The present disclosure relates to a human-machine multi-turn conversation method and system for human-machine interaction, and an intelligent apparatus. The method includes: S1, establishing a knowledge graph of user conversation behavior information; S2, determining, according to information currently input by a user, a node corresponding to the information currently input and at least one child node of the node in the knowledge graph; S3, calculating a support degree of the at least one child node relative to the node according to the number of times of querying the node and the number of times of querying both the at least one child node and the node in a historical query record of the knowledge graph; and S4, determining whether to output semantic information of the at least one child node by determining a size relation between the support degree and a preset support degree threshold value.