Skripsi

ANALISIS SENTIMEN PADA ULASAN APLIKASI MYPERTAMINA MENGGUNAKAN INDOBERT

---

The implementation of the MyPertamina application as a means of distributing subsidized fuel has triggered various responses from the public. User reviews on the Google Play Store platform serve as a vital data source that can be utilized to analyze public satisfaction. This study aims to classify the sentiment of MyPertamina application user reviews into positive and negative classes using the Fine-Tuning IndoBERT method. As a Transformer-based model pre-trained on an Indonesian corpus, IndoBERT possesses advantages in understanding complex semantic contexts. To maximize model performance, this research integrates a WeightedTrainer mechanism to address data imbalance challenges and applies Bayesian Optimization techniques through the Optuna framework to efficiently search for the most optimal hyperparameter combinations within a predefined search space. Based on experimental results from 20 trials, the best parameter configuration was obtained in Trial 16 with a combination including a learning rate of 2.60 x 10-5, batch size of 32, weight decay of 0.0449, adam epsilon of 1.77 x 10-8, warmup ratio of 0.1256, and a training duration of 2 epochs. Final evaluation of the model on the test data demonstrated excellent performance, achieving an accuracy rate of 92.98% and a weighted average F1-Score of 93.10%

---

Availability

Detail Information

Series Title

-

Call Number

T1936142026

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xxi, V-16 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.350 7

Content Type

Text

Media Type

unmediated

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Kecerdasan Buatan -- Analisis Sentimen

Specific Detail Info

-

Statement of Responsibility

KA

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment