Detecting Discrete Emotions In Text Using Neural Networks

In recent years, emotion detection in text has become more popular due to its vast potential applications in marketing, political science, psychology, human-computer interaction, artificial intelligence, etc. Access to a huge amount of textual data, especially opinionated and self-expression text, also played a special role in bringing attention to this field. In this work, we review the work that has been done in identifying emotion expressions in text and argue that although many techniques, methodologies, and models have been created to detect emotion in text, these methods, due to their handcrafted features and lexicon-based nature, are not capable of capturing the nuance of emotional language. By losing the information in the sequential nature of the text, and inability to capture the context, these methods cannot grasp the intricacy of emotional expressions, therefore, are insufficient to create a reliable and generalizable methodology for emotion detection. By understanding these limitations, we present our deep neural network methodology based on bidirectional GRU and attention mechanism and the fine-tuned transformer model (BERT) to show that we can significantly improve the performance of emotion detection models by capturing more informative text representation. Our results show a huge improvement over conventional machine learning methods on the same dataset with an average of 26.8 point increase in F-measure on the test data and a 38.6 point increase on a new dataset unseen by our model. We Show that a bidirectional-GRU with attention could perform slightly better than BERT. We also present a new methodology to create emotionally fitted embeddings and show that these embeddings perform up to 13% better in emotion similarity metrics.