Optimization Sentiments of Analysis from Tweets in myXLCare using Naïve Bayes Algorithm and Synthetic Minority Over Sampling Technique Method
Dedi Dwi Saputra, Windu Gata, Nia Kusuma Wardhani, Ketut Sakho Parthama, Hendra Setiawan, Sularso Budilaksono, Dimas Yogatama, Agus Hadiyatna, Endah Putri Purnamasari, Bryan Pratama, Deny Novianti · Journal of Physics Conference Series · 2020
Abstract Text mining can be used to classify opinions about complaints or not complaints experienced by XL customers. This study aims to find and compare classifications in the sentiments of analysis from the view of XL customers. This dataset was derived from tweets of XL customers written on myXLCare Twitter account. In text mining techniques, “transform case”, “tokenize”, “token filters by length”, “n-gram”, “stemming” were used to build classification and sentiments of analysis. Gataframework tools were used to help during preprocessing and cleansing processes. RapidMiner is used to help create the sentiment of analysis to search and compare two different classifications methods between datasets using the Naïve Bayes algorithm only and Naïve Bayes algorithm with Synthetic Minority Over-sampling Technique (SMOTE). The results of the two methods in this study found that the highest results were using the Naïve Bayes algorithm with Synthetic Minority Over-sampling Technique (SMOTE) with an accuracy of 86.33%, precision 82.85%, and recall ratio 92.38%.