Comparison of Scenario Pre-processing Performance on Support Vector Machine and Naïve Bayes Algorithms for Sentiment Analysis

Abstract

Television shows need a rating in their assessment, but public opinion is also required to complete it. Sentiment analysis is necessary for its completion. An essential step in sentiment analysis is pre-processing because, in public opinion, there are still many inappropriate writings. This study aims to compare the performance results using different pre-processing scenarios to get the best pre-processing performance on Support Vector Machine (SVM) and Naïve Bayes (NB) on sentiment analysis about the television show X Factor Indonesia. The stages used to start from literature study, problem analysis, design, data collection, pre-processing with two scenarios, word weighting with TF-IDF, classification using SVM and NB, then resulting accuracy from Confusion Matrix. The findings of this research are that optimal performance can be achieved using a comprehensive pre-processing scenario. This scenario should include the following steps: case-folding, removing emoji, cleansing, removing repetition characters, word normalization, negation handling, stopwords removal, stemming, and tokenization, with an accuracy of 79.44% on the SVM algorithm. This research shows that the complete pre-processing of the SVM algorithm is better in terms of accuracy, precision, recall, and F1-score.