<![CDATA[This research develops an automated three-category news classification system for the official news portal of the Regional Revenue Agency (Bapenda) of Surabaya City, using a Long Short-Term Memory (LSTM) deep learning model to reduce reliance on manual, inconsistent categorization. The training corpus combined original institutional data, web-scraped articles from other regional revenue agencies, and GPT-generated synthetic text, all validated by domain experts, then processed through a structured preprocessing pipeline. On an unseen test set of 513 authentic institutional articles, the model achieved an overall accuracy of 77.78% and a weighted F1-score of 0.78, driven mainly by strong performance on the majority "Important Information" class (F1-score 0.87). However, the macro-averaged F1-score was only 0.4434, reflecting substantially weaker performance on minority classes, particularly "Information Technology" (F1-score 0.069). The model was deployed as a human-in-the-loop recommendation feature within the agency's existing dashboard.]]>
Copyrights © 2026
