Toward Stacking Ensemble Based Bipartite Sentiment Classification of Hindi Movie Review Text
Ankita Sharma, Udayan Ghose · International Journal of Engineering Trends and Technology · 2024
Sentiment analysis has significantly progressed in resource-rich languages like English, but research in Hindi is still advancing. Regardless of being the third most spoken language globally, Hindi faces resource limitations. However, the growing use of technology and Hindi interfaces has led to abundant Hindi text on the web, presenting opportunities for researchers to extract valuable insights. The present work aims to evaluate the effectiveness of ensemble learning methods for bipartite sentiment classification of Hindi Movie Reviews (HMRs). This area has received relatively less attention from researchers. The study involves manually creating a binary HMR dataset comprising 6,000 reviews. Preprocessing and feature extraction are performed on the collected dataset. Several individual classification models are applied to the HMRs; subsequently, the predictions from these models are combined through a hard voting ensemble approach, and finally, an integrated two-layered stacking ensemble architecture is proposed and implemented in the present work. The preprocessed dataset undergoes classification using SVM, RF, DT, and KNN models in the first classification stage. The decisions from these four classifiers are then amalgamated to build and optimize the second-level estimators SVM and MLP. Ultimately, the meta-classifier provides the final prediction for the bipartite sentiment labels. The results demonstrate that the proposed model achieves the highest performance. Furthermore, the outcomes derived from this investigation have undergone rigorous statistical assessment through the application of the Friedman statistical test. The proposed framework has achieved the most elevated ranking in both the HMR and IIT-P movie review datasets, thereby providing substantial verification of the obtained results. Notably, this study is the first instance of the application of a statistical test for supplementary validation within the realm of the Hindi Review Sentiment Classification task.