Fine-Grained Entity Classification Technology for Data Standard Adaptation

The data types involved in the distributed manufacturing operation and maintenance of a full life model are highly complex, exhibiting significant differences in tems of structure, content, and standards. In practical application scenarios, numerous challenges persist in the classification and govermance of diverse source data. The proccss of govermance entails reorganizing the data: firstly, collecting the disparate data from each business domain; secondly, cleasing the original data based on its characteristics and addressing non-standardized elements; thirdly, aggregating the source data into a central repository (data lake) where it is managed according to its characteristics before being divided based on different standards. Our analysis reveals that current classification technologies fail to adequately addresss issues related to discrete structural variations within source data or achieve automation. To overcome this limitation and enable automated adaptation to standardized formats for efficient utillization of data resources we explore natural language processing methods and propose a combined feature etraction model leveraging BERT for semantic information extraction alongside CNN for integration discrete structural information.