Sentiment Analysis with Supervised Topic Modelling on Twitter Data Related to Indonesian Election 2024

Generally, The topic model is an unsupervised learning approach to find structures between topics and terms and relationships between documents and topics. In this research, we evaluated topics in the 2024 presidential election dataset related to their output i.e., sentiment polarity. This paper proposes a Supervised Latent Dirichlet Allocation (sLDA) approach to predict the observed sentiment outcome and adjust the topic model concurrently. We used the Twitter dataset related to the 2024 Indonesian election and the other tweets datasets as a comparison we got from the Kaggle website about the 2017 Jakarta regional head election. The Sentiment annotation process for each dataset uses the InSet Lexicon. This research contributes to exploring topics from Indonesian Twitter data and predicting the labels appearing from the resulting topics. The sLDA package from R is used. It is worth noting that in this study, we handled sentiment analysis as a regression task then the evaluation result of the model is in the form of a correlation and R2 between the prediction and the actual score sentiment. The results show that the correlation and R2 value increase as the number of topics increases. The highest correlation for the 2024 election dataset is 0.599, and R2 is 0.352 with 50 topics, and the highest correlation for the Jakarta regional election 2017 dataset is 0.574, and R2 is 0.307 with 25 topics.