Komparasi Kinerja Algoritma Machine Learning Untuk Analisis Sentimen Multi-Kelas Pada Data Teks Digital Terhadap Kebijakan Pemerintah

Abstract

The development of social media and digital platforms has increased the number of public opinions on various government policies, including the Free Nutritional Meals (MBG) program. Digital text data generated by the public contains important information regarding public perceptions and sentiments, however  the characteristics of unstructured data and large data volumes pose challenges in the analysis process. In addition, the integrity of class sentiment can also affect the performance of model classification. This study aims to compare the performance of several Machine Learning algorithms, namely Naive Bayes, Support Vector Machine (SVM), Random Forest, Decision Tree, and K-Nearest Neighbors (KNN) in multi-class sentiment analysis classification on digital text data related to the MBG program. The dataset used is 4,969 text data from Kaggle in the form of public comments on social media related to government policies. The research method is through data collection, text preprocessing, TF-IDF feature weighting, model training, and evaluation using a confusion matrix, accuracy, precision, recall, and F1-score. The preprocessing stages include case folding, tokenizing, stopword removal, and stemming to clean and normalize the text data. Sentiment classification was performed into three categories: positive, negative, and neutral. The results showed that the Random Forest algorithm performed best with an accuracy of 0.81, compared to SVM (0.77), Decision Tree (0.75), Naive Bayes (0.71), and K-Nearest Neighbors (KNN) (0.52). Furthermore, Random Forest also produced higher precision, recall, and F1-score values for most sentiment classes, making it more effective in recognizing sentiment patterns in digital text data on government policies related to the MBG program.