Daily, a massive amount of textual data is added to the Internet, making it difficult to extract relevant information. People can express their joy and sadness on social media. Natural language processing (NLP) sentiment analysis may determine a user’s emotional state, such as happiness, sadness, or rage, in social media posts. Sentiment analysis is a pivotal factor affecting various aspects. This work addresses the difficulty of sentiment analysis through the application of natural language processing and machine learning techniques. The news dataset we possess includes Bengali short stories. This study employed word vectorization techniques alongside a machine learning algorithm on the dataset. The objective is to evaluate the precision of sentiment analysis employing different word embeddings on Bengali short stories and other datasets. Various word vectorization techniques are used on three distinct datasets, including dataset 3 meticulously constructed by us. The TF-IDF vectorization method outperforms other vectorization techniques on Bengali data. The research uses various word vectorization techniques and machine learning algorithms, including Linear Regression, Random Forest, Naive Bayes and SVC, to examine three Bengali datasets, one of which is a manual compilation of Bengali short stories. The most effective vectorization approach identified was TF-IDF. The models achieved 81.22% accuracy on Dataset 1, 51.509% accuracy on Dataset 2 and 49.15% accuracy on Dataset 3 using TF-IDF.