Sentiment analysis is an essential component of natural language processing, which focuses on extracting subjective insights, like emotions and opinions from text. In this research, an innovative framework has been introduced for Twitter sentiment analysis that integrates Echo State Networks (ESN), Improved Student Psychology Based Optimization (ISPBO), and BERT embeddings. The proposed ESN-ISPBO-BERT model was evaluated on the SemEval-2016-1 and SemEval-2016-2 datasets and compared against other models, including SVM-Glove, CNN-BERT, LSTM, CNN, KNN, SVM, BERT, and GRU. The outcomes indicate that the suggested model exceeds all baseline models, and has achieved outstanding performance. On SemEval-2016-2, it achieves 98.82% accuracy, 98.79% precision, 98.96% recall, and 98.87% F1-score, while on SemEval-2016-1, it achieves 98.76% accuracy, 98.81% precision, 98.92% recall, as well as 98.86% F1-score. Moreover, the suggested model achieved the values of 98.51%, 98.42%, 98.87%, and 98.64% in terms of accuracy, precision, recall, and F1-score on Stanford Sentiment Treebank (SST-2). These outcomes indicate the efficiency of combining reservoir calculating, innovative optimization techniques, and contextual embeddings for the aim of sentiment analysis. The proposed model is a strong solution for the evaluation of Twitter data, with possible applications in brand monitoring, analyzing customer feedback, and tracking sentiment in real-time. This investigation represents the importance of hybrid strategies in tackling the issues of sentiment analysis on social media.