Sentiment classification is an essential component of natural language processing under deep learning. It is also an important task that finds application in numerous purposes such as customer experience analysis, customer retention analysis, and opinion mining. The intention of this study was to evaluate if the Epoch parameter in Deep Learning models has been effectively optimized in Sentiment Classification studies. Choosing the right number of epochs is important to properly tune model hyperparameters. In effect, the performance of deep learning model can be maximized. A review of 20 articles on deep learning sentiment classification conducted by the study shows that Epochs have not been adequately explored, with 10 (50%) studies having referencing Epochs and 0 (0%) studies applying Epoch optimization. There is a need to optimize deep learning models for sentiment classification, as shown in this study. Future research trends in epoch optimization have been presented in this study which focuses on adaptive and automated approaches to enhance training efficacy, prevent overfitting, and promote generalization performance. Future trends focus on those problems in which we can deal with optimizing the training epochs of the deeply diverse and complex applications for deep learning.