PRATAMA, KEVIN (2026) PERBANDINGAN INDOBERT-BASE, INDOBERTWEET, DAN XLM-R UNTUK KLASIFIKASI SENTIMEN POLITIK INDONESIA MENGGUNAKAN PSEUDO-LABELING. S1 thesis, Universitas Mercu Buana Jakarta.
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (JURNAL MAHASISWA)
Jurnal 41522010042 Kevin Pratama - KEVIN PRATAMA.pdf Restricted to Registered users only Download (2MB) |
Abstract
This study compared IndoBERT-base, IndoBERTweet, and XLM-R for sentiment classification of Indonesian Twitter/X replies to Joko Widodo’s posts using supervised learning and weak supervision. The initial dataset consisted of 62,615 comments collected from March 1, 2024 to October 23, 2025. After preprocessing, 48,549 valid records were obtained, consisting of 8,778 manually labeled and 39,771 unlabeled replies. Phase A used manually labeled data to train and evaluate the models, while Phase B expanded the training set through pseudo-labeling using the best Phase A model. Model performance was evaluated using macro-F1 as the primary metric, supported by accuracy, weighted-F1, and per-class F1. The results showed that IndoBERTweet achieved the best performance in Phase A with a macroF1 score of 0.7586 and remained the best-performing model in Phase B with a macro-F1 score of 0.7637. Bootstrap tests confirmed that IndoBERTweet significantly outperformed IndoBERT-base and XLM-R in Phase A. Pseudo-labeling descriptively improved macro-F1 across all models, but the Phase A to Phase B improvements were not statistically significant. The NEUTRAL class remained the most challenging due to short text, ambiguity, slang, sarcasm, and limited conversational context. Keywords: Sentiment Analysis, Twitter/X, IndoBERT, IndoBERTweet, XLM-R, pseudo-labeling Penelitian ini membandingkan IndoBERT-base, IndoBERTweet, dan XLM-R untuk klasifikasi sentimen balasan Twitter/X berbahasa Indonesia terhadap unggahan Joko Widodo menggunakan supervised learning dan weak supervision. Dataset awal terdiri atas 62.615 komentar yang dikumpulkan pada rentang 1 Maret 2024 hingga 23 Oktober 2025. Setelah preprocessing, diperoleh 48.549 data valid, terdiri atas 8.778 data berlabel manual dan 39.771 data tidak berlabel. Phase A menggunakan data berlabel manual untuk melatih dan mengevaluasi model, sedangkan Phase B memperluas data latih melalui pseudo-labeling menggunakan model terbaik dari Phase A. Evaluasi dilakukan menggunakan macro-F1 sebagai metrik utama, serta accuracy, weighted-F1, dan F1-score per kelas sebagai metrik pendukung. Hasil penelitian menunjukkan bahwa IndoBERTweet memperoleh performa terbaik pada Phase A dengan macro-F1 sebesar 0,7586 dan tetap menjadi model terbaik pada Phase B dengan macro-F1 sebesar 0,7637. Uji bootstrap menunjukkan bahwa IndoBERTweet secara signifikan mengungguli IndoBERT-base dan XLM-R pada Phase A. Pseudo-labeling meningkatkan macro-F1 seluruh model secara deskriptif, tetapi peningkatan dari Phase A ke Phase B belum signifikan secara statistik. Kelas NEUTRAL tetap menjadi kelas paling sulit karena teks pendek, ambigu, slang, sarkasme, dan keterbatasan konteks. Kata Kunci : Analisis Sentimen, Twitter/X, IndoBERT, IndoBERTweet, XLM-R, pseudo-labeling.
| Item Type: | Thesis (S1) |
|---|---|
| NIM/NIDN Creators: | 41522010042 |
| Uncontrolled Keywords: | Analisis Sentimen, Twitter/X, IndoBERT, IndoBERTweet, XLM-R, pseudo-labeling. |
| Subjects: | 000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 004 Data Processing, Computer Science/Pemrosesan Data, Ilmu Komputer, Teknik Informatika 000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 006 Special Computer Methods/Metode Komputer Tertentu > 006.7 Multimedia Systems/Sistem-sistem Multimedia > 006.75 Social Multimedia/Multimedia Social |
| Divisions: | Fakultas Ilmu Komputer > Informatika |
| Depositing User: | khalimah |
| Date Deposited: | 26 Aug 2026 04:29 |
| Last Modified: | 26 Aug 2026 04:29 |
| URI: | http://repository.mercubuana.ac.id/id/eprint/103413 |
Actions (login required)
 |
View Item |
