HERMAWAN, ALDI RAHMAN (2026) PERBANDINGAN SVM DAN HYBRID INDOBERTWEET-SVM UNTUK ANALISIS SENTIMEN PROGRAM MBG DI MEDIA SOSIAL. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
COVER.pdf Download (962kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
BAB1.pdf Restricted to Registered users only Download (98kB) |
|
|
Text (BAB II)
BAB2.pdf Restricted to Registered users only Download (331kB) |
|
|
Text (BAB III)
BAB3.pdf Restricted to Registered users only Download (265kB) |
|
|
Text (BAB IV)
BAB4.pdf Restricted to Registered users only Download (519kB) |
|
|
Text (BAB V)
BAB5.pdf Restricted to Registered users only Download (85kB) |
|
|
Text (DAFTAR PUSTAKA)
DAFTARPUSTAKA.pdf Restricted to Registered users only Download (91kB) |
|
|
Text (LMAPIRAN)
LAMPIRAN.pdf Restricted to Registered users only Download (227kB) |
Abstract
The Free Nutritious Meals Program is an Indonesian government policy that has attracted considerable public attention on social media. This study aims to analyze public sentiment toward the MBG Program and compare the performance of a Support Vector Machine (SVM) using Term Frequency–Inverse Document Frequency (TF-IDF) with a Hybrid IndoBERTweet–SVM model. Public sentiment is defined as the tendency of public opinion classified into positive, neutral, and negative categories, while model performance is assessed using accuracy, precision, recall, and F1-score. The research sample consists of 2,335 text data collected from Twitter/X and YouTube comments, comprising 545 positive, 569 neutral, and 1,221 negative instances. The data were selected using purposive sampling based on the relevance of posts and comments to the MBG Program. A quantitative comparative approach was employed through data collection, text cleaning, normalization, sentiment labeling, data splitting, model training, parameter optimization, and evaluation. In the first model, texts were represented using TF-IDF and classified with SVM. In the second model, contextual embeddings generated by IndoBERTweet were used as input features for SVM. The results show that the Hybrid IndoBERTweet–SVM achieved a test accuracy of 72.59% and a macro F1-score of 70.67%, whereas the TF-IDF-based SVM obtained a test accuracy of 61.88% and a macro F1-score of 59.58%. The hybrid model also produced higher F1-scores across all sentiment classes, although the neutral class remained the most difficult to classify. These findings indicate that IndoBERTweet contextual embeddings are more effective than TF-IDF weighting in representing Indonesian social media language and improving SVM performance in distinguishing positive, neutral, and negative sentiments. Keywords: Sentiment Analysis, Social Media, SVM, Transformer. Program Makan Bergizi Gratis adalah salah satu kebijakan pemerintah Indonesia yang banyak memperoleh perhatian masyarakat di media sosial. Penelitian ini bertujuan menganalisis sentimen publik terhadap Program MBG serta membandingkan kinerja model Support Vector Machine (SVM) berbasis Term Frequency–Inverse Document Frequency (TF-IDF) dengan model Hybrid IndoBERTweet–SVM. Sentimen publik dalam penelitian ini didefinisikan sebagai kecenderungan opini masyarakat yang dikelompokkan ke dalam kelas positif, netral, dan negatif, sedangkan kinerja model diukur menggunakan accuracy, precision, recall, dan F1-score. Sampel penelitian terdiri atas 2.335 data teks yang berasal dari Twitter/X dan komentar YouTube, dengan distribusi 545 data positif, 569 data netral, dan 1.221 data negatif. Data dipilih menggunakan teknik purposive sampling berdasarkan kesesuaian isi unggahan dan komentar dengan topik Program MBG. Analisis dilakukan menggunakan pendekatan kuantitatif komparatif melalui tahapan pengumpulan data, pembersihan teks, normalisasi, pelabelan, pembagian data, pelatihan model, optimasi parameter, dan evaluasi. Pada model pertama, teks direpresentasikan menggunakan TF-IDF dan diklasifikasikan dengan SVM, sedangkan pada model kedua, embedding kontekstual IndoBERTweet digunakan sebagai fitur masukan bagi SVM. Hasil penelitian menunjukkan bahwa model Hybrid IndoBERTweet–SVM memperoleh test accuracy sebesar 72,59% dan macro F1-score sebesar 70,67%, sedangkan model SVM berbasis TF-IDF memperoleh test accuracy sebesar 61,88% dan macro F1-score sebesar 59,58%. Model hybrid juga menghasilkan nilai F1-score yang lebih tinggi pada seluruh kelas sentimen, meskipun kelas netral tetap menjadi kategori yang paling sulit diklasifikasikan. Berdasarkan hasil tersebut, embedding kontekstual IndoBERTweet lebih efektif dibandingkan pembobotan TF-IDF dalam merepresentasikan bahasa media sosial Indonesia dan meningkatkan kemampuan SVM dalam membedakan sentimen positif, netral, dan negatif. Kata Kunci : Analisis Sentimen, Media Sosial, SVM, IndoBERTweet.
Actions (login required)
 |
View Item |
