A Transfer Learning Approach for Emotion Intensity Prediction in Microblog Text

Emotional expressions are an important part of daily communication between people. Emotions are commonly transferred non verbally through facial expressions, eye contact and tone of voice. With the rise in social media usage, textual communication in which emotions are expressed has also witnessed a great increase. In this paper automatic emotion intensity prediction from text is addressed. Different approaches are explored to find out the best model to predict the degree of a specific emotion in text. Experimentation was conducted using the dataset provided by SemEval-2018 Task 1: Affect in Tweets. Experiments were conducted to identify regression systems and parameter settings that perform consistently well for this problem space. The presented research highlights the importance of the Transfer Learning approach in inducing knowledge from state of the art models in sentiment analysis for use in the task of emotion intensity prediction. © Springer Nature Switzerland AG 2020.