Data Augmentation for Technical Standard Relation Extraction

The paper introduces a method for fine-grained relation extraction in grid technology standards, addressing challenges in manual annotation due to complex guidelines and large-scale dataset requirements. Data augmentation techniques, specifically word-level perturbation and sentence template methods, are applied to a limited annotated dataset to improve model generalization and efficiency. Experiments on Q/GDW 1168-2013 annotated data show that the GPLinker model has demonstrated its higher efficiency and accuracy in relation extraction, with the integrated use of word-level perturbation and sentence template augmentation significantly boosts the model's F1-score to 0.582925, indicating improved precision and recall in relation extraction from technical standards.