Deep Learning and Data Augmentation for Detecting Self-Admitted Technical Debt