Mandatory E-learning is a required training for Ministry of Finance’s employees through Kemenkeu Learning Center (KLC) as LMS, where text-based recapitulation reports for participant’s feedback are not available due to large volume of participant’s evaluation data. Sentiment analysis using text mining is necessary to classify the feedback into positive, negative, and neutral labels, enabling the recapitulation process to be automated, faster, and more accurate. Using Knowledge Discovery in Databases (KDD) framework, the process involves data selection and manual labeling, text preprocessing (data cleansing, case folding, stop word removal, stemming, tokenizing, filtering tokens by length), data transformation (TF-IDF weighting, cosine similarity measurement, and resampling using random undersampling/RUS to reduce majority label). Modeling phase compares the best combination of algorithms covers Support Vector Machine (SVM), Multinomial Naïve Bayes, K-Nearest Neighbor (KNN), and Random Forest using a 90:10 training-to-testing data ratio. This research show that SVM with cosine similarity is the best algorithm scenario, achieving accuracy, precision, recall, and f1-score for negative label of 97.01\%, 96.22\%, 95.82\%, and 96.02\%, respectively, within 48.71 seconds, which \textbf{can be leveraged} to improve quality of e-learning’s report faster, more accurate, and to be automated.