Data augmentation for equipment technical standard entity recognition

This paper addresses the problem of entity recognition in equipment technical standards by proposing a data augmentation method based on entity dictionaries and numeric entities, combined with the knowledge expansion capabilities of large models to enrich training data. First, this paper designed a data augmentation process based on entity dictionaries, including sample pre-labeling, entity dictionary construction, and entity replacement, ensuring that the generated samples accurately reflect the complexity of the domain. Second, to handle common numeric entities in technical standards documents (such as parameter values, standard references, clause numbers, etc.), this paper proposed specialized replacement strategies to ensure both numerical randomness and character coverage. Additionally, semantic large models were leveraged to expand measurement units, addressing the issue of limited unit variety in small sample datasets and enhancing the model's robustness and generalization ability. Experimental results show that, after systematic data augmentation, the model performance improved significantly compared to before augmentation. Although resource consumption increased, the model's ability to recognize and classify entities was effectively enhanced, providing a solid foundation for subsequent automated analysis and information extraction.