A Comparative Analysis of Feature Extraction Methods for Human Opinion Grouping Using Several Machine Learning Techniques

Opinion analysis is one kind of text mining that aims to find out human opinions and emotions towards any particular event, topic, or subject. Feature extraction is one of the most crucial steps in the text categorization process and an efficient feature extraction method can greatly enhance the accuracy of opinion mining. In this paper, we try to concentrate on comparing the performances of four well-known feature extraction techniques (i.e. BOW, TF-IDF, Skip-Gram, and CBOW) for binary opinion classification. To accomplish the task, two different datasets of hotel and restaurant have been considered and three machine learning models i.e. Naive Bayes (NB), Support Vector Machine (SVM), and Logistic Regression (LR) have been applied as supervised classifiers. In terms of classifier accuracy and mean squared error (MSE), the experimented result depicts that Skip-Gram technique shows the highest performance among the four feature extraction techniques for both datasets. In addition, SVM and LG classifiers perform the best for hotel and restaurant datasets respectively and our proposed approach comes out with an optimistic accuracy of 95.8%.