Open shivashankarrs opened 3 years ago
We will use data augmentation solutions (back-translation, MLM based sampling and any other simple approach.)
@afshinrahimi can you take lead on this task? We can discuss based on the solutions I have read so far, and shortlist few techniques/implementations to use.
Backtranslation and Language models used: https://www.aclweb.org/anthology/2020.emnlp-main.97.pdf (class imbalance)
Studies several augmentation approaches for imbalance: https://www.aclweb.org/anthology/2020.findings-emnlp.269.pdf (class imbalance)
Stereotypical Bias Removal for Hate Speech Detection Task using Knowledge-based Generalizations, https://arxiv.org/pdf/2001.05495.pdf (Fairness)
Implement ways to create synthetic samples (SMOTE) from hidden representations.
Related to the CIKM 2020 paper.