PDA: Data Augmentation with Preposition Words on Chinese text classification

While previous data augmentation methods have shown good results, some of them have limitations in applicability and complexity. In this paper, we propose a simple Chinese data augmentation technique called Preposition Data Augmentation (PDA). The principle of PDA is straightforward, which involves randomly inserting Chinese prepositions into the original sentences. The augmented samples maintain the same order as the original ones, preserving the semantic information and retaining all the input information. According to our experimental results on Chinese datasets, our PDA method demonstrates a certain advantage over AEDA [5], EDA [13], and back-translation [11] in terms of model generalization. We conducted experiments on various Chinese classification tasks, and the results indicate the advantages of our PDA data augmentation method.