Sentiment Analysis of Customer Reviews for Predictive Product Development in e-Commerce

Abstract

The recent expansion of online transactions, such as e-commerce, has led to the offering of many products and services via online platforms. Because there are so many items accessible while buying online, users struggle to choose the one that best fits their needs. The complex interactions between user and item properties have been the subject of several research in deep learning-based recommender systems (RSs). Since e-commerce has grown so rapidly in recent years, it has become more important to use user evaluations as a basis for purchasing decisions. In addition to assisting prospective clients in making well-informed judgements, reviews create trust and provide companies useful information. Sentiment analysis is a method used to analyse product evaluations, marketing campaigns, and consumer sentiment. Decisions about future marketing initiatives, product and service development, and customer service improvements may be influenced by this useful data. Predicting ratings on social media is often used to forecast product ratings based on user feedback. Convolutional neural networks (CNN), recurrent neural networks (RNN), and bi-directional long short-term memory (Bi-LSTM) are among the deep learning models that are extensively benchmarked in our research. These models are assessed using a variety of word embedding methods, including Word2Vec, FastText, and bi-directional encoder representations from transformers (BERT) and its variations.  In this study, we examine and compare the performance indicators of neural network-based models for consumer sentiment prediction using a dataset of product evaluations from consumers of an online women's clothing company.