情報処理学会 第81回全国大会講演要旨

Probabilistic Relational Models (PRM) extend Bayesian networks allowing the representation of information in a relational database, by introducing relations and individuals. However, learning a PRM structure from relational data is a more complex task than learning a Bayesian Network from “flat” data. The main difficulties that arise while learning a PRM are establishing what are the legal dependency structures, searching for possible structures, and scoring them. We propose a novel method for searching for a PRM structure using a score typically used to learn Bayesian networks and apply it to a real, large scale problem, achieving near state of the art performance.