Sentiment Analysis of Steam Reviews Using Transformer Models

In recent years, deep learning models such as BERT and RoBERTa have shown remarkable performance on various natural language processing tasks, including sentiment analysis. In this study, we compare the performance of BERT and RoBERTa models on the Steam Reviews dataset for sentiment analysis. The Steam Reviews dataset contains over 6.4 million reviews in English from the Steam platform, each labeled as positive or negative sentiment. We preprocessed the data, trained both models, tuned the hyperparameters, evaluated the models, performed statistical analysis, and visualized the resultson a subset of samples. Our findings suggest that the RoBERTa model outperforms the BERT model in terms of accuracy, precision, recall, and F1-score. This study highlights the effectiveness of RoBERTa in the sentiment analysis task and provides insight into the performance differences between BERT and RoBERTa models. We also discuss potential reasons for the observed differences in performance, such as the larger size of the RoBERTa model and its use of dynamic masking during training. Our study contributes to the growing body of research on the use of deep learning models for sentiment analysis tasks and provides valuable insights for practitioners seeking to choose between BERT and RoBERTa models for similar tasks. Overall, our results suggest that RoBERTa is a more effective model than BERT for sentiment analysis on the Steam Reviews dataset.