DEEP LEARNING DRIVEN SENTIMENT ANALYSIS OF E-COMMERCE CONSUMER IMPRESSIONS USING ADVANCED FETURE EXTRACTION TECHNIQUES

Abstract

E-commerce has emerged as one of the biggest players in the current digitized business environment, and this has led to the creation of large amounts of consumer data through consumer reviews and feedback. The objective of this master thesis is to identify the consumer impression in the e-commerce data by applying sophisticated feature extraction techniques and sentiment analysis based on deep learning approaches. This paper seeks to explore the elements of consumer sentiment as captured in online reviews, which is vital in increasing customer satisfaction and sales. The research problem seeks to establish the performance of different machine learning models in sentiment analysis of e-commerce reviews and feature extraction techniques such as TF-IDF and Word2Vec. The main goal is to identify which set of machine learning models and feature extraction methods gives the best accuracy and efficiency in sentiment analysis. The methodology includes a systematic review of the literature in order to identify the current sentiment analysis methods and their uses in e-commerce. The analysis utilises a collection of Amazon product reviews, which is first cleaned, tokenized, and balanced before being used in the study. Thus, four machine learning models, including Support Vector Machine (SVM), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Bidirectional Encoder Representations from Transformers (BERT), are chosen for the comparison. These models are then optimized and assessed with numerous evaluation metrics like accuracy, precision, recall, and F1 score. Empirical Findings show that deep learning models especially BERT exhibit higher accuracy than traditional machine learning models in the sentiment analysis task because they can analyze the context and language features of the text. BERT provided the highest accuracy thus showing its effectiveness in handling the sentiment analysis of consumer reviews. The study also focuses on the significance of feature selection where TF-IDF and Word2Vec improve the results of the model. The study outcome shows that the combination of the advanced feature extraction technique with the deep learning model is useful in developing a robust framework for sentiment analysis in the e-commerce context. This approach allows organizations to acquire a better understanding of customers’ tendencies and issues, which helps in decision-making and improves customer engagement. Further research will focus on the development of the hybrid models and live sentiment analysis to improve the overall performance and usability of the proposed approach for dynamic e-commerce scenarios. The study outcome shows that the combination of the advanced feature extraction technique with the deep learning model is useful in developing a robust framework for sentiment analysis in the e-commerce context. This approach allows organizations to acquire a better understanding of customers’ tendencies and issues, which helps in decision-making and improves customer engagement. Further research will focus on the development of the hybrid models and live sentiment analysis to improve the overal