Data Augmentation methods in natural language processing.

• Ano de defesa: 2021
• Autor(a) principal: Taynan Maier Ferreira
• Orientador(a): Anna Helena Reali Costa
• Banca de defesa: Aline Marins Paes Carvalho, Thiago Alexandre Salgueiro Pardo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade de São Paulo
• Programa de Pós-Graduação: Engenharia Elétrica
• País: BR
• Link de acesso: https://doi.org/10.11606/D.3.2021.tde-04112021-162156

Data Augmentation (DA) methods a family of techniques designed for synthetic gen eration of training data have shown remarkable results in various Deep Learning and Machine Learning tasks. Despite its widespread and successful adoption within the com puter vision community, DA techniques designed for natural language processing (NLP) tasks have exhibited much slower advances and limited success in achieving performance gains. As a consequence, with the exception of applications of back-translation to machine translation tasks, these techniques have not been as thoroughly explored by the wider NLP community. There is no unified view or comparative analysis between the various DA methods available. Furthermore, there still lacks a proper practical understanding of the relationship between DA and several important aspects of model design, such as training data and regularization parameters. In this work, we perform a comprehensive study of NLP DA techniques, comparing their relative performance under different settings in Sentiment Analysis tasks. We also propose Deep Back-Translation, a novel NLP DA technique. We perform qualitative and quantitative analysis of generated synthetic data, evaluate its performance gains and compare all of these aspects to previous existing DA procedures.