Comparison of Word2vec and CountVectorizer with Mutual Information in Support Vector Machine (SVM) for Public Sentiment Analysis

Social media is widely used today. Along with the development of social media, it makes it not only a means of communication but also a means of exchanging opinions. One of the social media that is widely used to exchange opinions is X (Twitter). X is widely used to express opinions, particularly on controversial issues, such as the relocation of IKN. Therefore, sentiment analysis is needed to analyse public opinion regarding this national issue. SVM is widely used to classify sentiment based on several required categories, such as positive or negative. However, SVM will work even more effectively if the features used have good quality. Therefore, feature extraction and selection are necessary to enhance SVM classification accuracy. The selection of appropriate feature extraction is very important for classification. Therefore, this study aims to compare two feature extractions, namely Word2Vec and CountVectorizer by adding Mutual Information feature selection to SVM in classifying public sentiment from X. The results show that SVM with Word2Vec and CountVectorizer is more effective than SVM with Mutual Information feature selection. The results show that SVM with Word2Vec feature extraction and Mutual Information feature selection is more effective overall with 84% accuracy, 90% precision, 90% recall, and 90% f1-score, compared to SVM with CountVectorizer feature extraction and Mutual Information feature selection which has 80% accuracy, 83% precision, 92% recall, and 87% f1-score.