Twitter sentiment classification using Naive Bayes based on trainer perception

Abstract

This paper presents strategy to classify tweets sentiment using Naive Bayes techniques based on trainers' perception into three categories; positive, negative or neutral. 50 tweets of 'Malaysia' and 'Maybank' keywords were selected from Twitter for perception training. In this study, there were 27 trainers participated. Each trainer was asked to classify the sentiment of 25 tweets of each keyword. Results from the classification training was then be used as the input for Naive Bayes training for the remaining 25 tweets. The trainers were then asked to validate the results of sentiment classification by the Naive Bayes technique. The accuracy of this study is 90% ± 14% measured by total number of correct per total classified tweets. © 2015 IEEE.