An empirical research on sentiment analysis using machine learning approaches

Nowadays users of social networks are very much interested in expressing their opinions about different sorts of products or services in social media which leads to the growth of user-generated web contents. Their reviews on social media have a significant impact on customers for making effective and optimal decisions for buying products or using services. In sentiment analysis, most of the used approaches are based on machine learning techniques. In this paper, the well-known methods of machine learning are reviewed and compared against each other. Then the comparative studies on the performance of these techniques on online user reviews that come from multiple industry domains are performed. The experiments involve many different data sets from various domains including Amazon, Yelp and IMDb. Well-known methods such as Support Vector Machine, Decision Tree, Bagging, Boosting, Random Forest and Maximum Entropy are implemented in the experiments. Based on the experimental results it is found that users can extract applicable information from review data sets for business intelligence and better product sales production, and that Boosting and Maximum Entropy outperform the other examined machine learning algorithms for detecting sentiments in online user reviews.