Text Data Augmentation Impact on Text Classification Models

This article presents research results on the impact of textual data augmentation on the classification model training process and its resulting quality. The importance of using text augmentation techniques for specific tasks is justified. This study examines how the different text augmentation techniques affect model's tendency to overfit and their impact on resulting accuracy. For that purpose, a sentiment classifier based on recurrent neural network was developed and tested. Experiments focus on the impact of the augmentation factor on classifying data using a small-sized dataset.