A Comparative Study of Machine Learning Algorithms for Sentiment Analysis of Game Reviews

Abstract

Sentiment analysis, also known as opinion mining, is the process of analysing a body of text to determine the sentiment expressed by it. In this study, Natural Language Processing techniques and Machine Learning algorithms have been applied to create multiple sentiment analysis models customized for the gaming domain to determine the sentiment of game reviews. The dataset was collected from Steam and Metacritic through the use of web API and web scraping. This was followed by text preprocessing, data labelling, feature extraction and finally model training. In the training phase, the effects of oversampling and hyperparameter tuning on the performance of the models have been evaluated. Through comparison between Support Vector Classifier (SVC), Multi-layer Perceptron Classifier (MLP), Extreme Gradient Boosting Classifier (XGB), Logistic Regression (LR) and Multinomial Naïve Bayes (MNB), it was evident that SVC had the most superior performance.