Sentiment analysis on E-commerce based product reviews using machine learning algorithms

Abstract

Opinions from others can be important when a choice is necessary, mainly when valuable resources such as time and wealth are involved. People often depend on their peers’ past experiences. The analysis of sentiments or opinions is a computation using text analysis of people’s emotions, thoughts, and feelings. In recent years, it became the most significant natural language processing and sentiment analysis research approach. Because it is founded on people’s views and as all people have different viewpoints to see things, it is becoming more popular every day. In this digital world, when a person decides to purchase a product or utilize a service, they get access to many customer evaluations. Still, it is a tiresome process to read and analyze them all. Moreover, when an organization wants to make a profit, find new possibilities, anticipate sales trends, and manage its reputation via public opinion or sell its product, it also needs to address many customer remarks accessible to its customers. Therefore, our goal is to show that it is feasible to do so with sentiment analysis approaches. With sentiment analysis, it is easy to analyze and extract a vast number of accessible data comments from both consumers that can aid in fulfilling the objectives of the organization. The dataset utilized here is obtained from the Ali-Express e-commerce website’s online product reviews. We use various data processing techniques such as tokenization, removal of punctuation marks and stop words, stemming, TF-IDF, and parts of speech tagging. The results of our study comprise several methods of machine learning techniques. In this work, we examined eight distinct types of Machine Learning Algorithms, including Naive Bayes Classifier, SVM, Random Forest Classifier, Logistic Regression (L.R.), K- nearest neighbors, XGBoost Classifier, Decision Tree Classifier, and Gradient Boosting, and compared their precision and accuracy to find the most accurate one. According to our analysis, Logistic Regression is performing better among all the other seven classifiers having an accuracy of around 98%, and KNN is performing lowest, having an accuracy of 46%. Other classifiers such as Decision Tree is, giving 69%, Gradient Boosting 63%, Naive Bayes and SVM 82%, and Random Forest and XGBoost are giving 80% of accuracy.