Learning Probabilistic Relational Models

摘要: A large portion of real-world data is stored in commercial relational database systems. In contrast, most statistical learning methods work only with "flat" representations. Thus, to apply these methods, we are forced convert our into a flat form, thereby losing much the structure present database. This paper builds on recent probabilistic models (PRMs), and describes how learn them from databases. PRMs allow properties an object depend probabilistically both other that related objects. Although significantly more expressive than standard models, such as Bayesian networks, show extend well-known for networks models. We describe parameter estimation -- automatic induction dependency model. Moreover, procedure can exploit retrieval techniques efficient datasets. experimental results real synthetic