Two simple and effective ensemble classifiers for Twitter sentiment analysis
Yeqing Yan, Hui Qiong Yang, Hui‐Ming Wang · 2017 Computing Conference · 2017
This article presents two simple yet effective ensemble learners for Twitter data sentiment classification. These two learners are built upon four off-the-shelf classifiers, namely, Mallet's MaxEnt and Naïve Bayes, SentiStrength, and Pattern from Textblob. Extensive evaluation of these algorithms on twelve Twitter datasets demonstrates that these two ensemble algorithms constantly deliver top classification accuracy, followed by MaxEnt and Naïve Bayes. Moreover, both ensemble learners are embarrassingly parallel and can be easily extended to handle huge amount of Twitter data. This article also presents a potential solution to the problem of lacking sufficient training data. Specifically, to effectively classify the tweets associated with product A, one can use tweets associated with other products in the same category of product A to train a classification model. This suggests the homogeneity of tweets from similar products in the context of sentiment classification.