Augmenting Data for Sarcasm Detection with Unlabeled Conversation Context
share
We present a novel data augmentation technique, CRA (Contextual Response Augmentation), which utilizes conversational context to generate meaningful samples for training. We also mitigate the issues regarding unbalanced context lengths by changing the input-output format of the model such that it can deal with varying context lengths effectively. Specifically, our proposed model, trained with the proposed data augmentation technique, participated in the sarcasm detection task of FigLang2020, have won and achieves the best performance in both Reddit and Twitter datasets.
READ FULL TEXT