This research introduces a novel dataset of user reviews in the mobile gaming domain, comprising over 251,000 reviews spanning 72 game types gathered from the Google Play Store. Leveraging advanced natural language processing (NLP) techniques, the dataset undergoes processing to serve as a valuable resource for developing sentiment analysis models. Additionally, this paper presents a new approach utilizing Transformer models for the sentiment analysis problem on this dataset, explicitly focusing on overall sentiment classification into Positive, Negative, or Neutral categories. In this paper, we incorporate emoji data into sentiment classification. The experimental results demonstrate that the inclusion of emojis leads to improved performance. Specifically, the RoBERTa model achieves the highest performance on the emojis dataset, with an Accuracy of 0.942, Loss of 0.146, and F1 Scores for Positive, Neutral, and Negative sentiments at 0.970, 0.800, and 0.930, respectively.