A Quantum Theory-Based Hybrid Neural Model to Improve Textual Intelligence for Threat Entity Classification

Abstract

Here we introduce an uncomplicated hybrid model for threat-entity classification: A typical neural network progresses in parallel with a quantum inspired course that rotates features and uses a wave-like attention. Sentence embedding we couple with some basic counts of entities, project back down in dimensionality, and balance the classes with SMOTE; Isolation Forest and Random Forest assist in capturing edge-case oddities. On a structured cyber-threat data set, the model achieves 97.2% accuracy and 0.97 F1-score (precision 0.96, recall 0.98, AUC 0.99), with some misses (~2.2% FN) and some false alarms (~4.3% FP). The trade-off is interpretability: the quantum-inspired pathway improves generalizability but makes the decisions harder to interpret.