Novel opinion mining system for movie rviews in Turkish

Abstract

Opinion Mining (OM) works on transferring the online available opinions into useful knowledge. In this paper, a novel opinion mining system of reviews in Turkish has been presented. The proposed system utilizes Word2Vec, which is one of the states of the art text feature extraction method, along with an ensemble learning algorithm for classification. The challenging and benchmark “IMDB Movies Reviews” dataset has been used for conducting the experimental comparison and verification. In addition, the performance of the proposed method is compared to some of the well-known machine learning algorithms like Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Naive Bayes (NB). The tested ensemble methods are the Random Forest (RF), AdaBoost Classifier, and Gradient-Boosting Classifier (GBC). The results of the conducted experiments using the dataset have shown that the performance of SVM, KNN, and NB are comparable. However, the performance, robustness, and stability of the system have been significantly improved by adapting the RF ensemble learning, along with the Word2Vec feature vector, and suitable pre-processing operations on the data. In addition, the proposed method is compared to one of the states of art ensemble methods and have shown superior performance with respect to it.