Zombies Eat Brains, You are Safe: A Knowledge Infusion based Multitasking System for Sarcasm Detection in Meme

Anonymous

Abstract: Sarcasm detection is, in itself, a challenging task in the field of Natural Language Processing (NLP), and the task even becomes more complex when the target is a meme. In this paper, we first hypothesize that sarcasm detection is closely associated with emotions present in the meme. We propose a deep learning-based multitask model to perform these two tasks in parallel, where sarcasm detection is the primary, whereas emotion recognition is considered as an auxiliary task. Furthermore, we propose a novel knowledge infusion (KI) method to get a sentiment-aware knowledge representation on top of our multitasking model. This sentiment-aware knowledge representation is obtained from a pre-trained parent model and subsequently, this representation is used via a novel Gating Mechanism to train our downstream multitasking model. For training and evaluation purposes, we created a large-scale dataset consisting of 7,416 sample Hindi memes as there was no readily available dataset for building such multimodal systems. We collect the Hindi 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. We demonstrate the effectiveness of our proposed work through extensive experiments. The experimental results show that our proposed system achieves a 64.48% macro F1-score, outperforming all the baseline models. Finally, we note that our proposed system is model agnostic and can be used with any downstream model in practice. We will make the resources and codes available.

Paper Type: long