Widiyatama, Oja and Rozak, Fatur and Rozie, Andri Fachrur and Arisal, Andria and Kurniasari, Dian (2026) BERT-Based Multi-Task Classification Model for Free Health Check-Up Program Analysis. In: 2025 International Conference on Computer, Control, Informatics and its Applications (IC3INA), 15-16 October 2025, Jakarta, Indonesia.

Text BERT-Based_Multi-Task_Classification_Model_for_Free_Health_Check-Up_Program_Analysis.pdf Download (3MB)

Abstract

Assesing public perception is vital for ensuring the effectiveness government program. This study analyzes public responses to the government’s Free Health Check-Up Program using data extracted from tweets on Platform X posted between October 20, 2024 and March 31, 2025. The study employs a multi-task classification approach in which the first task classifies tweets as either news or public service announcements, while the second task performs sentiment analysis. For comparison, single task and multi-class classification model were also developed as baselines, allowing a comprehensive evaluation across different strategies. The dataset was preprocessed using a combination of the NLTK method, custom-built dictionary, and the InSet Lexicon-Based method for automatic sentiment labeling. The classification model used was IndoBERT, which is a BERT-based model and was optimized using the Design of Experiment technique. The multi-task classification approach demonstrated optimal performance. Achieving Content type classification reached an accuracy of 87.05% and F1-score of 84.04%, while sentiment classification achieved an even higher accuracy of 89.76% and F1-score of 87.07%. The model was trained with dropout rate of 0.2, learning rate of 5e-5, batch size of 32, weight decay of 0.1, and 5 training epochs. These result demonstrate the potential of multi-task classification and the IndoBERT model for analyzing public sentiment on social media platforms.

Item Type:	Conference or Workshop Item (Paper)
Subjects:	Q Science > QA Mathematics
Divisions:	Fakultas Matematika dan Ilmu Pengetahuan Alam (FMIPA) > Prodi Matematika
Depositing User:	DIAN KURNIASARI
Date Deposited:	17 Apr 2026 02:42
Last Modified:	17 Apr 2026 02:42
URI:	http://repository.lppm.unila.ac.id/id/eprint/54836

Actions (login required)

View Item