CLASSIFICATION OF CUSTOMER SENTIMENTS BASED ON ONLINE REVIEWS: COMPARATIVE ANALYSIS OF MACHINE LEARNING AND DEEP LEARNING ALGORITHMS

Abstract

E-commerce's transformation of consumer behavior has increased the importance of understanding customer emotions, especially in the transition from traditional retail models to online platforms. The proliferation of online shopping has fundamentally changed not only shopping habits but also consumer interactions and purchase decisions. This research aims to compare and analyze the performance of various text mining and machine learning algorithms in the context of sentiment analysis and online review data. For this purpose, analyses were performed with a total of five supervised classification algorithms including Logistic Regression, Naive Bayes, Support Vector Machine, Random Forest, AdaBoost, and a deep learning model, CNN Model. The dataset used in the study includes customer reviews obtained from a women's clothing e-commerce platform. The missing data were completed by pre-processing the dataset. Count Vectorizer and TF-IDF vectorization were performed to transform the textual data. In addition, various text preprocessing steps were applied. According to the findings obtained from the research, AdaBoost and Naive Bayes algorithms were the most effective algorithms in terms of classifying customer sentiments. No significant difference was detected in terms of the vectorization method used. Although the CNN Model showed high performance, the generalizability of the model was considered low because overfitting was detected during the training of the model.