Abstarct: Aspect-baxsed sentiment analysis is a text analysis technique that classifies comments baxsed on aspects and identifies sentiments related to each aspect. This technique can be used to automatically analyze feedback from customers to different parts of goods or services and help employers to focus on points that need quality improvement. In this thesis, we will introduce a new architecture baxsed on deep learning for aspect-baxsed sentiment analysis, which aims to provide a method for text classification baxsed on aspect. Due to the fact that we need to analyze input words in parallel to analyze emotions, we used the attention encoder laxyer to provide better results with high accuracy. Here we have proposed two architectures, in the first model, two laxyers of attention encoders (which is a parallelizable and interactive alternative to LSTM and applied to calculate the hidden states of input embeddings) are used, and in the other model, one encoder laxyer is used. Attention has been used with several multi-head attention and point convolution transformation. The test of this architecture has been done on three different datasets of restaurants, laptops and Twitter, which comparison with modern methods of aspect-baxsed sentiment analysis will show the high accuracy of this method. As an example, the aspect-baxsed sentiment analysis on the restaurant dataset has shown 0.8054 percent accuracy in the first model and 0.8187 percent accuracy in the second model, which shows higher accuracy compared to similar methods in this field.