International Research Journal of Education and Technology

The dramatic evolution of memes on social media has raised the challenge of new content moderation challenges, most notably the identification of hate speech and objectionable content made up of a combination of visual and text data. This paper suggests a Data Analytics System for Offensive memes Text Classification using Optical Character Recognition (OCR), Natural Language Processing (NLP), and sentiment analysis to identify and moderate toxic meme content on the internet.

The system proposed performs text extraction from memes via OCR, processes it with text mining methods (tokenization, stop word elimination, stemming), and classifies sentiment via the VADER (Valence Aware Dictionary and sEntiment Reasoner) algorithm. The tiered moderation model applies escalating interventions—warnings for initial transgressions, image blocking for second offenses, and account suspension for habitual perpetrators—while informing users via SMS. Operates in Python and MySQL, the system is a real-world solution to battling hate speech online through memes. Transfer learning and multimodal analysis are potential future extensions to enhance detection performance for other cultural backgrounds. The work contributes to automatic content moderation through solving the specific issues of hate speech in memes through end-to-end OCR-NLP.