Comparative Analysis of Deep Learning Models for Multi-label Sentiment Classification of 2024 Presidential Election Comments

This study evaluates the effectiveness of various deep learning architectures in multi-label sentiment classification of YouTube comments related to the 2024 Indonesian presidential election. The dataset includes comments from debate videos featuring candidates Anies Baswedan, Prabowo Subianto, and Ganjar Pranowo. The research compares Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and a hybrid CNN-BiLSTM model. Preprocessing steps included normalization, removal of unwanted characters, case folding, stopwords removal, and text augmentation. Class imbalance was addressed using class weights in the loss function. Models were evaluated using accuracy, Area Under the Curve (AUC), and Hamming Loss. The Bi-LSTM model outperformed others with an average accuracy of $98 \%$ and an AUC of 0.92. Despite the potential of the hybrid model, it did not surpass the Bi-LSTM model. The Bi-LSTM model’s superior performance is attributed to its ability to capture long-term dependencies and contextual information. This model was integrated into a Flask web application for real-time sentiment analysis, demonstrating practical application and providing insights into public sentiment during the election. This research contributes to sentiment analysis in political contexts by showcasing the BiLSTM model’s effectiveness and addressing class imbalance issues.