File size: 4,346 Bytes
433993b f840887 433993b 69dd5d1 f335e4f 69dd5d1 9f52149 69dd5d1 c709e1a 69dd5d1 507d26b 69dd5d1 507d26b 69dd5d1 507d26b 5d65110 528c17b 5d65110 610b928 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 |
---
language: "en"
license: "cc-by-sa-4.0"
tags:
- text-classification
- hate-speech
widget:
- text: "Gay is okay."
---
# roberta-base-frenk-hate
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).
## Fine-tuning hyperparameters
Fine-tuning was performed with `simpletransformers`. Beforehand a brief hyperparameter optimisation was performed and the presumed optimal hyperparameters are:
```python
model_args = {
"num_train_epochs": 6,
"learning_rate": 3e-6,
"train_batch_size": 69}
```
## Performance
The same pipeline was run with two other transformer models and `fasttext` for comparison. Accuracy and macro F1 score were recorded for each of the 6 fine-tuning sessions and post festum analyzed.
| model | average accuracy | average macro F1|
|---|---|---|
|roberta-base-frenk-hate|0.7915|0.7785|
|xlm-roberta-large |0.7904|0.77876|
|xlm-roberta-base |0.7577|0.7402|
|fasttext|0.725 |0.707 |
From recorded accuracies and macro F1 scores p-values were also calculated:
Comparison with `xlm-roberta-base`:
| test | accuracy p-value | macro F1 p-value|
| --- | --- | --- |
|Wilcoxon|0.00781|0.00781|
|Mann Whithney U-test|0.00108|0.00108|
|Student t-test | 1.35e-08 | 1.05e-07|
Comparison with `xlm-roberta-large` yielded inconclusive results. `roberta-base` has average accuracy 0.7915, while `xlm-roberta-large` has average accuracy of 0.7904. If macro F1 scores were to be compared, `roberta-base` actually has lower average than `xlm-roberta-large`: 0.77852 vs 0.77876 respectively. The same statistical tests were performed with the premise that `roberta-base` has greater metrics, and the results are given below.
| test | accuracy p-value | macro F1 p-value|
| --- | --- | --- |
|Wilcoxon|0.188|0.406|
|Mann Whithey|0.375|0.649|
|Student t-test | 0.681| 0.934|
With reversed premise (i.e., that `xlm-roberta-large` has greater statistics) the Wilcoxon p-value for macro F1 scores for this case reaches 0.656, Mann-Whithey p-value is 0.399, and of course the Student p-value stays the same. It was therefore concluded that performance of the two models are not statistically significantly different from one another.
## Use examples
```python
from simpletransformers.classification import ClassificationModel
model_args = {
"num_train_epochs": 6,
"learning_rate": 3e-6,
"train_batch_size": 69}
model = ClassificationModel(
"roberta", "5roop/roberta-base-frenk-hate", use_cuda=True,
args=model_args
)
predictions, logit_output = model.predict(["Build the wall",
"Build the wall of trust"]
)
predictions
### Output:
### array([1, 0])
```
## Citation
If you use the model, please cite the following paper on which the original model is based:
```
@article{DBLP:journals/corr/abs-1907-11692,
author = {Yinhan Liu and
Myle Ott and
Naman Goyal and
Jingfei Du and
Mandar Joshi and
Danqi Chen and
Omer Levy and
Mike Lewis and
Luke Zettlemoyer and
Veselin Stoyanov},
title = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},
journal = {CoRR},
volume = {abs/1907.11692},
year = {2019},
url = {http://arxiv.org/abs/1907.11692},
archivePrefix = {arXiv},
eprint = {1907.11692},
timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
and the dataset used for fine-tuning:
```
@misc{ljubešić2019frenk,
title={The FRENK Datasets of Socially Unacceptable Discourse in Slovene and English},
author={Nikola Ljubešić and Darja Fišer and Tomaž Erjavec},
year={2019},
eprint={1906.02045},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/1906.02045}
}
```
|