Harnessing Advanced Learning for Sarcasm Detection

Sarcasm in social media and in various methods of communication attracts higher attention and create profound impact even more than the customary positive and negative responses. It is necessary to propose a prototype and henceforth a fully-fledged system as a solution to detect sarcasm, to spread positive etiquette and avoid misguidance and misinformation. Unlike other text classification tasks, sarcasm detection is bit more challenging as it relies more on contextual nuances rather than actual word sentiment. The work leverages 5 learning algorithms (Logistic Regression, Multi-head Attention Encoder, Siamese Networks, BiLSTM and BiGRU) with data extracted from SARC (Self Annotated Reddit Corpus) Dataset and News Headline dataset. The models use contextual nuances, linguistic features to detect and analyze sarcasm. An analysis on how the model’s performance varies based on the nature of text data (ie. formal and informal) has been made to learn about the nuanced nature of sarcasm.