Abstract:Bayesian network originates from people's research on uncertainty problems in the field of artificial intelligence, and it is an important tool for uncertainty problems inference and data analysis. Structure learning is the core content of Bayesian network research, and the K2 algorithm is one of the classical algorithms for structure learning. In order to solve the problem that the learning effect of K2 algorithm strongly depends on the node order, this paper proposes a new hybrid structure learning algorithm: double K2 algorithm. The algorithm firstly takes the node information as the initial node order, and obtains the initial network structure through the search strategy of K2 algorithm. Then, the modified node order is obtained by using topological ordering on the initial network structure. Finally, K2 algorithm obtains the optimal network structure through the modified node order. Experimental results show that the double K2 algorithm is better than other classical algorithms in accuracy and efficiency.