Article Per Year (5 Year)
p-Index From 2021 - 2026
0.408
P-Index
This Author published in this journals
All Journal SEMINAR TEKNOLOGI MAJALENGKA (STIMA)
Budiman
Fakultas Teknik, Universitas Majalengka
Computer Science & IT Control & Systems Engineering Industrial & Manufacturing Engineering Mechanical Engineering Transportation

Published : 2 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 2 Documents
Search

 1 
INTEGRASI FINE-TUNED INDOBERT DAN BERTOPIC UNTUK SENTIMENT-TOPIC MODELING OPINI PUBLIK PEMBANGUNAN IKN Wiwin Kartini; Harun Sujadi; Budiman
SEMINAR TEKNOLOGI MAJALENGKA (STIMA) Vol 10 (2026)
Publisher : Universitas Majalengka

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Using natural language processing methods, this study investigates public perceptions of the development of the Indonesian Capital City (IKN) on social media X. Previous studies predominantly relied on isolated sentiment classification or standalone topic modeling, which failed to explain the specific underlying context behind public sentiment. To address this gap, this study proposes a novel two-stage framework integrating Fine-Tuned IndoBERT (optimized via Low-Rank Adaptation/LoRA) for sentiment classification and BERTopic for topic extraction. A total of 12,000 raw comments were harvested via web crawling, preprocessed, validated through automated exact-match deduplication and noise filtering, and labeled into positive, negative, and neutral classes. Performance evaluation using Confusion Matrix and classification metrics revealed that Fine-Tuned IndoBERT achieved optimal performance with an accuracy of 91.83%, a macro precision of 91.84%, a macro recall of 91.54%, and a macro F1-score of 91.93%. The integration with BERTopic configured using Sentence Transformer, UMAP, HDBSCAN, and class-based TF-IDF (c-TF-IDF) components successfully extracted nine main topic clusters and mapped them against sentiment categories. The results confirm that positive sentiments dominate physical development, logistics, and digital transformation topics, whereas negative sentiments concentrate heavily on environmental impacts, budget transparency, and indigenous rights. This study provides an actionable, integrative decision-support model for public policy analysis.
ANALISIS SENTIMEN ULASAN PENGGUNA APLIKASI IDENTITAS KEPENDUDUKAN DIGITAL MENGGUNAKAN ALGORITMA NAIVE BAYES DENGAN PEMBOBOTAN TF-IDF Aprianti Safarani Aprianti Safarani; Budiman; Dadan Zaliluddin
SEMINAR TEKNOLOGI MAJALENGKA (STIMA) Vol 10 (2026)
Publisher : Universitas Majalengka

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Research on the Digital Population Identity (IKD) application found a dominant negative sentiment of 74.81% based on an analysis of 10.308 clean reviews from 10,842 raw data collected through web scraping on the Google Play Store from January 2025 to March 2026. This indicates that the level of public satisfaction is still low. Technical obstacles during the registration process and system stability that is often disrupted after application updates are the main user concerns. This study successfully created a sentiment classification model with an accuracy of 87.88% and a negative class recall of 0.99 using the Naive Bayes algorithm and Frequency Term-Inverse Document Frequency (TF-IDF) weighting. The results of the word feature extraction recommend that the registration system architecture should be thoroughly evaluated, each software update should be thoroughly tested, and server capacity should be increased to provide a more stable and efficient user experience in the future in population administration services.
Co-Authors Aprianti Safarani Aprianti Safarani Dadan Zaliluddin Harun Sujadi Wiwin Kartini