Structure learning method of Bayesian network with prior information
Yang Feng · Systems engineering and electronics · 2012
In the process of Bayesian network structure learning,how to gather prior information and use it effectively are very important for building an exact network structure.Therefore,according to the three stages of Bayesian network structure learning with prior information: information gathering,information fusion and the optimization of learned network structure.Firstly,the deficiency of currently available methods that is used for obtaining prior information is discussed,and then a new method of gathering prior structure information based on so-called belief map is proposed.Secondly,the minimum description length score is modified so that it can fuse uncertain prior information.Finaly,a simulated annealing method is revised appropriately according to the characteristics of the problem for searching the optimal structure.Experimental results show that the proposed method can improve the precision of structure learning efficiently.