Analisis Sentimen Hasil Pemilu (Quick Count) Calon Presiden dan Wakil Presiden 2024 di Media Sosial Media X Menggunakan Metode Bidirectional Long Short-Term Memory (BiLSTM)

Qurrotu Aini (Universitas Nurul Jadid, Jawa Timur)
M. Noer Fadli Hidayat (Universitas Nurul Jadid, Jawa Timur)
Abu Tholib (Universitas Nurul Jadid, Jawa Timur)

Journal of Computer System and Informatics (JoSYC)
Vol 5 No 3 (2024): May 2024

It is important to understand public opinions, attitudes and sentiments in relation to presidential and vice presidential candidates in the context of Indonesia's general elections. The fact that quick count results have become a major topic of conversation on social media, especially on platforms such as Twitter, shows how important it is to monitor people's views on election results. However, tweets that are free-form and use digital language are often difficult for the unfamiliar to understand, which can lead to the spread of misinformation or inaccurate views. Sentiment analysis is therefore key in understanding people's views on election results. This research proposes the use of the Bidirectional Long Short Term-Memory (BiLSTM) method to analyse sentiment related to the quick count results of the 2024 presidential and vice presidential elections on X social media. This sentiment analysis aims to classify texts into positive, negative, or neutral categories. The purpose of this study is to measure the sentiment value and accuracy of the BiLSTM method in sentiment analysis of election results. Data was collected by scraping X social media using the keywords "quick count results of 2024 presidential election" and "results of 2024 presidential election", resulting in 1348 tweets. Preprocessing included cleaning, case folding, normalisation, tokenisation, stopword removal, and stemming. Sentiments were labelled using the Vader lexicon dictionary. BiLSTM modelling was performed by dividing the data into 70% for training and 30% for testing. The results showed that neutral sentiment had the highest percentage at 92.86%, followed by positive sentiment at 3.83% and negative at 3.31%. The BiLSTM model achieved an accuracy of 86.89% with an overall accuracy of 97%. The highest precision, recall, and F1-score values were found in the neutral class, at 98%, 99%, and 99% respectively. This research proves that BiLSTM is an effective method for sentiment analysis of complex texts such as election results.