DeepEmotionNet: Emotion mining for corporate performance analysis and prediction

•A deep learning-based model that can learn contextual, syntactic, document-level, and domain-specific semantic features to bootstrap emotion detection performance.•A real-world downstream application of the proposed model to detect top executives’ emotions from their Twitter postings.•The relationships between top executives’ emotions detected by the proposed model and corporate financial performance.•The predictive power of top executives’ emotions detected by the proposed model on corporate financial performance.•Our results demonstrate that more accurate emotion detection using our model can translate into substantial improvements in the downstream explanatory and predictive analytics of corporate financial performance.