A Multi-Task Learning Approach to Sarcasm Detection (Student Abstract)

Abstract Sarcasm detection plays an important role in natural language processing as it has been considered one of the most challenging subtasks in sentiment analysis and opinion mining applications. Our work aims to detect sarcasm in social media sites and discussion forums, exploiting the potential of deep neural networks and multi-task learning. Specifically, relying on the strong correlation between sarcasm and (implied negative) sentiment, we explore a multi-task learning framework that uses sentiment classification as an auxiliary task to inform the main task of sarcasm detection. Our proposed model outperforms many previous baseline methods on an existing large dataset annotated with sarcasm.