This study investigates the use of deep learning models for sentiment analysis of Twitter data focused on the gaming community. It specifically compares the performance of a conventional Long Short- Term Memory model with a hybrid model that combines LSTM with the AdamW optimizer. The dataset, sourced from Kaggle, consists of gaming-related tweets that are often informal and sentimentally diverse, providing a challenging testbed for classification models. The objective was to determine whether the use of the AdamW optimizer—known for enhancing generalization through decoupled weight decay, a component of AdamW, assists in managing model complexity by preventing overfitting. Depending on the dataset, however, its impact might be fluctuating.—could improve the LSTM model’s performance. The findings revealed that the standard LSTM model slightly outperformed the LSTM + AdamW hybrid, with accuracies of 90.88% and 90.72%, respectively. The marginally lower performance of the hybrid model may stem from factors such as excessive regularization, suboptimal hyperparameters, or dataset-specific traits that favored the traditional training approach. Nevertheless, the hybrid model demonstrated strong performance and stable training behavior, showing potential for future refinement. This research contributes a novel comparison of LSTM and LSTM + AdamW for sentiment classification in a niche domain and highlights the impact of optimizer choice on model performance.