Optimized dynamic multi-head attention for abstractive text summarization

A novel architecture, dynamic multi-head attention-based LSTM (DMHA-LSTM), which enhances sequential modeling through an adaptive attention mechanism and paves the way for future integration with pre-trained language models and deployment in low-resource environments.