Volta at SemEval-2021 Task 6: Towards Detecting Persuasive Texts and Images using Textual and Multimodal Ensemble

Memes are one of the most popular types of content used to spread information\nonline. They can influence a large number of people through rhetorical and\npsychological techniques. The task, Detection of Persuasion Techniques in Texts\nand Images, is to detect these persuasive techniques in memes. It consists of\nthree subtasks: (A) Multi-label classification using textual content, (B)\nMulti-label classification and span identification using textual content, and\n(C) Multi-label classification using visual and textual content. In this paper,\nwe propose a transfer learning approach to fine-tune BERT-based models in\ndifferent modalities. We also explore the effectiveness of ensembles of models\ntrained in different modalities. We achieve an F1-score of 57.0, 48.2, and 52.1\nin the corresponding subtasks.\n