Purwitasari, Yuliani (2026) KOMPARASI ALGORITMA TOPIC MODELING DAN MACHINE LEARNING UNTUK KLASIFIKASI SENTIMEN BERBASIS ASPEK PADA ULASAN APLIKASI PLN MOBILE. Undergraduate thesis, UPN Veteran Jawa Timur.
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (COVER)
22082010168-cover.pdf Download (1MB) |
|
Text (BAB 1)
22082010168-bab1.pdf Download (749kB) |
|
Text (BAB 2)
22082010168-bab2.pdf Restricted to Repository staff only until 5 March 2029. Download (886kB) |
|
Text (BAB 3)
22082010168-bab3.pdf Restricted to Repository staff only until 5 March 2029. Download (1MB) |
|
Text (BAB 4)
22082010168-bab4.pdf Restricted to Repository staff only until 5 March 2029. Download (7MB) |
|
Text (BAB 5)
22082010168-bab5.pdf Download (272kB) |
|
Text (DAFTAR PUSTAKA)
22082010168-daftarpustaka.pdf Download (243kB) |
|
Text (LAMPIRAN)
22082010168-lampiran.pdf Restricted to Repository staff only Download (676kB) |
Abstract
Advances in information and communication technology have accelerated the digitalization of public services, including the transformation of PT PLN (Persero) services through the PLN Mobile application. Although the application has received high ratings on the Google Play Store and App Store, negative reviews still indicate user-reported issues. The large volume of unstructured reviews requires an automated approach using Aspect-Based Sentiment Analysis (ABSA) to more specifically identify key aspects and sentiments. This study aims to evaluate topic modeling methods (LDA and BERTopic), compare the performance of SVM and Random Forest classification algorithms, and determine the best combination for aspect-based sentiment analysis of PLN Mobile reviews. Indonesian language reviews were collected through scraping, followed by preprocessing, topic modeling, and multilabel classification with various data splitting scenarios and the application of SMOTE-Tomek Links to address class imbalance. The modeling results show that LDA yields two key aspects: user experience and service quality, while BERTopic yields three key aspects: user experience, service quality, and system reliability. In terms of evaluation, BERTopic achieved the highest coherence score (c_v) of 0.6707 with four topics, outperforming LDA which achieved 0.5543 with two topics. However, in the final implementation, BERTopic generated three main topics based on its automatic clustering process. The best classification performance was obtained using LDA + Random Forest + SMOTE-Tomek Links (80:20 split), achieving a Macro F1-Score of 0.769 and an AUC of 0.85. The validation model achieved a Macro F1-Score of 0.739, a Macro AUC of 0.85, and a Hamming Loss of 0.058. The model is implemented into a website-based system that is able to predict aspects and sentiments through text input or CSV files and displays visualizations of distribution, proportion, and sentiment trends.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics 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: | YULIANI PURWITASARI PURWITASARI | ||||||||||||
| Date Deposited: | 06 Mar 2026 06:39 | ||||||||||||
| Last Modified: | 06 Mar 2026 06:50 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/50193 |
Actions (login required)
 |
View Item |