ANALISIS SENTIMEN CHATGPT PADA TWITTER DENGAN MENGGUNAKAN ALGORITMA NAIVE BAYES CLASSIFIER (NBC) DENGAN LAPLACE ESTIMATOR

ADITYA NUGRAHA, . (2026) ANALISIS SENTIMEN CHATGPT PADA TWITTER DENGAN MENGGUNAKAN ALGORITMA NAIVE BAYES CLASSIFIER (NBC) DENGAN LAPLACE ESTIMATOR. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text File COVER_merged_compressed.pdf Download (377kB)
	Text File BAB 1.pdf Download (658kB)
	Text File BAB 2.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text File BAB 3_compressed.pdf Restricted to Registered users only Download (390kB) | Request a copy
	Text File BAB 4_compressed.pdf Restricted to Registered users only Download (665kB) | Request a copy
	Text File BAB 5_compressed.pdf Restricted to Registered users only Download (143kB) | Request a copy
	Text File Daftar Pustaka_compressed.pdf Download (180kB)
	Text File Lampiran_compressed.pdf Restricted to Registered users only Download (648kB) | Request a copy

Abstract

Perkembangan teknologi kecerdasan buatan, khususnya ChatGPT, telah memicu beragam respons masyarakat Indonesia yang banyak diungkapkan melalui media sosial Twitter/X. Analisis sentimen terhadap opini tersebut menjadi penting untuk memahami persepsi publik, namun klasifikasi teks menggunakan Naive Bayes Classifier (NBC) masih menghadapi kendala berupa zero probability problem dan ketidakseimbangan distribusi kelas yang dapat menurunkan kualitas prediksi. Penelitian ini bertujuan mengklasifikasikan sentimen tweet berbahasa Indonesia tentang ChatGPT ke dalam tiga kelas, yaitu Positif, Negatif, dan Netral, serta mengevaluasi pengaruh penerapan Laplace Estimator dan teknik resampling terhadap performa klasifikasi. Penelitian menggunakan 4.590 tweet yang diproses melalui tahapan preprocessing, pembobotan fitur menggunakan TF-IDF, serta penanganan ketidakseimbangan data menggunakan kombinasi Synthetic Minority Oversampling Technique (SMOTE) dan Random Undersampling. Evaluasi dilakukan pada empat konfigurasi model, yaitu NBC tanpa Laplace Estimator dan tanpa resampling, NBC dengan resampling, NBC dengan Laplace Estimator tanpa resampling, serta NBC dengan Laplace Estimator dan resampling. Hasil penelitian menunjukkan bahwa penerapan Laplace Estimator meningkatkan akurasi model pada data tanpa resampling dari 19,06% menjadi 72,66%. Namun, model tersebut masih menunjukkan kecenderungan memprediksi kelas mayoritas sehingga performanya kurang seimbang antarkelas. Sementara itu, model NBC dengan Laplace Estimator dan resampling dipilih sebagai model terbaik karena menghasilkan performa klasifikasi yang lebih seimbang dengan akurasi sebesar 64,72%, presisi 65,55%, recall 64,72%, dan F1-score 64,67%, serta F1-score antarkelas yang relatif merata (Negatif 0,67; Netral 0,61; Positif 0,66). Hasil penelitian menunjukkan bahwa kombinasi Laplace Estimator dan teknik resampling efektif dalam mengatasi zero probability problem serta mengurangi dampak ketidakseimbangan kelas, sehingga menghasilkan performa klasifikasi sentimen yang lebih seimbang pada data tweet berbahasa Indonesia. ***** The rapid advancement of artificial intelligence, particularly ChatGPT, has generated diverse public responses in Indonesia, many of which are expressed through the Twitter/X social media platform. Sentiment analysis is essential for understanding public perception; however, text classification using the Naive Bayes Classifier (NBC) is often challenged by the zero probability problem and class imbalance, both of which may reduce prediction performance. This study aims to classify Indonesian-language tweets about ChatGPT into three sentiment categories, namely Positive, Negative, and Neutral, while evaluating the impact of the Laplace Estimator and resampling techniques on classification performance. A dataset consisting of 4,590 tweets was processed through text preprocessing, TF IDF feature weighting, and class imbalance handling using a combination of the Synthetic Minority Oversampling Technique (SMOTE) and Random Undersampling. Four model configurations were evaluated: NBC without Laplace Estimator and without resampling, NBC with resampling, NBC with Laplace Estimator without resampling, and NBC with both Laplace Estimator and resampling. The results show that applying the Laplace Estimator increased the classification accuracy on the non-resampled dataset from 19.06% to 72.66%. However, the model remained biased toward the majority class, resulting in less balanced performance across sentiment categories. In contrast, the NBC model with Laplace Estimator and resampling was selected as the best-performing model because it achieved the most balanced classification performance, with an accuracy of 64.72%, precision of 65.55%, recall of 64.72%, and an F1-score of 64.67%, while producing relatively balanced class-wise F1-scores (Negative 0.67, Neutral 0.61, and Positive 0.66). These findings demonstrate that combining the Laplace Estimator with resampling effectively addresses the zero probability problem and mitigates the impact of class imbalance, thereby improving the overall balance of sentiment classification performance on Indonesian tweet datasets.

Item Type:	Thesis (Sarjana)
Additional Information:	1). Dr. Widodo, M.Kom. ; 2). Ressy Dwitias Sari, S.T., M.T.I.
Subjects:	Teknologi dan Ilmu Terapan > Teknik Komputer
Divisions:	FT > S1 Pendidikan Teknik Informatika Komputer
Depositing User:	Aditya Nugraha .
Date Deposited:	10 Aug 2026 07:32
Last Modified:	10 Aug 2026 07:32
URI:	http://repository.unj.ac.id/id/eprint/69102

Actions (login required)

View Item