Analisis Emosi pada Media Sosial Twitter Menggunakan Metode Multinomial Naive Bayes dan Synthetic Minority Oversampling Technique

Abstract

Twitter social media is often used to express one's emotions through tweets. Much research has been conducted on emotional analysis in the social media Twitter. Machine learning is a tool that is widely used to categorize emotions. However, an imbalance in the amount of data between classes is often a problem. So, this research aims to determine the performance of the combined Multinomial Naïve Bayes (MNB) and Synthetic Minority Oversampling Technique (SMOTE) methods for emotional analysis of tweets from the social media Twitter. Each tweet through data preprocessing in this research includes case folding, data cleaning, convert slangword, convert negation, tokenization, stopword removal, and stemming. For feature extraction the n-gram method is used and for feature weighting the term frequency method is used. Testing was carried out using K-Fold Cross Validation. Based on the test results, using SMOTE an average accuracy of 0.65 or 65% was obtained and an average f1-score value of 0.66 or 66%. Meanwhile, without SMOTE, an average accuracy of 0.64 or 64% was obtained and an average f1-score of 0.65 or 65%. Although in this study it can be shown that the results using SMOTE are 1% better in categorizing emotions. However, the results obtained are not optimal, and other methods of data balancing and machine learning still need to be studied.