Predicting Customer Sentiment From Product Reviews Using Machine Learning
Azan Bukhari, Zamima Batool, Vu Minh Phuc, Anchit Bijalwan · 2025
This study employs a Logistic Regression model for sentiment analysis on the IMDB dataset, achieving an accuracy of 89%. The dataset was preprocessed using TF-IDF vectorization, ensuring that sentiment-relevant words were emphasized. Model evaluation utilized metrics such as precision, recall, F1-score, confusion matrix, and the ROC curve, with an AUC of 0.96, demonstrat-ing the model's robustness in distinguishing positive and negative sentiments. A key novelty of this research lies in its interpretability, with feature coefficients re-vealing the most influential words for predictions. Words like “great” and “amaz-ing” were strongly associated with positive sentiment, while terms such as “worst” and “awful” indicated negative sentiment. Additionally, qualitative in-sights were derived through visualizations, including word clouds and bar charts, providing a comprehensive understanding of the dataset and model behavior.