Emoji Based Sentiment Classification Using Machine Learning Approach

Abstract

Online media platforms like Facebook, Twitter, and Instagram continue to influence our world. People today are more closely connected than ever before, and they exhibit such sophisticated personas. Ongoing studies have revealed a link between excessive social media use and depression. This research focuses on the development of features using Term frequency – Inverse Document Frequency (TF-IDF) and Bag-of-Word (Bow). The work will also put light on the generation of models utilizing a machine-learning technique. The dataset for this research using Twitter API. Only the English context was kept from the Tweets after filtration. It focuses on categorizing users’ mental health at the tweet level. About 20000 reviews make up this dataset. Using these reviews, emoji sentiment classification has been developed. Data is cleaned using a pre-processing approach before BoW and TF-IDF were used to extract features. Following to it, classifier deployment, training and assessment were carried out. Metrics for evaluation are used to gauge classifier accuracy. MultinomialNB (MNB) fared best in the field of Bag-of-words features among the three classifiers used to evaluate the accuracy, but Random Forest (RF) outperformed TF-IDF. In Bag-of-Words, we are able to classify data with an accuracy of 86% using TF-IDF Random Forest and 89% using multinomial NB and BoW.