Sentiment analysis is the most well-known and active field of data mining research. Twitter is one of the most helpful social media tools nowadays for gathering and disseminating people's ideas, emotions, and points of view. thoughts about particular entities. As a result, sentiment analysis in the field of natural language processing became intriguing. Even though a number of approaches for sentiment analysis have been developed, system efficiency and accuracy can always be increased. The suggested architecture is designed to meet it with a deep learning-based sentiment analysis and an efficient and effective feature selection based on optimisation. The sentiment 140 dataset is used in this study to evaluate the effectiveness of the proposed gated attention recurrent network (GARN) architecture. Pre-processing first purges and filters out the accessible dataset. The Log Term Frequency-based Modified Inverse Class Frequency (LTF-MICF) model is then used to extract the sentiment-based features from the preprocessed data using a term weight-based feature extraction method. In the third phase, a hybrid mutation-based white shark optimizer (HMWSO) is introduced for feature selection. The GARN architecture, which combines recurrent neural networks (RNN) and attention mechanisms, is used to categorise the sentiment classes, such as positive, negative, and neutral, using the selected features. The performance of the proposed and current classifiers is then compared. The evaluated performance measures are accuracy, precision, recall, and f-measure, and the acquired values for these metrics using the suggested GARN are accuracy, precision, recall, and f-measure, respectively