PERBANDINGAN PENERAPAN METODE SVM DAN NAIVE BAYES PADA ANALISIS SENTIMEN TERHADAP KASUS ISU PERTAMINA BBM OPLOSAN

Monica Belinda; Mia kamayani;

Perbandingan Penerapan Metode SVM dan Naive Bayes Pada Analisis
Sentimen Terhadap Kasus Isu Pertamina BBM Oplosan

Monica Belinda

Isu mengenai dugaan pencampuran BBM oleh Pertamina menimbulkan beragam reaksi di media Isu
mengenai dugaan pencampuran BBM oleh Pertamina menimbulkan beragam reaksi di media sosial.
Penelitian ini bertujuan untuk mengklasifikasikan sentimen masyarakat dari Twitter terkait isu
tersebut, serta membandingkan kinerja algoritma Support Vector Machine (SVM) dan Naive Bayes.

Sebanyak 424 tweet digunakan setelah proses pengumpulan, pelabelan otomatis, dan pra-
pemrosesan teks (tokenisasi, stopword removal, dan stemming). Data kemudian diubah menjadi

vektor numerik menggunakan metode TF-IDF.Hasil klasifikasi menunjukkan bahwa SVM
memberikan akurasi lebih tinggi (92%) dibandingkan Naive Bayes (87%). SVM juga
menunjukkan performa lebih seimbang pada kedua kelas sentimen, terutama dalam mengenali opini
negatif. Berdasarkan hasil ini, SVM dinilai lebih efektif untuk analisis sentimen isu publik yang
bersifat sensitif seperti kasus BBM oplosan.
Kata kunci: Sentimen Publik, BBM Oplosan, Twitter, SVM, Naive Bayes, TF-IDF, Machine
Learning

Comparison of the Application of SVM and Naive Bayes Methods in Sentiment

Analysis of the Pertamina Mixed Fuel Issue Case

Monica Belinda

The issue of alleged fuel mixing by Pertamina has caused various reactions in the media The issue
of alleged fuel mixing by Pertamina has caused various reactions in social media. This study aims
to classify public sentiment from Twitter regarding the issue, and to compare the performance of
the Support Vector Machine (SVM) and Naive Bayes algorithms. A total of 424 tweets were used
after the collection process, automatic labeling, and text pre-processing (tokenization, stopword
removal, and stemming). The data was then converted into a numeric vector using the TF-IDF
method. The classification results showed that SVM provided higher accuracy (92%) compared to
Naive Bayes (87%). SVM also showed more balanced performance in both sentiment classes,
especially in recognizing negative opinions. Based on these results, SVM is considered more
effective for sentiment analysis of sensitive public issues such as the case of mixed fuel.
Keywords: Public Sentiment, Mixed BBM, Twitter, SVM, Naive Bayes, TF-IDF, Machine Learning