A machine and DL approach for classifying customer sentiments from online shopping reviews in Bangla text

Abstract

Due to the widespread availability of the internet all across the world, people prefer shopping online rather than going to a shop. There are various online marketplaces available in Bangladesh, like Daraz, Pickaboo, Rokomari, Othoba, Bikroy, Food Panda, and Robi Shop. With the increasing quantity of customers on online shopping platforms, the number of product reviews also increases with it. Data is classified utilizing machine learning (ML), deep learning (DL), transfer learning, and other data mining algorithms to facilitate the customer’s comprehension of the primary subject of the review before making a purchase. Natural language processing techniques are employed to categorize data in any given language for such issues. There are no Bengali shopping review datasets available on online sites. So, we manually collected a dataset of 2,600 reviews. In this paper, reviews are classified into 5 categories (satisfied, very satisfied, not satisfied, fairly satisfied, and satisfied but delivery problem). DL (long short-term memory (LSTM) and convolutional neural network (CNN)) and ML (support vector machine (SVM), random forest (RF), gradient boosting (GB), and extreme gradient boosting (XGBoost)) model have been applied. Among the DL models, CNN has the best accuracy (91.27%), and the RF classifier provides the highest accuracy (84.39%) out of all the ML models.