Sentiment Analysis of Steam Video Game Reviews for Disco Elysium Using the Random Forest Tuning Method with Chi-Square Features

Video games have emerged as one of the most popular forms of entertainment worldwide, particularly through digital distribution platforms like Steam, which provides access to over 30,000 games. User reviews on Steam are crucial for game developers to understand consumer needs and improve product quality. This study aims to conduct sentiment analysis on user reviews for the game Disco Elysium on Steam using the Random Forest classification method, optimized through Chi-Square feature selection. The dataset comprises 5,000 manually labeled reviews categorized as positive or negative. Preprocessing steps include punctuation and number removal, lowercasing, stopword elimination, stemming, and tokenization. The TF-IDF method is then applied for word weighting, followed by Chi-Square for feature selection to reduce irrelevant features. The Random Forest model is subsequently applied with hyperparameter tuning via GridSearchCV and RandomizedSearchCV. Evaluation metrics, including accuracy, precision, recall, and F1-score, are calculated using a confusion matrix. Results indicate that stemming and hyperparameter tuning enhance model accuracy, while Chi-Square feature selection does not always significantly improve performance. RandomizedSearchCV outperforms GridSearchCV in terms of computational efficiency and optimal parameter exploration, proving more effective for this study's requirements. This research highlights the potential for further optimization in preprocessing techniques and feature selection to enhance text-based sentiment analysis. Findings from this study can guide game developers in efficiently understanding user sentiment from digital platform reviews, thus aiding in more informed product improvement and decision-making.