An Advanced Bert Transformer for Spoiler Detection in Extension System for Social Media

Abstract

A Spoiler Detection Extension System is a browser-based tool designed to identify and hide spoilers in online content, such as reviews, articles, and social media posts. It leverages Natural Language Processing (NLP) and machine learning models to analyze text and detect potential spoilers based on context, sentiment, and key phrases. The system then either blurs, hides, or warns users before displaying spoiler content. The detection of spoilers represents an increasingly vital task because social media users continue to intensively discuss movies and television shows as well as books through their platforms. The research presents DistilBERT-TSD as a transformer-based model for effective spoiler identification which unites YAMNET attribute extraction techniques with DistilBERT contextual components. Through deep learning approaches the model reaches high precision in spoiler detection where it stands above the citation TF-IDF + SVM and CNN and LSTM and BERT-based models. Real-world dataset tests validate that DistilBERT-TSD produces 92.5% accuracy alongside a 0.96 AUC-ROC score which surpasses existing criterion. Research shows that the element of self-attention weights with transformer hidden states and final contextual embeddings plays a vital role in the classification of spoilers. The results from epoch-based evaluation establish that training for ten epochs delivers the most suitable performance level combined with generalization capabilities. The research demonstrates that DistilBERT-TSD operates as a highly performing spoiler detection system which creates a foundation for evolving multi-modal spoiler detection on social media platforms.