Comparison of Accuracy between Convolutional Neural Networks and Naïve Bayes Classifiers in Sentiment Analysis on Twitter

Abstract

The needs and demands of the community for the ease of accessing information encourage the increasing use of social media tools such as Twitter to share, deliver and search for information needed. The number of large tweets shared by Twitter users every second, making the collection of tweets can be processed into useful information using sentiment analysis. The need for a large number of tweets to produce information encourages the need for a classifier model that can perform the analysis process quickly and provide accurate results. One algorithm that is currently popular and is widely used today to build classifier models is Deep Learning. Sentiment analysis in this research was conducted on English-language tweets on the topic Turkey Crisis 2018 by using one of the Deep Learning algorithms, Convolutional Neural Network (CNN). The resulting of CNN classifier model will then be compared with the Naive Bayes Classifier (NBC) classifier model to find out which classifier model can provide better accuracy in sentiment analysis. The research methods that will be carried out in this research are data retrieval, pre-processing, model design and training, model testing and visualization. The results obtained from this research indicate that the CNN classifier model produces an accuracy of 0.88 or 88% while the NBC classifier model produces an accuracy of 0.78 or 78% in the testing phase of the data test. Based on these results it can be concluded that the classifier model with Deep Learning algorithm produces better accuracy in sentiment analysis compared to the Naive Bayes classifier model.