TY - JOUR AU - Firda Rachmawati AU - Ulil Azmi AU - Rahmania Azwarini PY - 2025/03/11 Y2 - 2025/04/02 TI - Comparison of Lexicon-Based Methods and Bidirectional Encoder Representations for Transformers Models in Sentiment Analysis of Government Debt Market Movements JF - International Journal of Engineering and Computer Science Applications (IJECSA) JA - IJECSA VL - 4 IS - 1 SE - Articles DO - https://doi.org/10.30812/ijecsa.v4i1.4832 UR - https://journal.universitasbumigora.ac.id/index.php/IJECSA/article/view/4832 AB - The State Budget of Indonesia (APBN) is the main tool for implementing fiscal policies and serves as a budgeting guideline for development execution in Indonesia. One of the funding sources in budget financing is Debt Financing, which consists of Government Securities (SBN) issuance and Loans. Overall, SUN contributes IDR 5,824.34 trillion, highlighting its significant proportion in debt financing. Understanding public sentiment toward SUN is essential in developing effective government policies. This research conducts sentiment analysis on tweets from the social media X over the past 7.75 years to assess public perception and propose strategic recommendations. The aim of this research is to compare the BERT model and the Lexicon-Based method to determine which achieves the highest accuracy in sentiment analysis. The findings can help the government develop strategies for issuing SUN, especially in improving public involvement and investor trust. This research method is based on a deep learning pre-trained Bidirectional Encoder Representations from Transformers (BERT) model, specifically IndoBERT, with fine-tuning, and a Lexicon-Based approach utilizing the InSet lexicon. The results of this research are as follows: on the overall tweet dataset, the BERT model with optimal hyperparameters outperformed the Lexicon-Based method, achieving an accuracy of 70.28% compared to 55.77%. Similarly, on an annual basis, BERT exhibited higher accuracy than the Lexicon-Based method, except in 2021. Public sentiment on SUN in social media X is categorized as 49% positive, 30% neutral, and 21% negative. These findings indicate a generally favorable perception of SUN but also highlight areas for improvement in public communication. Overall, the BERT model demonstrates superior performance over the Lexicon-Based method. Considering the opportunities available, the government could leverage social media through Key Opinion Leaders and enhance transparency in explaining policies such as Tapera. This approach could maximize public participation in investing in SUN in Indonesia. ER -