Extract Sentiment from Customer Reviews: A Better Approach of TF-IDF and BOW-Based Text Classification Using N-Gram Technique

Abstract

To enhance customer and product services, customer reviews play a must-have role for any business organization in the digital era. Business organizations can improve customer and product services to their greatest level by further analyzing the customer’s demands. For this purpose, text mining helps to convert unstructured text into structured form for further analysis. Sentiment analysis is a concept of opinion mining which draws significant advantages by classifying the customer reviews as positive or negative. The principal goal of this paper is to perform opinion classification. Bag of Words (BOW) and Term Frequency-Inverse Document Frequency (TF-IDF) are used as feature extraction techniques along with N-Gram and Random Forest (RF), Support Vector Machine (SVM), and Naive Bayes (NB) and are regarded as machine learning classifiers for reviews classification. Again, voting ensemble classifier has also been applied in order to investigate better performance and customer reviews are considered to accomplish the job. Though all the classifiers and feature extraction techniques achieve good performance, voting ensemble performs better among the classifiers while TF-IDF performs better for feature extraction techniques. From the notable result of the experiment, it can be pointed out that the implemented technique of customer reviews analysis might be an effective use to reshape business products.