Detection of hatefulness in Memes using unimodal and multimodal techniques
Aruna B. Bhat, Vaibhav Kumar Varshney, Varun Bajlotra, Vishesh Gupta · 2022 6th International Conference on Intelligent Computing and Control Systems (ICICCS) · 2022
Memes have become a common way to spread humor, information, and ideas. However, some objectionable content in memes might transmit hatefulness among people. Identification of hatefulness in memes is thus essential and even challenging as it simultaneously involves analyzing two modes of information. We propose two different methods to classify hateful memes through this research work. The first method focuses on using BERT to improve the efficiency of existing text based models. The second method introduces a novel Multimodal technique which detects objects using pre-trained models like InceptionV3, ResNet50, and Xception and combines the modalities of text and image. The models created are lightweight and perform on par with complex transformer models.