Komparasi Algoritma Topic Modeling Dalam Multi-Aspect Sentiment Analysis Ulasan Aplikasi Seabank Menggunakan Random Forest

Zabina, Keysya Alifia (2026) Komparasi Algoritma Topic Modeling Dalam Multi-Aspect Sentiment Analysis Ulasan Aplikasi Seabank Menggunakan Random Forest. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22082010187_Cover.pdf Download (2MB)
	Text (Bab 1) 22082010187_Bab 1.pdf Download (370kB)
	Text (Bab 2) 22082010187_Bab 2.pdf Restricted to Repository staff only until 15 September 2028. Download (859kB)
	Text (Bab 3) 22082010187_Bab 3.pdf Restricted to Repository staff only until 15 September 2028. Download (470kB)
	Text (Bab 4) 22082010187_Bab 4.pdf Restricted to Repository staff only until 15 September 2028. Download (2MB)
	Text (Bab 5) 22082010187_Bab 5.pdf Download (171kB)
	Text (Daftar Pustaka) 22082010187_Daftar Pustaka.pdf Download (409kB)
	Text (Lampiran) 22082010187_Lampiran.pdf Restricted to Repository staff only Download (1MB)

Abstract

Topic modeling is one approach that can be used to identify aspects or topics present in multiple texts. Advances in topic modeling methods have led to traditional approaches based on statistics and linear algebra, such as Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF), as well as modern approaches like BERTopic, which utilize embeddings and transformers. This study aims to compare LDA, NMF, and BERTopic in identifying aspects in reviews of the SeaBank app and to perform sentiment analysis on each identified aspect. Topic modeling was evaluated using coherence scores and topic diversity to assess the quality and diversity of the generated topics. Sentiment analysis was performed using the Random Forest algorithm with TF-IDF and Word2Vec feature representations. To address the issue of class imbalance, this study compared the baseline, SMOTETomek, and Class Weighted approaches. The results show that the BERTopic method produced the best topic quality, with a coherence score of 0.691327 and a topic diversity score of 0.95. In the sentiment analysis stage, Scenario 5, which uses TF-IDF and SMOTETomek. delivered the best performance with an accuracy of 0.912 and an F1 score of 0.576. These results indicate that the selected topic modeling approach is capable of identifying relevant aspects and can serve as a foundation for multi-aspect sentiment analysis in reviews of digital banking apps.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Permatasari, Reisa	NIDN0014059203	reisa.permatasari.sifo@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:

Keysya Alifia Zabina

Date Deposited:

15 Sep 2026 04:01

Last Modified:

15 Sep 2026 04:01

URI:

https://repository.upnjatim.ac.id/id/eprint/60242