A Knowledge Infusion Based Multitasking System for Sarcasm Detection in Meme

Abstract

In this paper, we hypothesize that sarcasm detection is closely associated with the emotion present in memes. Thereafter, we propose a deep multitask model to perform these two tasks in parallel, where sarcasm detection is treated as the primary task, and emotion recognition is considered an auxiliary task. We create a large-scale dataset consisting of 7416 memes in Hindi, one of the widely spoken languages. We collect the memes from various domains, such as politics, religious, racist, and sexist, and manually annotate each instance with three sarcasm categories, i.e., i) Not Sarcastic, ii) Mildly Sarcastic or iii) Highly Sarcastic and 13 fine-grained emotion classes. Furthermore, we propose a novel Knowledge Infusion (KI) based module which captures sentiment-aware representation from a pre-trained model using the Memotion dataset. Detailed empirical evaluation shows that the multitasking model performs better than the single-task model. We also show that using this KI module on top of our model can boost the performance of sarcasm detection in both single-task and multi-task settings even further. Code and dataset are available at this link: https://www. iitp.ac.in/ ai-nlp-ml/resources.html#Sarcastic-Meme-Detection .