A Method for Constructing High-Quality Text Datasets in the Field of Molding of Polymer Materials

The construction of high-quality text datasets is of great significance for the efficient utilization of historical data and empirical knowledge in a vast amount of scientific materials. Limited by the complexity and high heterogeneity of texts related to the molding of polymer materials, traditional dataset construction methods are difficult to be directly applied to the construction of high-quality text datasets in the injection molding industry. This paper proposes a supervised method for constructing text datasets for the molding of polymer materials. Through high-quality text annotation and supervised data augmentation, it effectively reduces the generation of noisy data and provides high-quality data support for text analysis and model training in the molding of polymer materials. The method consists of four steps: automatic literature collection with source tracking, literature pre-processing based on task requirements, entity label definition and data annotation combined with industry characteristics, and text data augmentation with the integration of domain expertise. The first three steps aim to improve the quality of text corpora in the field of polymer material molding. The last step involves integrating domain knowledge into pre-trained large language models to conduct supervised text data expansion for polymer material molding, thus enhancing the text corpora in the field of polymer material molding.