Sentiment Analysis of the 2024 Presidential Candidates Using SMOTE and Long Short Term Memory

Numerous political leaders participate in elections since they are a crucial component of the political process. Since electability is an issue, steps are taken to make political candidates running in general elections more electable. The media, including internet news media, has emerged as one of the key strategies for raising electability. Reader comments can be analyzed for sentiment to provide an evaluation of political figures. However, because the comments contain unstructured content, particularly in Indonesian text, it is difficult to interpret the sentiments of different comments in online news media. In this research, an analysis of public sentiment towards the 2024 presidential candidates will be carried out which is expressed through the Twitter social network. There are several stages to carry out sentiment analysis, including the stages of data collection, data preprocessing, balancing the distribution of the number of datasets, and sentiment classification using the LSTM method with word2vec feature representation. The results of this study show that the LSTM method combined with SMOTE due to the limited amount of data is able to produce a fairly good LSTM model with an average accuracy of 89.42% and a loss value of 0.24, the ideal scenario is when the accuracy is high and the loss is minimal, in which case the LSTM model only exhibits minor errors on a subset of the data.