Update README.md
Browse files
README.md
CHANGED
@@ -17,48 +17,6 @@ widget:
|
|
17 |
|
18 |
Text classification model based on [`roberta-base`](https://huggingface.co/roberta-base) and fine-tuned on the [FRENK dataset](https://www.clarin.si/repository/xmlui/handle/11356/1433) comprising of LGBT and migrant hatespeech. Only the English subset of the data was used for fine-tuning and the dataset has been relabeled for binary classification (offensive or acceptable).
|
19 |
|
20 |
-
|
21 |
-
|
22 |
-
If you use the model, please cite the following paper on which the original model is based:
|
23 |
-
```
|
24 |
-
@article{DBLP:journals/corr/abs-1907-11692,
|
25 |
-
author = {Yinhan Liu and
|
26 |
-
Myle Ott and
|
27 |
-
Naman Goyal and
|
28 |
-
Jingfei Du and
|
29 |
-
Mandar Joshi and
|
30 |
-
Danqi Chen and
|
31 |
-
Omer Levy and
|
32 |
-
Mike Lewis and
|
33 |
-
Luke Zettlemoyer and
|
34 |
-
Veselin Stoyanov},
|
35 |
-
title = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},
|
36 |
-
journal = {CoRR},
|
37 |
-
volume = {abs/1907.11692},
|
38 |
-
year = {2019},
|
39 |
-
url = {http://arxiv.org/abs/1907.11692},
|
40 |
-
archivePrefix = {arXiv},
|
41 |
-
eprint = {1907.11692},
|
42 |
-
timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},
|
43 |
-
biburl = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},
|
44 |
-
bibsource = {dblp computer science bibliography, https://dblp.org}
|
45 |
-
}
|
46 |
-
```
|
47 |
-
|
48 |
-
and the dataset used for fine-tuning:
|
49 |
-
```
|
50 |
-
@misc{ljubešić2019frenk,
|
51 |
-
title={The FRENK Datasets of Socially Unacceptable Discourse in Slovene and English},
|
52 |
-
author={Nikola Ljubešić and Darja Fišer and Tomaž Erjavec},
|
53 |
-
year={2019},
|
54 |
-
eprint={1906.02045},
|
55 |
-
archivePrefix={arXiv},
|
56 |
-
primaryClass={cs.CL},
|
57 |
-
url={https://arxiv.org/abs/1906.02045}
|
58 |
-
}
|
59 |
-
```
|
60 |
-
|
61 |
-
|
62 |
## Fine-tuning hyperparameters
|
63 |
|
64 |
Fine-tuning was performed with `simpletransformers`. Beforehand a brief hyperparameter optimisation was performed and the presumed optimal hyperparameters are:
|
@@ -125,4 +83,48 @@ predictions, logit_output = model.predict(["Build the wall",
|
|
125 |
predictions
|
126 |
### Output:
|
127 |
### array([1, 0])
|
128 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
Text classification model based on [`roberta-base`](https://huggingface.co/roberta-base) and fine-tuned on the [FRENK dataset](https://www.clarin.si/repository/xmlui/handle/11356/1433) comprising of LGBT and migrant hatespeech. Only the English subset of the data was used for fine-tuning and the dataset has been relabeled for binary classification (offensive or acceptable).
|
19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
## Fine-tuning hyperparameters
|
21 |
|
22 |
Fine-tuning was performed with `simpletransformers`. Beforehand a brief hyperparameter optimisation was performed and the presumed optimal hyperparameters are:
|
83 |
predictions
|
84 |
### Output:
|
85 |
### array([1, 0])
|
86 |
+
```
|
87 |
+
|
88 |
+
## Citation
|
89 |
+
|
90 |
+
|
91 |
+
If you use the model, please cite the following paper on which the original model is based:
|
92 |
+
```
|
93 |
+
@article{DBLP:journals/corr/abs-1907-11692,
|
94 |
+
author = {Yinhan Liu and
|
95 |
+
Myle Ott and
|
96 |
+
Naman Goyal and
|
97 |
+
Jingfei Du and
|
98 |
+
Mandar Joshi and
|
99 |
+
Danqi Chen and
|
100 |
+
Omer Levy and
|
101 |
+
Mike Lewis and
|
102 |
+
Luke Zettlemoyer and
|
103 |
+
Veselin Stoyanov},
|
104 |
+
title = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},
|
105 |
+
journal = {CoRR},
|
106 |
+
volume = {abs/1907.11692},
|
107 |
+
year = {2019},
|
108 |
+
url = {http://arxiv.org/abs/1907.11692},
|
109 |
+
archivePrefix = {arXiv},
|
110 |
+
eprint = {1907.11692},
|
111 |
+
timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},
|
112 |
+
biburl = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},
|
113 |
+
bibsource = {dblp computer science bibliography, https://dblp.org}
|
114 |
+
}
|
115 |
+
```
|
116 |
+
|
117 |
+
and the dataset used for fine-tuning:
|
118 |
+
```
|
119 |
+
@misc{ljubešić2019frenk,
|
120 |
+
title={The FRENK Datasets of Socially Unacceptable Discourse in Slovene and English},
|
121 |
+
author={Nikola Ljubešić and Darja Fišer and Tomaž Erjavec},
|
122 |
+
year={2019},
|
123 |
+
eprint={1906.02045},
|
124 |
+
archivePrefix={arXiv},
|
125 |
+
primaryClass={cs.CL},
|
126 |
+
url={https://arxiv.org/abs/1906.02045}
|
127 |
+
}
|
128 |
+
```
|
129 |
+
|
130 |
+
|