SoftEDA: Rethinking Rule-Based Data Augmentation with Soft Labels
This addresses a specific issue in NLP data augmentation for classification tasks, but it is incremental as it builds on existing rule-based methods.
The paper tackled the problem of rule-based text data augmentation damaging text meaning and hurting model performance by proposing a technique using soft labels for augmented data, resulting in demonstrated effectiveness across seven classification tasks.
Rule-based text data augmentation is widely used for NLP tasks due to its simplicity. However, this method can potentially damage the original meaning of the text, ultimately hurting the performance of the model. To overcome this limitation, we propose a straightforward technique for applying soft labels to augmented data. We conducted experiments across seven different classification tasks and empirically demonstrated the effectiveness of our proposed approach. We have publicly opened our source code for reproducibility.