MULTILABEL ASPECT-BASED SENTIMENT ANALYSIS PADA DATA ULASAN MULTIBAHASA APLIKASI FLO PERIOD TRACKER MENGGUNAKAN MULTILINGUAL BERT DAN BERTOPIC

AFFAH, RIFDA NASYWATUL (2026) MULTILABEL ASPECT-BASED SENTIMENT ANALYSIS PADA DATA ULASAN MULTIBAHASA APLIKASI FLO PERIOD TRACKER MENGGUNAKAN MULTILINGUAL BERT DAN BERTOPIC. Undergraduate thesis, UPN "Veteran" Jawa Timur.

	Text 22082010009.-cover.pdf Download (1MB)
	Text (Bab 1) 22082010009.-bab1.pdf Download (494kB)
	Text (Bab 2) 22082010009.-bab2.pdf Restricted to Repository staff only until 5 March 2029. Download (596kB)
	Text (Bab 3) 22082010009.-bab3.pdf Restricted to Repository staff only until 5 March 2029. Download (671kB)
	Text (Bab 4) 22082010009.-bab4.pdf Restricted to Repository staff only until 5 March 2029. Download (2MB)
	Text (Bab 5) 22082010009.-bab5.pdf Download (236kB)
	Text (Daftar Pustaka) 22082010009.-daftarpustaka.pdf Download (261kB)
	Text (Lampiran) 22082010009.-lampiran.pdf Restricted to Repository staff only until 5 March 2029. Download (334kB)

Abstract

The development of the FemTech industry has increased the use of women's health applications such as Flo Period Tracker. Even though it has a high rating, there are still negative reviews that show dissatisfaction with certain aspects. This study aims to apply Aspect-Based Sentiment Analysis (ABSA) to multilingual reviews to identify aspects and sentiment trends. Aspect extraction was performed using BERTopic and resulted in six main aspects, including user experience, cycle tracking accuracy, app subscription, app usefulness, data management, and cycle period changes. Sentiment classification was performed using Multilingual BERT (mBERT) into three categories: positive, negative, and neutral. The experiment involved variations in data splitting (70:30, 80:20, 90:10), batch size (16 and 32), and data imbalance handling methods in the form of Cross Entropy Loss, Focal Loss, and Random Oversampling combined with Balanced Bagging Classifier (ROS BBC). The best results were obtained in the ROS BBC scenario with a 70:30 split and a batch size of 16 because it was able to provide balanced and more stable performance in handling minority classes, with an accuracy of 0.87, precision of 0.64, recall of 0.72, f1-score of 0.67, and AUC of 0.93. The model was then implemented in a web-based system for automatic aspect-based sentiment analysis.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Wahyuni, Eka Dyar	NIDN0001128406	ekawahyuni.si@upnjatim.ac.id
Thesis advisor	Sugata, Tri Luhur Indayanti	NIDN8948770671230422	tri.luhur.fasilkom@upnjatim.ac.id

Subjects:

Q Science > Q Science (General)
Q Science > QA Mathematics > QA76.6 Computer Programming
T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Information Systems

Depositing User:

Rifda Nasywatul

Date Deposited:

06 Mar 2026 03:48

Last Modified:

06 Mar 2026 07:21

URI:

https://repository.upnjatim.ac.id/id/eprint/50192