Penerapan Fine-Tuning IndoBERT Untuk Aspect-Based Sentiment Analysis (ABSA) Pada Ulasan Pengguna Aplikasi JAKI

Faradillah, Safna (2026) Penerapan Fine-Tuning IndoBERT Untuk Aspect-Based Sentiment Analysis (ABSA) Pada Ulasan Pengguna Aplikasi JAKI. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22082010095-cover.pdf Download (492kB)
	Text (Bab 1) 22082010095-bab1.pdf Download (94kB)
	Text (Bab 2) 22082010095-bab2.pdf Restricted to Repository staff only until 20 July 2029. Download (539kB)
	Text (Bab 3) 22082010095-bab3.pdf Restricted to Repository staff only until 20 July 2029. Download (250kB)
	Text (Bab 4) 22082010095-bab4.pdf Restricted to Repository staff only until 20 July 2029. Download (6MB)
	Text (Bab 5) 22082010095-bab5.pdf Download (14kB)
	Text (Daftar Pustaka) 22082010095-daftarpustaka.pdf Download (160kB)
	Text (Lampiran) 22082010095-lampiran.pdf Restricted to Repository staff only until 20 July 2029. Download (144kB)

Abstract

Along with the rapid advancement of digital transformation, the adoption of digital platforms in public services has also increased, one of which is the JAKI application. The JAKI (Jakarta Kini) is a digital public service application developed by the Provincial Government of DKI Jakarta that has received thousands of diverse user reviews on the Google Play Store. Therefore, aspect-based sentiment analysis is necessary to gain a deeper understanding of user opinions regarding specific aspects. This study employs a fine-tuned IndoBERT model, utilizing the ABSA approach, to analyze JAKI app reviews on the Google Play Store, with a focus on predefined aspects: functional, service, and usability, and classifying sentiments into positive, negative, and neutral categories. The modeling process was conducted by comparing the performance of the XGBoost model and fine-tuned IndoBERT model, combined with hyperparameter optimization scenarios, two data split ratios (70:15:15 and 80:10:10), and three data imbalance techniques, namely Near Miss, SMOTE, and Class Weighting. The evaluation results indicate that the fine-tuned IndoBERT model with a learning rate of 2×10⁻⁵, a batch size of 16, and 10 epochs, combined with the Class Weighting technique using the 70:15:15 data split ratio, achieved the best performance, yielding the highest average macro F1-score. Specifically, the service and usability aspects each achieved a macro F1-score of 76%, while the functional aspect achieved 49%, due to the extremely severe class imbalance. The results of this study indicate that the selection of appropriate parameters, data splitting ratios, and techniques for handling imbalanced data can have a significant impact on the performance of the fine-tuned IndoBERT model in conducting aspect-based sentiment analysis on reviews of the JAKI app.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Najaf, Abdul Rezha Efrat	NIDN0029099403	rezha.efrat.sifo@upnjatim.ac.id
Thesis advisor	Pahlawan, Muhammad Reza	NIDN0716059801	muhammad_reza.si@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.6 Computer Programming
T Technology > T Technology (General)
T Technology > T Technology (General) > T58.6-58.62 Management Information Systems

Divisions:

Faculty of Computer Science > Departemen of Information Systems

Depositing User:

safna faradillah -

Date Deposited:

20 Jul 2026 04:38

Last Modified:

20 Jul 2026 06:46

URI:

https://repository.upnjatim.ac.id/id/eprint/56456