Contextual Analysis of Immoral Social Media Posts Using Self-attention-based Transformer Model

Abstract

Immoral posts detection on social media is a serious issue in this digital era. This matter wants advanced natural language processing (NLP) methods to address user-generated text's difficult semantics and context. Incorporating advanced deep learning (DL) techniques improves the model's aptitude to handle challenges such as slang, sarcasm, and vague expressions. This work suggests a deep contextual analysis framework using a self-attention-based transformer model to detect immoral contents on soil networks efficiently. The model captures complex contextual associations and semantic nuances by harnessing the strength of self-attention mechanisms. The proposed technique enables proper differentiation between moral and immoral content. The framework is assessed on two benchmark datasets, SARC and HatEval. The experiment shows the highest F1-score, 98.10%, on the SARC dataset. While on HatEval, the model achieved 97.34%, representing greater performance than state-of-the-art approaches. The results highlight the efficiency of self-attention-based DL models in delivering efficient, scalable, and ethical answers for observing and modifying harmful content on social media networks.