Sentiment Analysis of Amazon Musical Instruments Reviews using Machine Learning Techniques

This paper delves into sentiment analysis of Amazon musical instruments reviews, employing a comprehensive data science methodology. With a dataset comprising 10,261 rows and 9 columns from Kaggle, the analysis navigates through data preprocessing, exploratory data analysis (EDA), and the crucial task of labeling products based on customer ratings. The sentiment labels—Positive for ratings > 3.0, Neutral for ratings = 3.0, and Negative otherwise—pave the way for subsequent text preprocessing. N-gram analysis becomes a pivotal exploration, unraveling insights from unigrams to trigrams. The impact of retaining "not" in the list of stop words is noted, influencing the sentiment interpretation. Visualization through word clouds further illuminates sentiments. Feature engineering involves dropping insignificant columns, encoding the target variable, TF-IDF vectorization, and resampling the dataset using SMOTE. Model selection unfolds with a thorough evaluation of six classification models through 10-fold cross-validation, culminating in the identification of Logistic Regression as the optimal choice. Hyperparameter tuning refines the model, achieving a notable accuracy of 94.80% on the test set. The conclusion synthesizes key findings: the dataset's focus on guitars, the efficacy of Logistic Regression in sentiment analysis, and the necessity for considering alternative cross-validation methods like Stratified K-Fold. Additionally, the suggestion to explore data scaling is made for future enhancements. The paper illuminates the nuances of sentiment in musical instrument reviews, providing valuable insights for both sellers and consumers in the e-commerce domain.