<![CDATA[National elections are critical components of a functioning democracy, as they empower individuals to influence their nation's trajectory through active participation. In the contemporary digital landscape, social media platforms such as Twitter have emerged as pivotal forums for political discourse, enabling the expeditious dissemination of public sentiments. Nevertheless, assessing sentiment on Twitter poses various difficulties, primarily due to the casual nature of the language, the prevalent use of slang, sarcasm, and emoticons that often convey implicit emotional undertones. The present study puts forth a sentiment analysis framework that utilizes the Bidirectional Encoder Representations from Transformers (BERT) model, particularly IndoBERT, augmented with insights derived from emoticons. Emoticons are classified into three sentiment categories: positive, negative, and neutral. The dataset under consideration is composed of Indonesian tweets that have been pre-labeled and that pertain to the 2024 national election. Two configurations of the model were evaluated: a foundational IndoBERT model that relies solely on text, and an enhanced model that includes binary emoticon features. The experimental findings reveal that the emoticon-inclusive model attained a higher accuracy (78.5%) as opposed to the baseline model (77.7%) and demonstrated enhanced sensitivity in distinguishing neutral and negative sentiments. This finding suggests that emoticons offer valuable contextual information, thereby enhancing the accuracy of sentiment classification. The strategic integration of emoticons and BERT for Indonesian political sentiment analysis has received scant attention, rendering this method a novel addition to the field. The findings underscore the potential benefits of integrating text-based deep learning systems with emoticon characteristics to more effectively capture intricate emotional expressions in social media, particularly during political campaigns.]]>
