The application of natural language processing has proven to be particularly advantageous in the realm of sentiment analysis, particularly when dealing with vast and intricate amounts of unstructured data found within social media platforms. These insights contribute to making informed decisions. Among its multifaceted uses, social media serves as platform for individuals to articulate themselves through tweets and posts. The sentiment of written content, specifically within the realm of social media data, holds the potential to be dissected to extract opinions, emotions, and significant perspectives. Effectively assessing sentiment in publicly available social media data faces challenges from both theoretical and technological sources. While various methodologies have been developed over time, their effectiveness has been constrained by their focus on small datasets, rendering them inadequate in addressing the aforementioned complexities optimally. The recommended approach offers a solution to these challenges, encompassing critical facets such as data collection, feature encoding, feature selection, data pre-processing, and classification, as previously delineated. Notably, the feature encoding stage adopts a hybrid technique that combines the incorporation of bi-gram and tri-gram words. Rigorous testing conducted across multiple benchmark datasets to comprehensively evaluate the performance of the proposed framework. Significantly, the suggested method not only delivers comparable outcomes but, in various instances, superior results while necessitating less intricate computational processes. The average accuracy results attained using the multilayer perceptron neural network ranged from 89 to 91. The methodologies presented in paper are poised to substantially elevate the efficacy of sentiment analysis across diverse blog and social media content.