The primary issue with the prior strategy is that it uses sparse values, which increases classification noise and increases the likelihood of false positives. The proposed approach places considerable emphasis on efficient feature weighting using an attention mechanism and an XG Boost classifier. We will be using the sentiment database, which contains three classes: negative, positive, and neutral, to train using XG-Boost and grey wolf optimization. In the world today, we have grown accustomed to a continuous flow of data. Major social media platforms such as Twitter, Facebook, and Instagram confront a significant challenge because of spam accounts. Such accounts are used to trick unwary legitimate users into clicking on continuing repeated posts via bots. This can have a significant impact on the user experience on these sites. Much time and effort has been invested in developing effective methods for detecting different sentiment of tweets. This issue is effectively resolved through tweet sentiment analysis or classification. In this paper we have develop a methodology to reduce features dimension and extract only efficient features and reduce problem of features noise because of high dimension which increases the overlapping of features. In this paper reducing the noise of features because of quantity of features and reduce error in detection of sentiments.The results reveal an improvement in accuracy of 4-5%, recall of 3-4%, and precision of 2-3% over previous classifiers and approaches.

Tweet, Twitter, social media, deep learning, sentiment analysis