Dimensionality Reduction for Sentiment Classification using Machine Learning Classifiers

Sentiment analysis intends to identify the opinion either positive or negative given by clients or users from review documents. Sentiment analysis utilizing machine learning strategies faces the issue of high dimensionality of the feature vector. Consequently, a feature reduction strategy is required to dispose of the unessential and noisy elements from the feature vector. Feature reduction techniques selects the prominent features for reducing size of the feature set. The features which are nearly distributed presented by different class in the feature vector, make complexity for the classifier to draw a clear decision boundary. In this work, we proposed two different approaches (i.e., Term Presence Count (TPC) and Term Presence Ratio (TPR)) to remove those redundant features in positively and negatively tagged documents with nearly equal distribution. We applied four machine learning-based classification techniques including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Naive Bayes (NB) for sentiment classification using movie review dataset. Finally, the classifiers are evaluated in terms of accuracy, precision, recall, and Average F-measure. Experimental results manifest that the feature dimension reduced to approximately 83% by our proposed method while improving the classification performance.