Public opinion sentiment analysis has become increasingly important for understanding social and commercial trends, especially with the surge of user-generated content on social media and online platforms. Existing sentiment analysis models, such as BERT and LSTM, have shown effectiveness in capturing contextual information but often face challenges related to computational complexity and efficiency. In this paper, we introduce DeepSentOpt, a new deep learning model that merges an optimized attention mechanism with a precision optimization strategy to enhance both accuracy and efficiency in sentiment analysis. Extensive experiments on three benchmark datasets—Sentiment140, IMDB Movie Reviews, and Amazon Product Reviews—demonstrate that DeepSentOpt consistently surpasses current state-of-the-art models in classification accuracy while substantially lowering training time and reducing model size. Our findings confirm the model’s effectiveness across varied text data, ranging from brief tweets to longer reviews, underscoring its practicality for real-world sentiment analysis tasks. Additionally, DeepSentOpt’s lighter computational footprint makes it well-suited for deployment on resource-limited devices, such as mobile phones and edge computing platforms. Future research will investigate its potential for multilingual sentiment analysis and integration with domain-specific knowledge.