<![CDATA[Background: Digital government transformation requires efficient and integrated management of SPBE documents usingartificial intelligence (AI).Objective: To develop an AI chatbot to solve the problems of manual searches, human error, and difficulties in extractingtext from complex PDF documents in SPBE.Methods: The study combined Forward Chaining (for structured questions) and the IndoBERTQA Transformer model(for open-ended questions). Text was extracted using PyPDF2. The evaluation used 150 question-context-answer datapairs and two Forward Chaining scenarios.Result: Forward Chaining produced fast and fixed responses, while IndoBERTQA provided more contextual answers(Cosine Similarity of 0.730). However, the low F1-score (0.316) shows that prediction accuracy is still a major weaknessand needs priority improvement. The system successfully sped up information access, although it still struggles withextracting tables and non-text elements.Conclusion: Integrating both methods provides a good balance between the chatbot’s speed and flexibility. Furtherdevelopment on dataset quality, model optimization, and document extraction is highly necessary to improve the system’saccuracy.]]>
Copyrights © 2026
