An Intelligent Electronic System for Offensive Meme Detection using BERT and VGG-19

With user-generated content continuing to proliferate on social media, the application of offensive content in memes has created a critical problem that requires sophisticated means of Classification. This paper designs a novel framework that involves deep learning models for image and text classification. Specifically, the system utilizes Optical Character Recognition (OCR) technology to extract text from images, and afterward, applies the BERT (Bidirectional Encoder Representations from Transformers) model to analyze text and the VGG-19 convolutional neural network to analyze images. The framework can provide effective integration of various data types and thus enable the identification of offensive memes quickly and accurately. Experimental analysis on benchmarking datasets was used to measure the success of this approach for meme classification and its performance and stability. Multi-modal offensive content moderation was the identified issue, and an issue to the wider field of automated method to content study and enhanced user protection has been provided.