This paper explores improving emotion recognition in software engineering communication through data augmentation. It summarizes that developers often express emotions in communications, but current tools perform poorly at emotion recognition due to a lack of large, high-quality datasets. The paper manually annotates 2000 comments with emotions, extends an emotion taxonomy, and analyzes errors in existing tools. It then evaluates three data augmentation strategies and finds that augmenting data while preserving sentiment polarity improves tool performance the most. The paper contributes annotated data and data augmentation source codes to aid future emotion recognition research.