In order to better detect hidden hate comments or posts, the paper introduces sarcasm-based rationale (emoticons or text segments that indicate sarcasm) combined with hate/offensive rationale. HateXplain, an established benchmark hate dataset, was used to create a dataset by labeling texts and choosing justifications based on sarcasm. Then, the recently created dataset was used with the current cutting-edge model. When sarcasm rationale was combined with hate/offensive rationale in a newly formed attention proposed in the data’s preprocessing step, the model’s F1-score increased by 0.01. Furthermore, a noteworthy enhancement was noted in explainability metrics like faithfulness and plausibility with the new data.
https://www.mdpi.com/2076-3417/14/11/4898