Sentiment analysis helps businesses connect with potential customers more effectively. However, limitations remain in capturing word context relation-ships and improving accuracy. In recent years, research has increasingly focused on addressing this challenge using transformer architectures, yielding promising results. To tackle this issue, this study proposes an innovative sentiment analysis model for data review. BERT-base transforms input text into tokens for semantic feature representation. Then, LSTMs enhance contextual understanding. Finally, an attention mechanism extracts key textual information, providing a more comprehensive understanding of text sequences and enabling the quick identification of the most relevant content. Experimental results on three datasets: Twitter Airline, IMDB, and Sentiment140, demonstrate accuracies of 93.15%, 96.07%, and 89.84%, respectively.