Developing an Emotion Recognition Tool for Tweets

Abstract

This research in the domain of sentiment analysis aims to maintain high effectivity and applicability by not just understanding how the user feels, but by accurately quantifying their emotions. With the desired end product being a web application, the visualization that is received as an output aids the user in comparing the intensity of their emotions of anger, fear, joy and sadness demonstrated in the input tweets. This web application was built with the use of a dataset from the SEMEVAL-2018 competition. Training and testing of the dataset using 6 machine learning algorithms and their evaluation using performance metrics including R2, MAE, MSE and RMSE led us to arrive at the result that 'Support Vector Regressor' was the best performing algorithm for anger, sadness and fear while the Gradient Boosting algorithm performed best for joy. To that end, the web application uses the following algorithm for evaluating the respective emotions. In addition to developing a prediction model, the research also involved extensive data visualization an analysis that conveyed the most used words and hashtags when the user experiences each of the aforementioned emotions. To highlight the research's accomplishment- the author has been able to create a fairly accurate and relatively quick sentiment analysis model for public opinion.