Sarcasm Detection Framework Using Context, Emotion and Sentiment Features
Sarcasm detection is an essential task that can help identify the actual sentiment in user-generated data, such as discussion forums or tweets. Sarcasm is a sophisticated form of linguistic expression because its surface meaning usually contradicts its inner, deeper meaning. Such incongruity is the essential component of sarcasm, however, it makes sarcasm detection quite a challenging task. In this paper, we propose a model, that incorporates different features to capture the incongruity intrinsic to sarcasm. We use a pre-trained transformer and CNN to capture context features, and we use transformers pre-trained on emotions detection and sentiment analysis tasks. Our approach outperformed previous state-of-the-art results on four datasets from social networking platforms and online media.