Sentiment Exploring on Feedback of E-commerce Data Using Machine Learning Algorithms

In today’s fast-growing Internet world, customer ratings and reviews play an essential role in online buying on e-commerce websites such as Amazon, Flipkart, and others. Sentiment analysis is crucial for increasing customer satisfaction on e-commerce sites since it contains a lot of consumer feedback. In this work, we have used Amazon Women's E-Commerce Clothing Reviews dataset. We have used CountVectorizer and TF-IDF and trained the data on five machine learning (ML) classifiers, namely logistic regression (LR), multinomial Naive Bayes (MNB), Bernoulli Naive Bayes (BNB), support vector machine (SVM), random forest (RF), and AdaBoosting (AB). When comparing the ML model’s accuracy scores concerning the CountVectorizer, it was discovered that MNB and LR models had the highest accuracy of 0.94, while RF had the lowest accuracy of 0.90. SVM achieved the maximum accuracy of 0.94 using the TF-IDF approach, and MNB achieved the lowest accuracy of 0.89. The accuracy, precision, recall, F1-score, and AUC-ROC curve help us to determine the performance of the ML algorithms. To examine the dataset’s attributes and comprehend the relationships between the variables, many statistical techniques were applied.