Cross Domain Sentiment Analysis on E-Commerce Datasets using Machine Learning and Ensemble Learning Approaches

Abstract

There is an enormous amount of data available on internet on online user reviews about products feedback. This data is priceless for industries and businesses to take forthcoming decisions and develop marketing strategies for upcoming products based on users’ interest. By making an analysis on available users’ review data, organizations can understand the market trend and provide better service that can significantly scale up the business. This analysis on users’ sentiments is achieved through data mining approaches where a model is trained on any particular domain data source and used on the new data to predict polarity of user interest. In some instances, cross domain sentiment analysis is very useful when labeled reviews are not abundant for model training. Though, there exist some works which address the problem of cross domain sentiment analysis but still there is need of a comprehensive research on usability of machine learning (ML) algorithms in this field and efficiency evaluation of them in cross domain sentiment analysis. In this work, we use various ML algorithms and ensemble classifiers to test their efficient in classifying different domain datasets in training and testing phase. Also, we propose an efficiency approach for feature extraction to enhance the prediction accuracy. We experimented our proposed work using various ML and ensemble approaches and evaluate the experiment results on various parameters like F1-Score, precision, recall, accuracy, AUC and ROC.