Compare Machine Learning Models in Text Classification Using Steam User Reviews
Youchen Miao, Zeyu Jin, Yumeng Zhang, Yuchen Chen, Junren Lai · 2021
Text Classification and Sentiment Analysis of game reviews are viewed as important parts in not only academic fields but also in game studies. In this paper, with more than 400 thousand game reviews on Steam platform, we preprocess the data using different libraries (sklearn, nltk, and spaCy) and use them as inputs to build three sentiment classification models based on different algorithms (Naive Bayes, SVM, and Random Forest). In contrast to previous studies that only focus on different sentiment analysis models, our paper also highlights the use of different APIs to preprocess the data and their corresponding model performance. The results show that no matter which API we choose, Random Forest models always perform the best. However, in terms of training time, Naive Bayes is the fastest. This work can be used to apply grid search for researchers to automatically find the optimum API before conducting sentiment analysis in the future.