Sentiment Analysis of Twitter Posts on COVID-19 Vaccines

Abstract

This paper presents the development of a system for analyzing Twitter data about COVID-19 vaccines in Serbia. The analysis includes a sentiment detection system and a topic detection system related to the tweets. For the project's development, a labeled set of English tweets was used, along with a collection of Serbian tweets that were translated into English. Two approaches were tested for sentiment detection: supervised learning by training a model on a labeled dataset of English tweets about COVID-19. The best performance in sentiment detection was achieved using convolutional neural networks, which achieved an accuracy of 59%. It was identified that there are 38% positive, 37% negative, and 25% neutral Serbian tweets in the dataset. The topic model achieved a coherence score of 0.45, identifying fifteen topics.