Optimization of Artificial Intelligence Natural Language Processing Model Based on Deep Neural Network
Mingrui Jia · IEEE Access · 2025
This paper proposes a novel framework integrating Hierarchical Relational Transformer (HRT) and Adaptive Semantic Calibration (ASC) to address challenges in semantic coherence, transferability, and interpretability, especially in multilingual and ambiguous environments. The HRT enhances traditional transformer models by incorporating cross-layer adaptive composition and dynamic token interactions, enabling deeper relational encoding and better contextual understanding. ASC refines inference through uncertainty-aware decoding and hierarchical memory scheduling, balancing syntactic fluency, semantic precision, and computational efficiency. Together, HRT and ASC support a broad range of NLP tasks, such as reasoning, summarization, and dialogue, with improved accuracy, consistency, and generalizability. Experimental results show significant improvements in interpretability and cross-task stability across multiple benchmarks, offering a scalable pathway for future NLP advancements.