Using structural learning algorithms to combine database knowledge with prior information from domain experts can be helpful when building a Bayesian network. If the prior information is too hazy to be encoded as properties that are local to families of variables, conventional learning algorithms cannot easily incorporate it. In object-oriented fields like computer networks, large pedigrees, and genetic analysis, for instance, conventional algorithms do not take advantage of prior information about repetitive structures. A strategy for conducting structural learning in object-oriented domains is presented in this paper. It is argued that this method supports a natural approach for expressing and incorporating prior information provided by domain experts, and it is demonstrated that this method is more efficient than conventional algorithms in such domains.

Knowledge fusion, object orientation, structural learning, and Bayesian networks