Comparison of machine learning and deep learning algorithms on sentiment analysis in game reviews

Abstract

This study aims to analyze player review sentiment for the Marvel Rivals game on Steam using machine learning and deep learning algorithms, including Random Forest, Naive Bayes, XGBoost, and Bi-LSTM. The research was conducted within the CRISP-DM framework, which encompasses understanding the business problem, data exploration, data preparation, model building, and evaluation and implementation. Player review data was collected through web scraping, then preprocessed to clean and reformat the text before being used to train a sentiment classification model. Model evaluation was performed using metrics such as accuracy, precision, recall, and F1-score to identify the most effective model. The results indicated that Bi-LSTM was the best performing model, achieving an accuracy of 89% and an F1-score of 0.72 for negative sentiment. Hyperparameter tuning on real data contributed significantly to this performance. Conversely, applying SMOTE to balance the dataset actually reduced the performance of the Bi-LSTM model, suggesting that parameter optimization is more effective than synthetic data balancing, particularly for deep learning models.