Sentiment analysis (SA) is a useful NLP task. There are hundreds of Arabic sentiments analysis systems. However, because of the morphological nature of the Arabic languages, there are still many challenges that need more work. In this paper, two classifiers have been used: Naive Bayes and CNN-LSTM models. The experiments are conducted on Arabic tweets dataset that consists of 58k tweets written in several dialects, the same preprocessing steps have been done before fitting the models. The experimental results show that the deep Learning CNN-LSTM classifier fits better for this task which achieved an accuracy of 98% while Naive Bayes achieved 87.6%.