Different people have embraced social media as an official and easy form of communication. On social media sites such as Twitter and Facebook, users post messages, attach photos, and videos. As a result, a large amount of sentimentally rich data is produced. The application of sentiment analysis, for example, to ascertain consumer feedback regarding a specific firm or product. The techniques employed to examine these feelings are machine learning techniques. However, the amount of data utilised for training and testing, data domains, noise, and the curse of dimensionality alter sentiment analysis’s effectiveness. This work’s objective is to produce a novel sentiment analysis model for social media data from Facebook and Twitter that incorporates part-of-speech tagging and dimensionality reduction techniques. The proposed method of hybrid Grey Wolf Optimisation with Back Propagation Neural Network (GWOBPNN) algorithm is tested against sentiment analysis models, and its performance is compared with both. The methods used to test this model include Naïve Bayes, K-Nearest Neighbour, and Support Vector Machine algorithms. According to experimental data, the model uses machine learning approaches to increase sentiment analysis performance. Our newly developed method performs better than the existing methods when measuring factors like accuracy, precision, recall, and F Measure are taken into account.