - pl
8
+ license:
9
  - other
10
  multilinguality:
11
  - monolingual
 
19
  pretty_name: Polish-Political-Advertising
20
  ---
21
 
22
+ # Polish-Political-Advertising
23
 
24
  ## Info
25
 
 
27
 
28
  > We achieved a 0.65 inter-annotator agreement (Cohen's kappa score). An additional annotator resolved the mismatches between the first two annotators improving the consistency and complexity of the annotation process.
29
 
30
+ ## Tasks (input, output and metrics)
31
+
32
+ Political Advertising Detection
33
+
34
+ **Input** ('*tokens'* column): sequence of tokens
35
+
36
+ **Output** ('tags*'* column): sequence of tags
37
+
38
+ **Domain**: politics
39
+
40
+ **Measurements**: F1-Score (seqeval)
41
+
42
+ **Example:**
43
+
44
+ Input: `['@k_mizera', '@rdrozd', 'Problemem', 'jest', 'mała', 'produkcja', 'dlatego', 'takie', 'ceny', '.', '10', '000', 'mikrofirm', 'zamknęło', 'się', 'w', 'poprzednim', 'tygodniu', 'w', 'obawie', 'przed', 'ZUS', 'a', 'wystarczyło', 'zlecić', 'tym', 'co', 'chcą', 'np', '.', 'szycie', 'masek', 'czy', 'drukowanie', 'przyłbic', 'to', 'nie', 'wymaga', 'super', 'sprzętu', ',', 'umiejętności', '.', 'nie', 'będzie', 'pit', ',', 'vat', 'i', 'zus', 'będą', 'bezrobotni']`
45
+
46
+ Input (translated by DeepL): `@k_mizera @rdrozd The problem is small production that's why such prices . 10,000 micro businesses closed down last week for fear of ZUS and all they had to do was outsource to those who want e.g . sewing masks or printing visors it doesn't require super equipment , skills . there will be no pit , vat and zus will be unemployed`
47
+
48
+ Output: `['O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'B-WELFARE', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'B-WELFARE', 'O', 'B-WELFARE', 'O', 'B-WELFARE', 'O', 'B-WELFARE']`
49
+
50
+
51
+ ## Data splits
52
+
53
+ | Subset | Cardinality |
54
+ |:-----------|--------------:|
55
+ | train | 1020 |
56
+ | test | 341 |
57
+ | validation | 340 |
58
+
59
+ ## Class distribution
60
+
61
+ | Class | train | validation | test |
62
+ |:--------------------------------|--------:|-------------:|-------:|
63
+ | B-HEALHCARE | 0.237 | 0.226 | 0.233 |
64
+ | B-WELFARE | 0.210 | 0.232 | 0.183 |
65
+ | B-SOCIETY | 0.156 | 0.153 | 0.149 |
66
+ | B-POLITICAL_AND_LEGAL_SYSTEM | 0.137 | 0.143 | 0.149 |
67
+ | B-INFRASTRUCTURE_AND_ENVIROMENT | 0.110 | 0.104 | 0.133 |
68
+ | B-EDUCATION | 0.062 | 0.060 | 0.080 |
69
+ | B-FOREIGN_POLICY | 0.040 | 0.039 | 0.028 |
70
+ | B-IMMIGRATION | 0.028 | 0.017 | 0.018 |
71
+ | B-DEFENSE_AND_SECURITY | 0.020 | 0.025 | 0.028 |
72
+
73
  ## License
74
 
75
  [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
76
 
77
+ ## Links
78
+
79
+ [HuggingFace](https://huggingface.co/datasets/laugustyniak/political-advertising-pl)
80
+
81
+ [Paper](https://aclanthology.org/2020.winlp-1.28/)
82
+
83
  ## Citing
84
 
85
  > ACL WiNLP 2020 Paper