Abstarct: With the advancement of technology in the Internet, one of the most important security challenges and risks is phishing attacks and fraud by Fishers. Phishing is a type of cyber attack, which always tries to get information such as username, password, bank account information and the like by forging a website, email address and convincing the user to enter this information. Although phishing involves psychological deception and relies on human error instead of software or hardware errors, users play a major role in preventing them from being caught in this type of attack. On the other hand, it is not possible to detect suspicious domains by users and current phishing detection systems often can not adapt to new attacks and have low detection accuracy and high false positive rates. Graph-baxsed methods are one of the methods for identifying malicious domains. The main challenge in this dissertation is how to increase the accuracy of recognizing these domains with Graph-baxsed method and deep learning. Therefore, in this dissertation, Graph-baxsed phishing detection system using deep learning is presented. The key idea of this method is to extract IP from the domain, define the relationship between the domains, their weight and also convert the data to vector by Node2vec algorithm. Then, using CNN and DENSE deep learning models, the classification and identification operation is performed. The results showed that the Graph-baxsed method and the use of deep learning have a favorable effect on identifying these domains. Thus that the proposed method has 99% accuracy in the first data set, which has increased by 1% compared to the previous best method, BP.