An Automated Machine Learning Approach for Sentiment Classification of Bengali E-Commerce Sites

Abstract

E-commerce reviews and comments about specific products disclose consumer’s perceptions as well as attitudes. This attitudes expressed by the consumer’s seem to be most useful for the new customers who is interested on any product. Meanwhile, an ever increasing number of reviews and comments are being stored daily and the amount of people buying goods online are increasing in a great extent. User emotions record associated with every products is beneficial for both the makers as well as customers. But with the increasing number of datasets in the e-commerce websites, it has almost become impossible for manual analysis and without automated machine learning approach it can’t be even imagined. Therefore, improvement in the field of machine learning approaches must be accomplished. Realizing the worth of this, this work proposes a hybrid machine learning approach by incorporating different machine learning approaches. Own Bangla StopWords database consisting of approximately 900words have also been concentrated in this work. Initially, input data are tokenized using python NLTK library and filtered using StopWords created. Then conversion of data to numerical from string are conducted using TF-IDF (Term Frequency–Inverse Document Frequency) information retrieval mechanism and finally trained using K-nearest neighbor (KNN) with Support Vector Machine (SVM). Our proposed approach Normalization along with StopWords filtering embedded KNN based SVM demonstrates superiority after a comparative study with Principle Component Analysis (PCA) with Convolutional neural network (CNN), Random Forest, Logistic Regression.