<![CDATA[Abstrak - Ulasan pelanggan pada Google Maps dapat dimanfaatkan untuk mengetahui persepsi pelanggan terhadap kualitas produk dan layanan suatu perusahaan. Namun, banyaknya jumlah ulasan dan data yang tidak terstruktur menyebabkan proses analisis secara manual menjadi kurang efektif. Penelitian ini bertujuan untuk melakukan analisis sentimen pada ulasan pelanggan Aming Coffee menggunakan metode pembobotan Term Frequency–Inverse Document Frequency (TF-IDF) dan algoritma machine learning. Data penelitian diperoleh melalui proses scraping sebanyak 4.000 ulasan pelanggan dari Google Maps Aming Coffee. Tahapan penelitian meliputi preprocessing data yang terdiri atas cleaning, case folding, tokenisasi, normalisasi, stopword removal, dan stemming, kemudian dilakukan pembobotan TF-IDF, pembagian data, klasifikasi menggunakan algoritma Naïve Bayes, Support Vector Machine (SVM), dan Decision Tree C4.5, serta penerapan Synthetic Minority Over-sampling Technique (SMOTE) pada data latih untuk menangani ketidakseimbangan kelas. Evaluasi model dilakukan menggunakan accuracy, precision, recall, F1-score, dan confusion matrix. Hasil penelitian menunjukkan bahwa model tanpa SMOTE memperoleh akurasi tertinggi, yaitu Naïve Bayes sebesar 90,63%, diikuti SVM sebesar 90,50%, dan C4.5 sebesar 88,25%. Setelah penerapan SMOTE, akurasi model mengalami penurunan, namun nilai precision meningkat sehingga model menjadi lebih mampu memperhatikan kelas sentimen minoritas. Hasil Exploratory Data Analysis (EDA) menunjukkan bahwa sentimen positif mendominasi ulasan pelanggan Aming Coffee dengan kata yang paling sering muncul antara lain “kopi”, “coffee”, “enak”, “mantap”, dan “ramai”. Penelitian ini menunjukkan bahwa kombinasi TF-IDF, machine learning, dan SMOTE dapat digunakan untuk analisis sentimen ulasan pelanggan serta memberikan informasi yang bermanfaat dalam evaluasi kualitas layanan dan pengambilan keputusan bisnis. Kata kunci : Analisis Sentimen; TF-IDF; Naïve Bayes; Support Vector Machine; Decision Tree C4.5; SMOTE; Abstract - Customer reviews on Google Maps can be utilized to understand customer perceptions of a company's products and services. However, the large volume of unstructured reviews makes manual analysis less effective. This study aims to perform sentiment analysis on Aming Coffee customer reviews using the Term Frequency–Inverse Document Frequency (TF-IDF) weighting method and machine learning algorithms. Research data were collected through scraping 4,000 customer reviews from Google Maps. The research stages included data preprocessing consisting of cleaning, case folding, tokenization, normalization, stopword removal, and stemming, followed by TF-IDF weighting, data splitting, sentiment classification using Naïve Bayes, Support Vector Machine (SVM), and Decision Tree C4.5 algorithms, as well as the implementation of Synthetic Minority Over-sampling Technique (SMOTE) on training data to address class imbalance. Model evaluation was conducted using accuracy, precision, recall, F1-score, and confusion matrix. The results showed that models without SMOTE achieved the highest accuracy, with Naïve Bayes reaching 90.63%, followed by SVM at 90.50% and C4.5 at 88.25%. After applying SMOTE, model accuracy decreased, while precision increased, indicating improved attention to minority sentiment classes. Exploratory Data Analysis (EDA) results revealed that positive sentiment dominated Aming Coffee customer reviews, with frequently occurring words including “kopi,” “coffee,” “enak,” “mantap,” and “ramai.” This study demonstrates that the combination of TF-IDF, machine learning algorithms, and SMOTE can be effectively applied to sentiment analysis and provide useful insights for service quality evaluation and business decision-making. Keywords: Sentiment Analysis; TF-IDF; Naïve Bayes; Support Vector Machine; Decision Tree C4.5; SMOTE;]]>
