A Data Analytics and Hybrid Expert System Framework Application in the E Commerce Industry for User Product Rating Analysis

Abstract

In the rapidly evolving world of e-commerce, the influence of online reviews on consumer newlinepurchasing decisions cannot be overstated. Online reviews serve as a cornerstone for newlineinformed decision-making, offering potential buyers a wealth of information on the quality, newlinefunctionality, and user satisfaction associated with products. Amazon, a global leader in newlinee-commerce, exemplifies this trend by integrating extensive user-generated product reviews newlineinto its platform. These reviews are not only beneficial to consumers but also vital for newlinebusinesses aiming to understand customer sentiment and improve product offerings. newlineA significant aspect of utilizing these reviews is the accurate classification of user ratings. newlineThis task involves analyzing textual reviews to predict the numerical ratings provided by newlineusers, which range from one to five stars. The classification of these ratings is crucial newlinefor several reasons. Firstly, it aids in summarizing the general sentiment towards a newlineproduct. Secondly, it helps in identifying specific areas of improvement for products. newlineLastly, accurate rating classification can enhance the recommendation systems used by newlinee-commerce platforms, thereby improving the overall user experience. newlineThis study focuses on a comprehensive comparative analysis of several machine learning newline(ML) and deep learning (DL) algorithms for user rating classification in Amazon food newlinereviews. The algorithms evaluated include Naive Bayes, Logistic Regression, Support newlineVector Machines (SVM), Long Short-Term Memory (LSTM) networks, Convolutional newlineNeural Networks (CNN), and introduces Deep Bidirectional Recurrent Neural Networks newline(DBRNN) as a novel approach. Naive Bayes and Logistic Regression are traditional ML algorithms known for their newlinesimplicity and efficiency. Naive Bayes, based on Bayes theorem, assumes independence newlinebetween features, making it less effective for complex datasets but still a strong baseline newlinemodel. Logistic Regression, which predicts probabilities of categorical outcomes, is robust newlineand interpretable.