This contains the ReclaimedQueerLex dataset by Rebecca Dorn et al. [1] and a Python script, which preprocesses the dataset and evaluates it using a hate speech detection model, fine-tuned on LGBTQ+ reclaimed language called the RHS model [2].
In the end the code for a LateX table is outputed with the Precision, Recall and F1-Score of hate and non-hate labels, given by Detoxify, Perspective and the RHS model.
[1] Rebecca Dorn, et al. (2024). Harmful Speech Detection by Language Models Exhibits Gender-Queer Dialect Bias. In Proceedings of the 4th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO '24). Association for Computing Machinery, New York, NY, USA, 1–12. doi.org/10.1145/3689904.3694704.
[2] Eszter Zsisku, et al. (2024). Hate Speech Detection and Reclaimed Language: Mitigating False Positives and Compounded Discrimination. In Proceedings of the 16th ACM Web Science Conference, 241–249, doi.org/10.1145/3614419.3644025.