Understanding Public Opinions of Government Measures Against COVID-19 Through Twitter Sentiment Analysis

Abstract

This study performs sentiment analysis on Twitter data to classify user opinions regarding Indonesian government policies during the COVID-19 pandemic. Prior research applying deep learning techniques like long short-term memory (LSTM) networks for sentiment analysis of social media data is limited. A dataset of 1000 tweets was collected using keywords related to government policies on vaccination, public activity restrictions, health protocols, and online learning. The tweets were preprocessed and word embeddings were generated using Word2Vec. An LSTM model was developed for sentiment classification and policy categorization. The results indicate that the combination of skip-gram Word2Vec and LSTM achieves an accuracy of 88% for sentiment analysis, outperforming other methods. Health protocols garnered the most positive sentiments while distance learning garnered the most negative sentiments. The study demonstrates the effectiveness of LSTM networks for sentiment classification of social media data to gain insights into public opinions.