In recent years, the combination of knowledge map and question and answer system are more and more applied in professional fields. In this paper, we firstly constructed a knowledge graph of traditional Chinese medical records which contains four entities: herbs, prescriptions, diseases, symptoms and their relationships; Secondly, we use Node2vec algorithm to represent the graph; Finally, we match questions and answers based on KNN. Experiment result shows that with the output of five prescriptions, our system’s accuracy reaches 93.4%, which achieves good results.