Mimeme Attribute Classification using LDV Ensemble Multimodel Learning

Abstract

One of the most common types of social networking interaction is memes. Memes are innately multimodal, so studying and processing them is a hot issue currently. This study's analysis of the DV dataset comprises classifying memes according to their irony, humour, motive, and overarching mood. The effectiveness of three different creative transformer-based strategies has been carefully examined. The DV Dataset used here is created by own meme data for this implementation analysis of hateful memes. Out of all of our strategies, the proposed ensemble model LDV obtained a macro F1 score of 0.737 for humour classification, 0.775 for motivation classification, 0.69 for sarcasm classification, and 0.756 for overall sentiment of the meme.