Electronic Theses and Dissertation

TikTok merupakan salah satu platform media sosial yang sangat populer di Indonesia
dan banyak digunakan untuk promosi produk, termasuk produk skincare seperti
sunscreen. Banyaknya komentar pengguna pada video TikTok menjadi sumber data
yang potensial untuk dianalisis guna memahami sentimen publik terhadap suatu merek.
Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi otomatis yang
mampu mengelompokkan komentar TikTok ke dalam tiga kategori sentimen, yaitu
positif, netral, dan negatif. Data komentar dikumpulkan melalui proses web scraping
dari enam brand sunscreen populer, yakni Wardah, Azarine, Skintific, The Originote,
Madame Gie, dan Facetology. Total data yang digunakan berjumlah 13.913 komentar,
yang telah melalui proses pra-pemrosesan dan augmentasi data menggunakan teknik
backtranslation untuk menangani ketidakseimbangan kelas. Pelabelan komentar
dilakukan secara otomatis menggunakan model Generative AI (GPT-4o mini) yang
kemudian divalidasi secara manual. Teks komentar dikonversi menjadi vektor numerik
menggunakan metode TF-IDF sebelum digunakan untuk pelatihan tiga algoritma
klasifikasi, yaitu Support Vector Machine (SVM), Naive Bayes (MultinomialNB), dan
Random Forest. Proses tuning hyperparameter dilakukan dengan GridSearchCV untuk
mendapatkan performa optimal. Hasil evaluasi menunjukkan bahwa model SVM dan
Random Forest memiliki performa terbaik dengan akurasi masing-masing 80,78% dan
80,74%, serta F1-score sebesar 0.8062 dan 0.8061. Sementara itu, model Naive Bayes
menunjukkan hasil F1-score sebesar 0.6378. Hasil klasifikasi ditampilkan dalam
bentuk visualisasi interaktif berbasis Streamlit, seperti pie chart dan word cloud
berdasarkan masing-masing sentimen. Penelitian ini menunjukkan bahwa kombinasi
teknik preprocessing, augmentasi data, dan tuning hyperparameter dapat meningkatkan
efektivitas model klasifikasi dalam menganalisis sentimen komentar TikTok. Sistem
yang dikembangkan berpotensi menjadi alat bantu bagi pelaku industri maupun peneliti
untuk memahami persepsi konsumen terhadap suatu produk secara efisien.
Kata kunci : TikTok, Sentimen, Web Scraping, TF-IDF, SVM, Naive Bayes, Random
Forest, Augmentasi

TikTok is one of the most popular social media platforms in Indonesia and is widely used for product promotion, including skincare products such as sunscreen. The large volume of user comments on TikTok videos serves as a valuable data source for analyzing public sentiment toward a particular brand. This study aims to develop an automatic classification system capable of categorizing TikTok comments into three sentiment classes: positive, neutral, and negative. Comment data was collected through a web scraping process from six popular sunscreen brands: Wardah, Azarine, Skintific, The Originote, Madame Gie, and Facetology. A total of 13,913 comments were used, which underwent preprocessing and data augmentation using the backtranslation technique to address class imbalance. The labeling process was conducted automatically using a Generative AI model (GPT-4o mini), and manually validated to ensure label accuracy. The text comments were transformed into numerical vectors using the TF-IDF method before being used to train three classification algorithms: Support Vector Machine (SVM), Naive Bayes (MultinomialNB), and Random Forest. Hyperparameter tuning was performed using GridSearchCV to achieve optimal model performance. Evaluation results showed that the SVM and Random Forest models achieved the best performance with accuracy scores of 80.78% and 80.74%, and F1-scores of 0.8062 and 0.8061, respectively. Meanwhile, the Naive Bayes model yielded an F1-score of 0.6378. The classification results were visualized interactively using a Streamlit-based application, including pie charts and word clouds for each sentiment category. This study demonstrates that the combination of preprocessing techniques, data augmentation, and hyperparameter tuning can significantly improve the performance of sentiment classification models in analyzing TikTok comments. The developed system has the potential to be a useful tool for industry practitioners and researchers in understanding consumer perceptions of a product efficiently. Keywords : TikTok, Sentiment, Web Scraping, TF-IDF, SVM, Naive Bayes, Random Forest, Augmentation