Analisis Sentimen Penanganan Covid-19 Menggunakan Metode Long Short-Term Memory Pada Media Sosial Twitter

Abstract

Social media can be used to convey people's aspirations for government policies. Several government policies regarding the regulation of the handling of Covid-19 often elicit responses and criticism from the public, especially on Twitter social media. The aspirations conveyed can contain positive or negative responses. To find out the representation of public sentiment based on these responses, it is necessary to do a sentiment analysis technique. The Long Short-Term Memory (LSTM) method is a deep learning method that can be used in sentiment analysis. LSTM is used because it has the advantage of being able to store large amounts of information in memory cells. Before carrying out classification modeling, the dataset must go through the process of case folding, punctuation removal, normalization, and stopword removal. This aims to ease the training process by eliminating characters or words that are not needed. Next, the word is vectorized using FastText, the goal is to change the string data type to an array vector, so that the word can be processed in the LSTM. The final performance of the model is measured based on the value of precision, recall, accuracy, and f Measure. Based on testing the dropout layer parameters on the hidden layer against 10 fold cross validation, the average accuracy of model testing resulting from all k folds is 72.4%. with the maximum model performance achieved at k fold = 9, when using a dropout layer of 0.4, the values ​​achieved are precision, recall, accuracy, and f measure respectively: 76.74%, 80.49%, 78.31%, 78.57%.