Aryaputra, Al-Faiz Azzam (2026) Komparasi Kinerja IndoBERT dan IndoRoBERTa untuk Analisis Emosi dan Sentimen Berbasis Topik Ulasan pada Aplikasi Kedai Kopi Terpopuler di Indonesia. Undergraduate thesis, UPN Veteran Jawa Timur.
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Cover)
22082010022.-cover.pdf Download (1MB) |
|
Text (Bab 1)
22082010022.-bab1.pdf Download (395kB) |
|
Text (Bab 2)
22082010022.-bab2.pdf Restricted to Repository staff only until 17 July 2029. Download (687kB) |
|
Text (Bab 3)
22082010022.-bab3.pdf Restricted to Repository staff only until 17 July 2029. Download (4MB) |
|
Text (Bab 4)
22082010022.-bab4.pdf Restricted to Repository staff only until 17 July 2029. Download (8MB) |
|
Text (Bab 5)
22082010022.-bab5.pdf Download (236kB) |
|
Text (Daftar pustaka)
22082010022.-daftarpustaka.pdf Download (211kB) |
|
Text (Lampiran)
22082010022.-lampiran.pdf Restricted to Repository staff only Download (1MB) |
Abstract
User reviews of coffee shop applications can contain more than one aspect within a single text, and each aspect can have more than one emotion as well as one specific sentiment. This condition requires an approach capable of analyzing each aspect separately. This study aimed to identify aspects in user reviews, analyze the performance of IndoBERT and IndoRoBERTa, determine the best model, and deploy it into an analysis system. The study employed the CRISP-DM framework using 8,018 reviews collected from five coffee shop applications on the Google Play Store. Topics in the reviews were extracted using Latent Dirichlet Allocation and subsequently interpreted as aspects, while aspect, emotion, and sentiment labels were determined through majority voting from nine annotators. IndoBERT and IndoRoBERTa were trained to perform multi-label emotion classification and aspect-based sentiment analysis, then evaluated using k-fold cross validation and test data. LDA in the stemmed scenario produced three topics with a coherence score of 0.4612 that were interpreted as the Order and Fulfillment Process, Product Quality and Outlet Experience, and Customer Digital Experience aspects. On the test data, IndoBERT with class weighting and a 70:30 data split achieved an average macro precision of 0.7254, macro recall of 0.7860, macro F1-score of 0.7334, and accuracy of 0.6096. The macro F1-score was higher than that of IndoRoBERTa, which achieved 0.7289. The IndoBERT model was then deployed into a system capable of accepting text input or CSV files and displaying analysis results in the form of tables and visualizations. The results showed that IndoBERT delivered the best performance in aspect-based emotion and sentiment analysis on the dataset used.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics > QA76.6 Computer Programming Q Science > QA Mathematics > QA76.87 Neural computers T Technology > T Technology (General) |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Information Systems | ||||||||||||
| Depositing User: | Al-Faiz Azzam Aryaputra | ||||||||||||
| Date Deposited: | 20 Jul 2026 02:07 | ||||||||||||
| Last Modified: | 20 Jul 2026 03:10 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/56034 |
Actions (login required)
 |
View Item |