| Abstract | Probabilistic graphical models, in particular Bayesian net-works, are useful models for representing statistical pat-
terns in propositional domains. Recent work develops ef-
fective techniques for learning these models directly from
data. However these techniques apply only to attribute-value
(i.e., flat) representations of the data. Probabilistic relational
models (PRMs) allow us to represent much richer depen-
dency structures, involving multiple entities and the rela-
tions between them; they allow the properties of an entity
to depend probabilistically on properties of related entities.
PRMs represent a generic dependence, which is then instan-
tiated for specific circumstances, i.e., for a particular set of
entities and relations between them. Friedman et al. showed
how to learn PRMs from relational data, and presented tech-
niques for learning both parameters and probabilistic depen-
dency structure for the attributes in a relational model. Here
we examine the benefit that class hierarchies can provide
PRMs. We show how the introduction of subclasses allows
us to use inheritance and specialization to refine our models.
We show how to learn PRMs with class hierarchies (PRM-
CH) in two settings. In the first, the class hierarchy is pro-
vided, as part of the input, in the relational schema for the
domain. In the second setting, in addition to learning the
PRM, we must learn the class hierarchy. Finally we dis-
cuss how PRM-CHs allow us to build models that can repre-
sent models for both particular instances in our domain, and
classes of objects in our domain, bridging the gap between a
class-based model and an attribute-value-based model.
|
