Data and code repository to evaluate multilingual fairness for hate speech detection for the LREC 2020 paper Multilingual Twitter corpus and baselines for evaluating demographic bias in hate speech recognition.
-
Install required packages;
- Install conda;
- Install pytorch;
- Final:
pip install -r requirements.txt.
-
Process pre-trained word embeddings;
- Follow the instructions.
-
Run python scripts
- Test logistic regression classifier:
python lr.py; - Test RNN classifier:
python rnn.py; - Test CNN classifier:
python cnn.py; - Test BERT classifier:
python bert.py.
- Test logistic regression classifier:
To request non-binary demographic labels or if you have any issues, please email xiaolei.huang@colorado.edu.
If you use our corpus in your publication, please kindly cite this paper):
@inproceedings{huang2020-lrec,
title = "Multilingual Twitter Corpus and Baselines for Evaluating Demographic Bias in Hate Speech Recognition",
author = "Huang, Xiaolei and
Linzi, Xing and
Dernoncourt, Franck and
Paul, Michael J.",
booktitle = "Proceedings of the Twelveth International Conference on Language Resources and Evaluation ({LREC} 2020)",
month = may,
year = "2020",
address = "Marseille, France",
publisher = "European Language Resources Association (ELRA)",
url = "https://arxiv.org/pdf/2002.10361.pdf",
abstract = "Existing research on fairness evaluation of document classification models mainly uses synthetic monolingual data without ground truth for author demographic attributes. In this work, we assemble and publish a multilingual Twitter corpus for the task of hate speech detection with inferred four author demographic factors: age, country, gender and race/ethnicity. The corpus covers five languages: English, Italian, Polish, Portuguese and Spanish. We evaluate the inferred demographic labels with a crowdsourcing platform, Figure Eight. To examine factors that can cause biases, we take an empirical analysis of demographic predictability on the English corpus. We measure the performance of four popular document classifiers and evaluate the fairness and bias of the baseline classifiers on the author-level demographic attributes.",
}