Enhanced Naive Bayes Classifier for real-time sentiment analysis with SparkR
Young Gyo Jung, Kyung Tae Kim, Byungjun Lee, Hee Yong Youn · 2016
Correct and fast sentiment analysis of continuously generated data such as Twitter message is very important for providing real-time customized service to the users. While Naive Bayes Classifier(NBC) is the most popular classifier employed for sentiment analysis, the existing studies on it have been based on single server environment. Consequently, they are not adequate for handling real-time stream data. In this paper, thus, we propose a scheme adopting the Laplace Smoothing technique with Binarized NBC for enhancing the accuracy, and employing SparkR for speed-up via distributed and parallel processing. Computer simulation with Sentiment140 reveals that the proposed approach consistently allows higher accuracy than the existing schemes. It also identifies that the SparkR environment allows faster training than R.