<![CDATA[Background: The increasing complexity of cybersecurity mandates, such as BSSN and Kominfo regulations, presents a significant challenge for automated compliance auditing in Indonesia. Traditional NLP models often struggle with the semantic gap between formal regulatory language and raw technical system telemetry, leading to high false-negative rates in security monitoring. Objective: The purpose of this research to develop a robust deep learning framework to automate cybersecurity compliance assessments while addressing the linguistic challenges of the Indonesian regulatory landscape. The primary goal is to enhance the detection of non-compliant system behaviors by bridging the gap between documentation and real-time logs. Methods: The proposed framework utilized a Transformer-based BERT architecture integrated with a novel Directional Augmentation (DA) mechanism. The methodology follows a four-phase process: (1) data collection of 8,240 labeled points, (2) an Anti-Leak Grouping Strategy to prevent data memorization, (3) implementation of DA through Semantic Polarization and Technical Jargon Injection, and (4) model training and evaluation. Result: The findings of this research are indicate that the proposed framework significantly outperformed the baseline BERT model. Test Accuracy rose from 90.15% to 95.72%, while Validation Accuracy improved from 91.20% to 96.88%. The final model achieved an Overall Accuracy of 94.39%, maintaining a balanced F1-score and effectively reducing False Negatives to only 32 cases in the detection of security violations, Conlussion : Integrating Directional Augmentation into BERT optimizes Indonesian cybersecurity auditing by synchronizing BSSN/Kominfo regulatory language with technical telemetry through semantic polarization and jargon injection.]]>
Copyrights © 2026
