Conceptual Analysis of Product Evaluations Using Deep Learning

Abstract

The process of classifying a small text piece into positive, negative or neutral is called as Sentiment Analysis. For example, if a message is considered, then the meaning of the sentence is identified and then analysis of the sentence has to be done. After analysis, we decide that the sentence is either positive, either negative or neutral. This is referred to as Sentiment Analysis. Nowadays, most of the people prefer the online products compared to the direct buyers. So, the online marketing is growing very high. In fact, the online marketing is fully satisfying the customer needs so that it occupies a great place in people's heart. They are also satisfying the demands of customers with high quality and low price. These are the primary reasons for the online marketing to be in a successful position. If an e-commerce company has an ability to gather information about the customer thoughts and behavior, then the online marketing will be more effective than now. In this paper, we have taken a particular e-commerce dataset called Flipkart and classify the polarity of the comments by using some of the classifiers namely Support Vector Machine Classifier, Guassian Naïve Bayes Classifier and Random Forest Classifier and Multilayer Perceptron Classifier. The existing approach is that the comments were classified based on the attitude of the customer. But, now the proposed approach has been implemented with the help of Multilayer Perceptron (MLP) Neural Network Classifier which is simulated by the SPYDER tool. The accuracy for all the four algorithms namely, Support Vector Machine, Random Forest, Naïve Bayes and Multilayer Perceptron (MLP) Neural Network algorithm has been computed and the best accuracy has been predicted by comparing them. Here, the Multilayer Perceptron (MLP) shows the best accuracy of all others and the accuracy is 99.94%. Moreover, some performance metrics has been determined here. The performance metrics namely Precision, Recall and F-measure has been evaluated for each algorithm individually and then the comparison is made among them. Then, the ROC curve is measured for the classes designated by us for revealing the recognition of parameters between two diagnostic groups. It is predicted individually for each class. At last, the Confusion Matrix is going to enumerate for each algorithm distinctly. It presents the actual and predicted values in a tabular format thereby measuring the performance of a classifier.