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.