Text Data Augmentation Using Generative Adversarial Networks, Back Translation and EDA

Abstract

Data or information augmentation techniques have been explored in NLP to create greater textual information for training. But, the overall performance advantage of present strategies is regularly marginal. Text augmentation can generate additional variations of the original text and improve the generalization ability of a machine learning model that processes natural language text data. The paper represents the performance of Generative Adversarial Networks for the overall performance in text classification. The results show that the Generative models give the best overall performance advantage over the EDA or back translation accuracy. The proposed models process text augmentation using GANs compared to methods like Easy Data Augmentation and back translation. The goal of EDA is to generate new, semantically similar sentences from an existing sentence, to increase the size and diversity of a dataset for training natural language processing models. Back Translation is an improved method with increased accuracy. Generative Adversarial Networks (GANs) generate new, synthetic data that is similar to a given training dataset using neural network architecture. In the context of text augmentation, GANs can be used to generate new, realistic text similar to a given text dataset.