<p>Today more than ever, social media like Twitter is significant. However, it has transformed into a public bulletin board where regular people, businesses, and even well-known figures like celebrities regularly post their opinions and take part in discussions. Also, Twitter persists as a useful tool for friends to stay in touch. Despite the widespread adoption of Twitter globally, it is more important than ever to be able to do trustworthy analyses of opinions to assess public opinion and perceptions on a variety of topics. Emojis are often used to convey emotions or thoughts that are challenging to put into words concisely. This paper establishes a new way of analyzing opinions by merging deep learning-based sentiment analysis approaches and integrating them with emojis. In this line of action, we propose the Giza reformed deep attention model for both textual as well as emoji-based feature extraction to classify user sentiments. To construct an enhanced opinion mining, a feature extraction and data preparation process based on multi-sets is carried out, followed by the application of the Giza pyramid construction optimizer to feature selection. It significantly reduces the dimensions to lead toward improved accuracy in predictions. This paper proposes the reformed temporal attention-based gated convoluted network (RefTAC) for effective sentiment classification based on the selected features. We validate the effectiveness of our proposed approach on popular Twitter datasets, and the results show impressive performance: 99.1% accuracy on the Airlines dataset, 99% accuracy on the Crowd Flower dataset, and 99% accuracy on the Apple dataset. Our method is better in terms of accuracy and efficiency than the state-of-the-art methods, which shows that adding emojis to sentiment analysis works well.</p>