Human opinions and feelings can be studied and analyzed in various fields. Sentiment analysis divides data into neutral, positive, and negative categories to classify a writer's or speaker's attitude toward various topics or events. This study uses a hybrid approach that combines (Particle_Swarm_Optimization PSO) with machine learning classifiers (Artificial Neural_ Networks ANN, Naïve Base NB, and Support.Vector.Machine SVM). Following the preprocessing phase of each data collection, a Convolution Neural Network (CNN) will be used to create feature vectors, preparing the raw data as a stored labeled dataset with three classes (positive, negative, and natural). After that, with two hidden layers whose parameters were optimized by PSO, this dataset will be prepared for classification using NB, SVM, and ANN. The best results will be obtained with the proposed Artificial Neural Network A.N.N which is close to 99% Acc. value, about SVM Acc. equal to 92%, and NB 89%, better than using the same ML algorithms