民机装配工艺规范文本的弱监督命名实体识别方法OA

In high-end equipment manufacturing,such as civil aircraft manufacturing,a significant amount of unstructured assembly process specification texts still rely heavily on manual interpretation in practical applications.This leads to high comprehension costs and challenges in machine-based intelligent processing.To facilitate efficient utilization of assembly process information,this paper proposes a named entity recognition method tailored for assembly process texts,aiming to extract structured information.On one hand,this approach enhances user interpretation efficiency and reduces human comprehension discrepancies;on the other hand,it transforms the information into machine-readable knowledge data,thereby promoting the intelligent development of assembly process design.A prototype space-based recognition model is introduced,which employs weakly supervised learning with small-scale data to avoid dependence on costly large-scale anno-tated datasets.Meanwhile,by enhancing the prototype space through contrastive learning optimization and semantic mod-eling,the model's ability to recognize entity boundaries and contextual information is strengthened,enabling superior per-formance under few-shot conditions.Experimental results demonstrate that the proposed method achieves satisfactory rec-ognition performance across multiple categories,validating its practicality and potential for broader application in assembly process text processing.