The information flow of the brain functional network can reflect the causal relationship among different brain regions. Studying the causal relationship of brain regions under different emotional states is not only necessary to reveal the mechanism of emotion generation， but also to effectively analyze， identify and monitor the generation of emotion. In the field of artificial intelligence， effective emotion recognition will directly affect the human-computer interaction process. This paper used the DEAP emotional data to construct a directed network by the method of transfer entropy， discussed the flow direction and change rules of information flow under different emotional states， as well as the connection patterns existing in different brain regions. It is found that the intensity of information flow at each node is different under different emotional states. The higher the arousal， the greater the intensity of information flow. There are also significant differences in how information flows and connects. At the same time， this study further demonstrates that dependence truly exists between arousal and valence （i.e.， two dimensions of emotion）， and provides an important reference direction and research idea for emotion recognition.