In this paper, based on the characteristics of teaching English polite phrases, the acquired research data are subjected to the segmentation operation, and a graph convolutional neural network is used to iteratively train them, finally completing the modeling of the knowledge graph of English polite phrases and determining the parameters of the model. The constructed knowledge graph is ready to be introduced into the English polite language teaching activities, and a model for teaching English polite language has been designed based on it. Simulation experiments and independent sample t-tests are used to validate and analyze the knowledge graph model and the teaching model, respectively. Upon analysis, it was found that the model in this paper could quickly and accurately identify the number of times the English verbs of polite phrases appeared, which was reflected in the fact that Promise appeared five times and Praise appeared four times. After the teaching intervention, the English test scores of students in Group E (Mean: 15.994) were significantly higher than those of Group C (Mean: 12.152), and the difference between the scores of the two groups was statistically significant (t=6.811, P=0.002<0.05), which confirmed that the English politeness language teaching model of the Knowledge Graph has a facilitating effect on the students’ English scores, and at the same time, summarized the effect of English politeness language education in cultural communication. This study improves the quality of English polite language teaching and students’ learning effect so that the informatization of English polite language teaching in colleges and universities can be further developed and plays a role in promoting global cultural communication.