cornelius commited on
Commit
036c406
1 Parent(s): 0b23108

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +122 -19
README.md CHANGED
@@ -1,47 +1,150 @@
1
  ---
 
 
 
 
 
 
2
  tags:
3
- - generated_from_keras_callback
4
- model-index:
5
- - name: partypress-monolingual-austria
6
- results: []
 
 
7
  ---
8
 
9
- <!-- This model card has been generated automatically according to the information Keras had access to. You should
10
- probably proofread and complete it, then remove this comment. -->
11
 
12
- # partypress-monolingual-austria
13
 
14
- This model is a fine-tuned version of [cornelius/partypress-monolingual-austria](https://huggingface.co/cornelius/partypress-monolingual-austria) on an unknown dataset.
15
- It achieves the following results on the evaluation set:
16
 
17
 
18
  ## Model description
19
 
20
- More information needed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
 
22
  ## Intended uses & limitations
23
 
24
- More information needed
25
 
26
- ## Training and evaluation data
27
 
28
- More information needed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29
 
30
  ## Training procedure
31
 
32
- ### Training hyperparameters
33
 
34
- The following hyperparameters were used during training:
35
- - optimizer: None
36
- - training_precision: float32
37
 
38
- ### Training results
39
 
 
40
 
 
41
 
42
- ### Framework versions
 
 
 
 
 
 
 
43
 
44
  - Transformers 4.28.0
45
  - TensorFlow 2.12.0
46
  - Datasets 2.12.0
47
  - Tokenizers 0.13.3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-sa-4.0
3
+ language:
4
+ - de
5
+ metrics:
6
+ - accuracy
7
+ pipeline_tag: text-classification
8
  tags:
9
+ - partypress
10
+ - political science
11
+ - parties
12
+ - press releases
13
+ widget:
14
+ - text: 'Immissionsschutzgesetz muss ein Klagerecht für BürgerInnen beinhalten: "Es ist seit Jahren bekannt, welche Maßnahmen zur Reduktion der Feinstaubbelastung gesetzt werden müssen. Diese neuerlich bloß aufzuzählen, wie es jetzt Minister Berlakovich tut, hilft den Betroffenen nicht", kritisiert die Grüne Umweltsprecherin Christiane Brunner die jüngsten Aussagen des Umweltministers zur Problematik Feinstaub.'
15
  ---
16
 
17
+ # PARTYPRESS monolingual Austria
 
18
 
 
19
 
20
+ Fine-tuned model, based on [dbmdz/bert-base-german-cased](https://huggingface.co/dbmdz/bert-base-german-cased). Used in Erfort et al. (2023), building on the PARTYPRESS database. For the downstream task of classyfing press releases from political parties into 23 unique policy areas we achieve a performance comparable to expert human coders.
 
21
 
22
 
23
  ## Model description
24
 
25
+ The PARTYPRESS monolingual model builds on [dbmdz/bert-base-german-cased](https://huggingface.co/dbmdz/bert-base-german-cased) but has a supervised component. This means, it was fine-tuned using texts labeled by humans. The labels indicate 23 different political issue categories derived from the Comparative Agendas Project (CAP):
26
+ | Code | Issue |
27
+ |--|-------|
28
+ | 1 | Macroeconomics |
29
+ | 2 | Civil Rights |
30
+ | 3 | Health |
31
+ | 4 | Agriculture |
32
+ | 5 | Labor |
33
+ | 6 | Education |
34
+ | 7 | Environment |
35
+ | 8 | Energy |
36
+ | 9 | Immigration |
37
+ | 10 | Transportation |
38
+ | 12 | Law and Crime |
39
+ | 13 | Social Welfare |
40
+ | 14 | Housing |
41
+ | 15 | Domestic Commerce |
42
+ | 16 | Defense |
43
+ | 17 | Technology |
44
+ | 18 | Foreign Trade |
45
+ | 19.1 | International Affairs |
46
+ | 19.2 | European Union |
47
+ | 20 | Government Operations |
48
+ | 23 | Culture |
49
+ | 98 | Non-thematic |
50
+ | 99 | Other |
51
+
52
+ ## Model variations
53
+
54
+ There are several monolingual models for different countries, and a multilingual model. The multilingual model can be easily extended to other languages, country contexts, or time periods by fine-tuning it with minimal additional labeled texts.
55
 
56
  ## Intended uses & limitations
57
 
58
+ The main use of the model is for text classification of press releases from political parties. It may also be useful for other political texts.
59
 
60
+ The classification can then be used to measure which issues parties are discussing in their communication.
61
 
62
+ ### How to use
63
+
64
+ This model can be used directly with a pipeline for text classification:
65
+
66
+ ```python
67
+ >>> from transformers import pipeline
68
+ >>> tokenizer_kwargs = {'padding':True,'truncation':True,'max_length':512}
69
+ >>> partypress = pipeline("text-classification", model = "cornelius/partypress-monolingual-Austria", tokenizer = "cornelius/partypress-monolingual-Austria", **tokenizer_kwargs)
70
+ >>> partypress("Your text here.")
71
+ ```
72
+
73
+ ### Limitations and bias
74
+
75
+ The model was trained with data from parties in Austria. For use in other countries, the model may be further fine-tuned. Without further fine-tuning, the performance of the model may be lower.
76
+
77
+ The model may have biased predictions. We discuss some biases by country, party, and over time in the release paper for the PARTYPRESS database. For example, the performance is highest for press releases from Ireland (75%) and lowest for Poland (55%).
78
+
79
+ ## Training data
80
+
81
+ The PARTYPRESS multilingual model was fine-tuned with about 3,000 press releases from parties in Austria. The press releases were labeled by two expert human coders.
82
+
83
+ For the training data of the underlying model, please refer to [dbmdz/bert-base-german-cased](https://huggingface.co/dbmdz/bert-base-german-cased)
84
 
85
  ## Training procedure
86
 
87
+ ### Preprocessing
88
 
89
+ For the preprocessing, please refer to [dbmdz/bert-base-german-cased](https://huggingface.co/dbmdz/bert-base-german-cased)
 
 
90
 
91
+ ### Pretraining
92
 
93
+ For the pretraining, please refer to [dbmdz/bert-base-german-cased](https://huggingface.co/dbmdz/bert-base-german-cased)
94
 
95
+ ### Fine-tuning
96
 
97
+ We fine-tuned the model using about 3,000 labeled press releases from political parties in Austria.
98
+
99
+ #### Training Hyperparameters
100
+
101
+ The batch size for training was 12, for testing 2, with four epochs. All other hyperparameters were the standard from the transformers library.
102
+
103
+
104
+ #### Framework versions
105
 
106
  - Transformers 4.28.0
107
  - TensorFlow 2.12.0
108
  - Datasets 2.12.0
109
  - Tokenizers 0.13.3
110
+
111
+
112
+ ## Evaluation results
113
+
114
+ Fine-tuned on our downstream task, this model achieves the following results in a five-fold cross validation that are comparable to the performance of our expert human coders. Please refer to Erfort et al. (2023)
115
+
116
+ ### BibTeX entry and citation info
117
+
118
+ ```bibtex
119
+ @article{erfort_partypress_2023,
120
+ author = {Cornelius Erfort and
121
+ Lukas F. Stoetzer and
122
+ Heike Klüver},
123
+ title = {The PARTYPRESS Database: A New Comparative Database of Parties’ Press Releases},
124
+ journal = {Research and Politics},
125
+ volume = {forthcoming},
126
+ year = {2023},
127
+ }
128
+ ```
129
+
130
+ ### Further resources
131
+
132
+ Github: [cornelius-erfort/partypress](https://github.com/cornelius-erfort/partypress)
133
+
134
+ Research and Politics Dataverse: [Replication Data for: The PARTYPRESS Database: A New Comparative Database of Parties’ Press Releases](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2FOINX7Q)
135
+
136
+
137
+
138
+ ## Acknowledgements
139
+
140
+ Research for this contribution is part of the Cluster of Excellence "Contestations of the Liberal Script" (EXC 2055, Project-ID: 390715649), funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Austria´s Excellence Strategy. Cornelius Erfort is moreover grateful for generous funding provided by the DFG through the Research Training Group DYNAMICS (GRK 2458/1).
141
+
142
+ ## Contact
143
+
144
+ Cornelius Erfort
145
+
146
+ Humboldt-Universität zu Berlin
147
+
148
+ [corneliuserfort.de](corneliuserfort.de)
149
+
150
+