Explainable Emotion Recognition from Tweets using Deep Learning and Word Embedding Models
Abdulqahar Mukhtar Abubakar, Deepa Gupta, Suja Palaniswamy
Abstract
Social media such as Twitter made it possible for people to express their mood through text, images, audios or videos. Understanding these emotions becomes vital for better human and computer interactions. Natural Language processing play an important role in classifying the textual emotions of people. This research proposes a method to classify emotions from text into six different categories namely anger, fear, joy, love, sadness, and surprise from the Emotion Recognition dataset, using state of the art (SOTA) pre-trained word embeddings and deep learning models. The results of the experiments demonstrate that DistilBert and CNN attained an F-score of 98%. The explainability modules explain the training and prediction of the proposed model by analyzing the contextual contribution of words in classification.