Identification of Trolling in Memes Using Convolutional Neural Networks
Manohar Gowdru Shridara, Daniel Hládek, Matúš Pleva, Renát Haluška · 2023
In this paper, we applied a convolutional neural network based on classifier to classify the given input image whether it was trolled image or a normal image, which creates a suggested priority based on its input, which includes vectored image data and emotion values. We applied the traditional methods of natural language processing to preprocess the text. In particular, we utilized the Tesseract library to extract text from meme images for the classification of troll memes. The tokenization method is used to prepossess meme image text data. We extracted the features from the texts by Word2Vec modelling. The classification was done only on the basis of the displayed text. We do not consider memes that do not contain any text. The results of our study indicate that the suggested model performs effectively with 98 percent accuracy rates.