We conduct experiments with the datasets of SemEval 2014 task4 to evaluate our model, the SemEval 2014 datasets consist of reviews in two categories: Restaurant and Laptop, and the reviews contains three labels of sentiment polarity: {positive, negative, neutral}