Pradipta, Arsa Cahaya (2026) Klasifikasi Emosi dan Pemodelan Topik pada Dataset Multi-Bahasa Ulasan Aplikasi Provider Telekomunikasi Menggunakan mBERT dan BERTopic. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
22082010015.-cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
22082010015.-bab1.pdf Download (285kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
22082010015.-bab2.pdf Restricted to Repository staff only until 26 May 2029. Download (674kB) |
|
|
Text (Bab 3)
22082010015.-bab3.pdf Restricted to Repository staff only until 26 May 2029. Download (329kB) |
|
|
Text (Bab 4)
22082010015.-bab4.pdf Restricted to Repository staff only until 26 May 2029. Download (6MB) |
|
|
Text (Bab 5)
22082010015.-bab5.pdf Download (119kB) | Preview |
|
|
Text (Daftar pustaka)
22082010015.-daftarpustaka.pdf Download (137kB) | Preview |
|
|
Text (Lampiran)
22082010015.-lampiran.pdf Restricted to Repository staff only Download (714kB) |
Abstract
The growth of digital telecommunication services leads to high user interaction through application reviews. Despite the very high number of downloads, many reviews contain technical complaints written using code-mixing and informal language. This study aims to apply emotion classification and topic modelling to detect emotions and complaint topics of telecommunication provider application users, using a review dataset from three applications such as MyTelkomsel, MyIM3, and MyXL collected from the Google Play Store. Two main approaches were used, the mBERT transformer model for classifying six basic emotions and the BERTopic model for problem topic modelling. The results showed that the mBERT model with normalization preprocessing scenario and hyperparameter tuning achieved the best performance. The mBERT model demonstrated an Accuracy of 0.72 and a Macro F1-Score of 0.59, and was more sensitive in handling class imbalance compared to resampling techniques. In topic modelling, BERTopic with a minimum cluster size of 17 produced a Coherence score of 0.6175 and a Stability score of 0.6253, identifying the dominance of anger and sadness on main topics such as app performance, pricing, and balance deduction. Furthermore, the web-based system developed using Flask was able to classify emotions and topics, supporting both single text and CSV file inputs.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA76.6 Computer Programming Q Science > QA Mathematics > QA76.87 Neural computers |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Information Systems | ||||||||||||
| Depositing User: | Arsa Cahaya Pradipta | ||||||||||||
| Date Deposited: | 26 May 2026 02:29 | ||||||||||||
| Last Modified: | 26 May 2026 02:29 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/51992 |
Actions (login required)
 |
View Item |