Social Media Sentiment Analysis Using K-Means and Naïve Bayes Algorithm

摘要

Opinions are a major influence when making decisions for individuals or organizations. A collection of opinions can be extracted to gain useful knowledge. This knowledge is used as a source of information which can be used as a consideration in decision making. The extraction of knowledge from text has been known as text mining. Text mining has any kinds of algorithm to extract information from collected text, such as K-Means, K-Nearest Neighbors, Naïve Bayes, and the others. One of the sources of opinion is from social media, especially Facebook and Twitter. On Facebook and Twitter, many people have been writing their opinions about many things. This very much data are difficult to analyze thoroughly. In this paper, K-Means and Naïve Bayes algorithm are developed to analyze public opinions or sentiments. Outlier removal is also added to this analysis. Opinions are taken from Facebook and Twitter. The accuracy of the system is tested 10 times at k different points for each k value (k=6, 7, 8, 9 and 10). As the result, the combination of K-Means and Naïve Bayes has lower accuracy than the accuracy produced by Naïve Bayes without the combination of K-Means, but almost same accuracies. The accuracy of Naïve Bayes algorithm is from 80.526%-82.500%, while the combination of Naïve Bayes and K-Means has 80.323%-81.523% accuracy.