Robust Optimized Deep Learning-Based Phishing Detection Framework for Semantic Web Systems Using Boosted Triangular Topology Aggregation Optimization

The growing prevalence of phishing attacks threatens the security of semantic web systems, necessitating more adaptive and precise detection mechanisms. This study proposes a novel phishing detection framework that combines deep learning and nature-inspired optimization to improve detection effectiveness. The framework integrates a comprehensive natural language processing pipeline with term frequency-inverse document frequency tokenization to transform email content into informative numerical vectors. A long short-term memory network is employed to model sequential and contextual dependencies. Critical hyperparameters are dynamically optimized through the newly introduced dynamic boosted triangular topology aggregation optimizer, which enhances search adaptability through cooperative interaction-driven adaptation, vortex-guided local refinement, and dynamic scout-based diversification.