Mixing Approach for Text Data Augmentation Based on an Ensemble of Explainable Artificial Intelligence Methods

To improve the accuracy and robustness of a model, text data augmentation is utilized to expand data. Among various text data augmentation methodologies, the method of mixing two or more data to generate augmented data is one of the most used methodologies due to its intuition. However, existing methodologies of data mixing have a critical disadvantage in that the importance of mixed-up words cannot be considered by simply applying a weighted summation. Thus, in this study, we propose a novel mixing-based text augmentation approach based on explainable artificial intelligence to consider the importance of mixed-up words. Experimental results confirmed that the proposed method outperforms the existing augmentation method. To the best of our knowledge, the proposed study is the first study considering the importance of mixed-up words among mixing-based approaches.