Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/168630
Title: Analisis Sentimen Kebijakan Pembangunan Ibu Kota Nusantara di Media Sosial X Menggunakan Bidirectional Long Short-Term Memory
Other Titles: Analysis of the Nusantara Capital City Development Policy on Social Media X using Bidirectional Long Short Term Memory
Authors: Julianto, Mochamad Tito
Najib, Mohamad Khoirun
Nugraha, Wanda
Issue Date: 2025
Publisher: IPB University
Abstract: Pembangunan Ibu Kota Nusantara (IKN) merupakan salah satu terobosan kebijakan pemerintah yang memicu beragam respons publik sehingga menjadi penting untuk memahami bagaimana pandangan publik terhadap kebijakan ini. Tujuan penelitian ini mencakup mengeksplorasi kecenderungan sentimen, membangun model analisis sentimen menggunakan metode Bidirectional Long Short-Term Memory (BiLSTM) yang dioptimalkan melalui hyperparameter tuning, dan melakukan evaluasi terhadap kinerja model sebelum dan sesudah tuning. Temuan dari proses eksplorasi distribusi sentimen menunjukkan dominasi sentimen negatif sebesar 52%, diikuti netral sebesar 34%, dan positif 14%. Model awal yang telah dibangun menghasilkan akurasi uji sebesar 80% dan rata-rata akurasi sebesar 82% pada 10-fold cross validation. Meskipun akurasi uji tetap 80%, model hasil tuning mengalami peningkatan recall pada kelas negatif dan positif, yang penting dalam konteks analisis sentimen kebijakan publik. Rata-rata akurasi setelah tuning meningkat menjadi 84% pada 10-fold cross validation menunjukkan model memiliki kemampuan generalisasi yang lebih baik.
The development of the new capital city, Ibu Kota Nusantara (IKN) is a groundbreaking government policy that has triggered diverse public responses, making it crucial to understand how the public views this policy. The objectives of this study include exploring sentiment trends, building a sentiment analysis model using the Bidirectional Long Short-Term Memory (BiLSTM) method optimized through hyperparameter tuning, and evaluating the model's performance before and after tuning. Findings from the sentiment distribution exploration process indicate a dominance of negative sentiment at 52%, followed by neutral at 34%, and positive at 14%. The initial model that has been built produced a test accuracy of 80% and an average accuracy of 82% in 10-fold cross-validation. Although the test accuracy remained at 80%, the tuned model experienced increased recall in both the negative and positive classes, which is important in the context of public policy sentiment analysis. The average accuracy after tuning increased to 84% in 10-fold cross-validation, indicating the model has better generalization capabilities.
URI: http://repository.ipb.ac.id/handle/123456789/168630
Appears in Collections:UT - Mathematics

Files in This Item:
File Description SizeFormat 
cover_G5401211045_6f6ffa6692934bb0af742ddf63984fe1.pdfCover6.37 MBAdobe PDFView/Open
fulltext_G5401211045_20c28d723c4c4363a959e837e7f51439.pdf
  Restricted Access
Fulltext10.2 MBAdobe PDFView/Open
lampiran_G5401211045_2a67c71e6645479d8e82f21780347f94.pdf
  Restricted Access
Lampiran7.27 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.