Optimizing Text Preprocessing for Accurate Sentiment Analysis on E-Wallet Reviews

This Research aims to optimize preprocessing techniques in sentiment analysis of reviews for the E-Wallet Dana application on the Google Play Store. Text preprocessing is a crucial step in Natural Language Processing (NLP) that affects the accuracy and efficiency of sentiment analysis. This study employs various preprocessing methods, including stopwords removal, stemming, and lemmatization, to clean and prepare the review data before analysis. The results show that lemmatization techniques significantly improve accuracy compared to basic preprocessing techniques such as stopwords removal and stemming. With proper preprocessing optimization, sentiment analysis can provide more accurate and informative results, which can be used to enhance the application's quality and user experience. This study uses SVM classification testing models with 4 kernels, where the highest results were achieved with cleaning, case folding, tokenization, and lemmatization techniques at 100% for Linear; 100% for RBF, 99% for Polynomial, and 99.50% for Sigmoid with an average accuracy of 99.63%. Highlights: Preprocessing Optimization: Lemmatization significantly improves sentiment analysis accuracy compared to basic techniques like stemming or stopword removal. High SVM Accuracy: The best preprocessing combination achieved an average accuracy of 99.63% across multiple SVM kernels, with linear and RBF kernels reaching 100%. Real-World Dataset: Analysis used 1000 authentic reviews from Google Play Store, highlighting practical insights for improving E-Wallet services like DANA. Keywords: DANA, Google Play Store, Preprocessing, Sentiment Analysis