Portal de Periódicos da CAPES

Hybrid Model using Stack-Based Ensemble Classifier and Dictionary Classifier to Improve Classification Accuracy of Twitter Sentiment Analysis

2020; Volume: 8; Issue: 7 Linguagem: Inglês

10.30534/ijeter/2020/02872020

2347-3983

Sangeeta Rani,

Text and Document Classification Technologies

Ensemble classifiers are widely used for the enhancement of accuracy of twitter sentiment classification.In the present research, a hybrid model based on stack based ensemble classifiers and dictionary based classifier is used for tweet classification as positive and negative.To enhance accuracy of classification, sentiment score retrieved from dictionary based classifier is added to the feature vector to get enhanced feature set and the hybrid stack based ensemble model is implemented on this enhanced feature set.Three machine learning classifiers svmRadial, C5.0, NB are used to build stacked based ensemble classifier using GLM and RF as Meta learners.Three data sets viz.Kaggle -US Airline Twitter Sentiment Data Set, Sentiment 140 Twitter Data Set, and Real time manually labeled data set related to 'Clean India Mission' are used for the implementation of the proposed model.Caret library of R Studio is used for creating the stack based ensemble of classifiers.The results show that the proposed hybrid model that used sentiment score as one of the features in feature set performed better with an accuracy of 0.8742223 for Kaggle -US Airline Twitter Sentiment Data Set, 0.8881453 for data set related to 'Clean India Mission' and 0.9953593 for Sentiment 140 Twitter Data Set, as compared to machine learning classifiers and other ensemble classifiers.