| Abstract [eng] |
In today’s digital world, understanding online customer reviews is essential for evaluating user satisfaction and improving business products and services. This thesis explores the potential of machine learning (ML) and natural language processing (NLP) methods to analyse and predict customer satisfaction based on user reviews. Using data from Wix, a leading website building platform, the research aims to identify key topics in customer feedback, predict review star ratings, and assess the effectiveness of sentiment analysis as an alternative measure of satisfaction. The thesis begins by reviewing existing literature on customer satisfaction analysis, ML and NLP methods. The methodology combines sentiment analysis, topic modelling, and supervised ML models. Reviews are preprocessed using advanced NLP techniques, including tokenisation, lemmatisation, and vectorisation, followed by training models such as Logistic Regression, Support Vector Machine, and XGBoost to predict star ratings. The results show that advanced machine learning models can effectively predict star ratings using only the text from reviews. Sentiment analysis also proved useful in measuring customer satis faction, aligning closely with traditional star ratings. Topic modelling uncovered common themes in customer feedback, such as usability problems, feature requests, and customer support experiences, offering valuable insights for improving Wix’s products and services. This study contributes to the field of automated review analysis by demonstrating how traditional and modern NLP techniques can work together effectively. The results offer practical value for businesses that seek to use customer feedback to improve user experience and satisfaction. |
