Sentiment Learning from Imbalanced Dataset: An Ensemble Based Method
Vinodhini Gopalakrishnan, Chandrasekaran Ramaswamy · 2014
More people are buying products online and expressing their opinions on the products through online reviews. Sentiment analysis is used to extract opinion related information from the reviews and the extracted results can benefit both consumers and manufacturers. Much work on machine learning based sentiment classification has been carried out on balanced datasets. However, the real time sentiment analysis is a challenging machine learning task, due to the imbalanced nature of positive and negative sentiments. Sentiment analysis becomes complex when learning from imbalanced data sets, very few minority class instances cannot present sufficient information and result in performance degrading significantly. Modifying the data distribution or the classifier are the traditional approaches for dealing with the class imbalance problem. In this work, we propose to apply a combination of both approaches. We propose a modification in ensemble based bagging algorithm and also in sampling method used for data distribution, so as to solve class imbalance problem to improve the classification performance. We found that the modified bagged ensemble makes an improvement in predicting performance in terms of the receiver operating characteristic curve (ROC). The results also show that the modified bagging model performs better in terms of area under the receiver operating characteristic curve (AUC) in imbalanced dataset.