Comparison of Classical and Neural Network-Based Models for Sentiment Analysis of Spotify Reviews

Abstract

Digital music streaming platforms such as Spotify have transformed how users access and consume music, generating large volumes of user reviews that reflect satisfaction and dissatisfaction with the service. Sentiment analysis of these reviews can provide valuable insights for developers and service providers. Although a number of studies have performed sentiment analysis on Spotify reviews, most of them are limited to comparing only classical algorithms. As a result, there is still a lack of evaluation that compares several classical algorithms and a simple neural-network-based model under the same feature representation and standardized text preprocessing on the same dataset. Therefore, this study aims to: (1) build sentiment classification models for Spotify reviews using classical machine learning algorithms, namely Naive Bayes, Support Vector Machine (SVM), and Logistic Regression; (2) develop and evaluate a Multi-Layer Perceptron (MLP) model using the same TF-IDF feature representation; and (3) compare the performance of all models. The experimental results show that Logistic Regression consistently achieved the best overall performance, with the highest accuracy (87.85%), precision (87.79%), and F1-score (87.65%), slightly outperforming Linear SVM and clearly surpassing Naive Bayes and MLP. Although the MLP model obtained a slightly higher recall than Naive Bayes, its overall performance remained lower than that of Logistic Regression and Linear SVM.