Dual-channel LSTM-GRU Network with Mayfly-optimized Attention for Multilingual Text Categorization

This study presents the Hybrid Attention-optimized Mayfly Network (HAMNet), an advanced text classification framework that integrates three scientifically grounded innovations: (1) a dual-channel LSTM-GRU architecture for robust sequential modeling, (2) a multi-head attention mechanism with adaptive feature fusion, and (3) Mayfly Optimization for automated hyperparameter tuning. Through rigorous experimental validation, HAMNet demonstrates state-of-the-art performance, achieving an accuracy of 0.9684 and an F1-score of 0.9683 on benchmark datasets — surpassing competitive baselines by 2.4–5.7% in absolute metrics. The model’s design incorporates systematic ablation studies to verify each component’s contribution, while cross-lingual tests confirm its adaptability to both Chinese and English corpora. The optimization framework exhibits 30% greater efficiency than conventional approaches while maintaining stable convergence properties. Meticulously designed experimental protocols and reproducible configurations underscore the scientific rigor of this work, which advances the field of neural text classification through its harmonized integration of architectural innovation and methodologically sound optimization strategies.