A Chinese Chemical Fire Named Entity Recognition Model Incorporating Text Hierarchical Structure Features and Multi‐Head Attention Mechanisms

The Chinese chemical fire named entity recognition (NER) task often faces great challenges such as rare Chinese characters, complex word structures, and varied utterance structures. Thus, this paper presents MulFA‐NER, an entity recognition model that integrates text hierarchical structural features and a multi‐head attention mechanism. The MulFA‐NER model employs three distinct hierarchical text‐structure channels at the character, word, and sentence levels to facilitate input distributed feature representation. Then deep learning models such as multi‐head attention and Bidirectional Long Short‐Term Memory Network (BiLSTM) are used for feature fusion and sequence feature encoding respectively. Subsequently, the Conditional Random Field (CRF) model is employed to accurately decode labels, thereby enabling the recognition of named entities in complex Chinese text corpora. The experimental results, using Chinese chemical disaster accident reports, show that the MulFA‐NER model exhibits superior performance compared to the six baseline models, achieving precision, recall, and F1 values of 94.92%, 95.42%, and 94.91%, respectively. Furthermore, through a comprehensive analysis of the model's performance dimensions, ablation experiments, and generalisation tests, it can be verified that the MulFA‐NER model demonstrates remarkable robustness and adaptability in the domain of Chinese text NER.