a7c4163b33286261b24c72fd3d1707c9-Supplemental-Datasets_and_Benchmarks.pdf

These datasets enable large-scale study of abuse detection for these languages. Anonymized comments: To further address privacy concerns, we anonymize our dataset. We combine thehate and offensivecategories in these datasets for training a binary classification model. We showthepercentage (%)ofemoticons present inourdatasetMACDinTable12. Infuture work,we will investigate in detail about the impact of emoticons on abuse detection. However,duetothe limited scale and diversity of abuse detection datasets in Indic languages, development of these models for Indic languages has been severely impeded.